Release: v4.8.1

* Fix torchscript tests

* Better test

* Remove bogus print

.circleci/TROUBLESHOOT.md

@@ -1,7 +0,0 @@
-# Troubleshooting
-
-This is a document explaining how to deal with various issues on Circle-CI. The entries may include actually solutions or pointers to Issues that cover those.
-
-## Circle CI
-
-* pytest worker runs out of resident RAM and gets killed by `cgroups`: https://github.com/huggingface/transformers/issues/11408

.circleci/config.yml

@@ -1,162 +1,499 @@
 version: 2.1
-setup: true
 orbs:
-    continuation: circleci/continuation@0.1.0
+    gcp-gke: circleci/gcp-gke@1.0.4
+    go: circleci/go@1.3.0
+
+# TPU REFERENCES
+references:
+    checkout_ml_testing: &checkout_ml_testing
+        run:
+            name: Checkout ml-testing-accelerators
+            command: |
+                git clone https://github.com/GoogleCloudPlatform/ml-testing-accelerators.git
+                cd ml-testing-accelerators
+                git fetch origin 5e88ac24f631c27045e62f0e8d5dfcf34e425e25:stable
+                git checkout stable
+    build_push_docker: &build_push_docker
+        run:
+            name: Configure Docker
+            command: |
+                gcloud --quiet auth configure-docker
+                cd docker/transformers-pytorch-tpu
+                if [ -z "$CIRCLE_PR_NUMBER" ]; then docker build --tag "$GCR_IMAGE_PATH:$CIRCLE_WORKFLOW_JOB_ID" -f Dockerfile --build-arg "TEST_IMAGE=1" . ; else docker build --tag "$GCR_IMAGE_PATH:$CIRCLE_WORKFLOW_JOB_ID" -f Dockerfile --build-arg "TEST_IMAGE=1" --build-arg "GITHUB_REF=pull/$CIRCLE_PR_NUMBER/head" . ; fi
+                docker push "$GCR_IMAGE_PATH:$CIRCLE_WORKFLOW_JOB_ID"
+    deploy_cluster: &deploy_cluster
+        run:
+            name: Deploy the job on the kubernetes cluster
+            command: |
+                go get github.com/google/go-jsonnet/cmd/jsonnet && \
+                export PATH=$PATH:$HOME/go/bin && \
+                kubectl create -f docker/transformers-pytorch-tpu/dataset.yaml || true && \
+                job_name=$(jsonnet -J ml-testing-accelerators/ docker/transformers-pytorch-tpu/bert-base-cased.jsonnet --ext-str image=$GCR_IMAGE_PATH --ext-str image-tag=$CIRCLE_WORKFLOW_JOB_ID | kubectl create -f -) && \
+                job_name=${job_name#job.batch/} && \
+                job_name=${job_name% created} && \
+                echo "Waiting on kubernetes job: $job_name" && \
+                i=0 && \
+                # 30 checks spaced 30s apart = 900s total.
+                max_checks=30 && \
+                status_code=2 && \
+                # Check on the job periodically. Set the status code depending on what
+                # happened to the job in Kubernetes. If we try max_checks times and
+                # still the job hasn't finished, give up and return the starting
+                # non-zero status code.
+                while [ $i -lt $max_checks ]; do ((i++)); if kubectl get jobs $job_name -o jsonpath='Failed:{.status.failed}' | grep "Failed:1"; then status_code=1 && break; elif kubectl get jobs $job_name -o jsonpath='Succeeded:{.status.succeeded}' | grep "Succeeded:1" ; then status_code=0 && break; else echo "Job not finished yet"; fi; sleep 30; done && \
+                echo "Done waiting. Job status code: $status_code" && \
+                pod_name=$(kubectl get po -l controller-uid=`kubectl get job $job_name -o "jsonpath={.metadata.labels.controller-uid}"` | awk 'match($0,!/NAME/) {print $1}') && \
+                echo "GKE pod name: $pod_name" && \
+                kubectl logs -f $pod_name --container=train
+                echo "Done with log retrieval attempt." && \
+                gcloud container images delete "$GCR_IMAGE_PATH:$CIRCLE_WORKFLOW_JOB_ID" --force-delete-tags && \
+                exit $status_code
+    delete_gke_jobs: &delete_gke_jobs
+        run:
+            name: Delete GKE Jobs
+            command: |
+                # Match jobs whose age matches patterns like '1h' or '1d', i.e. any job
+                # that has been around longer than 1hr. First print all columns for
+                # matches, then execute the delete.
+                kubectl get job | awk 'match($4,/[0-9]+[dh]/) {print $0}'
+                kubectl delete job $(kubectl get job | awk 'match($4,/[0-9]+[dh]/) {print $1}')
+
+
 
-parameters:
-    nightly:
-        type: boolean
-        default: false
 
 jobs:
-    # Fetch the tests to run
-    fetch_tests:
+    run_tests_torch_and_tf:
         working_directory: ~/transformers
         docker:
-            - image: cimg/python:3.7.12
+            - image: circleci/python:3.6
+        environment:
+            OMP_NUM_THREADS: 1
+            RUN_PT_TF_CROSS_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
         parallelism: 1
         steps:
             - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-torch_and_tf-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install GitPython
-            - run: pip install .
-            - run: mkdir -p test_preparation
-            - run: python utils/tests_fetcher.py | tee tests_fetched_summary.txt
+            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
+            - save_cache:
+                key: v0.4-{{ checksum "setup.py" }}
+                paths:
+                    - '~/.cache/pip'
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf ./tests/ -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
-                  path: ~/transformers/tests_fetched_summary.txt
-            - run: |
-                if [ -f test_list.txt ]; then
-                    cp test_list.txt test_preparation/test_list.txt
-                else
-                    touch test_preparation/test_list.txt
-                fi
-            - run: python utils/tests_fetcher.py --filter_pipeline_tests
-            - run: |
-                if [ -f test_list.txt ]; then
-                    mv test_list.txt test_preparation/filtered_test_list.txt
-                else
-                    touch test_preparation/filtered_test_list.txt
-                fi
-            - run: python utils/tests_fetcher.py --filters tests examples | tee examples_tests_fetched_summary.txt
-            - run: |
-                  if [ -f test_list.txt ]; then
-                      mv test_list.txt test_preparation/examples_test_list.txt
-                  else
-                      touch test_preparation/examples_test_list.txt
-                  fi
+                  path: ~/transformers/tests_output.txt
             - store_artifacts:
-                  path: test_preparation/test_list.txt
-            - store_artifacts:
-                  path: ~/transformers/test_preparation/filtered_test_list.txt
-            - store_artifacts:
-                  path: test_preparation/examples_test_list.txt
-            - run: python .circleci/create_circleci_config.py --fetcher_folder test_preparation
-            - run: |
-                  if [ ! -s test_preparation/generated_config.yml ]; then
-                      echo "No tests to run, exiting early!"
-                      circleci-agent step halt
-                  fi
-            - run: cp test_preparation/generated_config.yml test_preparation/generated_config.txt
-            - store_artifacts:
-                  path: test_preparation/generated_config.txt
-            - continuation/continue:
-                  configuration_path: test_preparation/generated_config.yml
+                  path: ~/transformers/reports
 
-    # To run all tests for the nightly build
-    fetch_all_tests:
+    run_tests_torch_and_flax:
         working_directory: ~/transformers
         docker:
-            - image: cimg/python:3.7.12
+            - image: circleci/python:3.6
+        environment:
+            OMP_NUM_THREADS: 1
+            RUN_PT_FLAX_CROSS_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
         parallelism: 1
         steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-torch_and_flax-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .
-            - run: |
-                  mkdir test_preparation
-                  echo "tests" > test_preparation/test_list.txt
-                  echo "tests" > test_preparation/examples_test_list.txt
-            - run: python utils/tests_fetcher.py --filter_pipeline_tests
-            - run: mv test_list.txt test_preparation/filtered_test_list.txt
-            - run: python .circleci/create_circleci_config.py --fetcher_folder test_preparation
-            - run: cp test_preparation/generated_config.yml test_preparation/generated_config.txt
+            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
+            - save_cache:
+                key: v0.4-{{ checksum "setup.py" }}
+                paths:
+                    - '~/.cache/pip'
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_flax ./tests/ -m is_pt_flax_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
-                  path: test_preparation/generated_config.txt
-            - continuation/continue:
-                  configuration_path: test_preparation/generated_config.yml
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_torch:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-torch-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision,timm]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
+            - save_cache:
+                  key: v0.4-torch-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -n 3 --dist=loadfile -s --make-reports=tests_torch ./tests/ | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_tf:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-tf-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
+            - save_cache:
+                  key: v0.4-tf-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_tf ./tests/ | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_flax:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                keys:
+                    - v0.4-flax-{{ checksum "setup.py" }}
+                    - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: sudo pip install .[flax,testing,sentencepiece]
+            - save_cache:
+                  key: v0.4-flax-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_flax ./tests/ | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_pipelines_torch:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            OMP_NUM_THREADS: 1
+            RUN_PIPELINE_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-torch-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
+            - save_cache:
+                  key: v0.4-torch-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch -m is_pipeline_test ./tests/ | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_pipelines_tf:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            OMP_NUM_THREADS: 1
+            RUN_PIPELINE_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-tf-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
+            - save_cache:
+                  key: v0.4-tf-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf ./tests/ -m is_pipeline_test | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_custom_tokenizers:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            RUN_CUSTOM_TOKENIZERS: yes
+            TRANSFORMERS_IS_CI: yes
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-custom_tokenizers-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: pip install .[ja,testing,sentencepiece,jieba]
+            - run: python -m unidic download
+            - save_cache:
+                  key: v0.4-custom_tokenizers-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -s --make-reports=tests_custom_tokenizers ./tests/test_tokenization_bert_japanese.py | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_examples_torch:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.6
+        environment:
+            OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-torch_examples-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,torch,sentencepiece,testing]
+            - run: pip install -r examples/pytorch/_tests_requirements.txt
+            - save_cache:
+                  key: v0.4-torch_examples-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: TRANSFORMERS_IS_CI=1 python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_torch ./examples/pytorch/ | tee examples_output.txt
+            - store_artifacts:
+                  path: ~/transformers/examples_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_hub:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        environment:
+            HUGGINGFACE_CO_STAGING: yes
+            RUN_GIT_LFS_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-hub-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get install git-lfs
+            - run: |
+                git config --global user.email "ci@dummy.com"
+                git config --global user.name "ci"
+            - run: pip install --upgrade pip
+            - run: pip install .[torch,sentencepiece,testing]
+            - save_cache:
+                  key: v0.4-hub-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -sv ./tests/ -m is_staging_test
+
+    build_doc:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.6
+        steps:
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-build_doc-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+            - run: pip install --upgrade pip
+            - run: pip install ."[docs]"
+            - save_cache:
+                  key: v0.4-build_doc-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: cd docs && make html SPHINXOPTS="-W -j 4"
+            - store_artifacts:
+                path: ./docs/_build
+
+    deploy_doc:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.6
+        steps:
+            - add_ssh_keys:
+                fingerprints:
+                    - "5b:7a:95:18:07:8c:aa:76:4c:60:35:88:ad:60:56:71"
+            - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-deploy_doc-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+            - run: pip install --upgrade pip
+            - run: pip install ."[docs]"
+            - save_cache:
+                  key: v0.4-deploy_doc-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: ./.circleci/deploy.sh
 
     check_code_quality:
         working_directory: ~/transformers
         docker:
-            - image: cimg/python:3.7.12
-        resource_class: large
+            - image: circleci/python:3.6
+        resource_class: medium
         environment:
             TRANSFORMERS_IS_CI: yes
-            PYTEST_TIMEOUT: 120
         parallelism: 1
         steps:
             - checkout
             - restore_cache:
                   keys:
-                      - v0.5-code_quality-{{ checksum "setup.py" }}
-                      - v0.5-code-quality
+                      - v0.4-code_quality-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
+            - run: pip install isort
             - run: pip install .[all,quality]
             - save_cache:
-                  key: v0.5-code_quality-{{ checksum "setup.py" }}
+                  key: v0.4-code_quality-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: black --check --preview examples tests src utils
+            - run: black --check examples tests src utils
             - run: isort --check-only examples tests src utils
             - run: python utils/custom_init_isort.py --check_only
-            - run: python utils/sort_auto_mappings.py --check_only
             - run: flake8 examples tests src utils
-            - run: doc-builder style src/transformers docs/source --max_len 119 --check_only --path_to_docs docs/source
-            - run: python utils/check_doc_toc.py
-
-    check_repository_consistency:
-        working_directory: ~/transformers
-        docker:
-            - image: cimg/python:3.7.12
-        resource_class: large
-        environment:
-            TRANSFORMERS_IS_CI: yes
-            PYTEST_TIMEOUT: 120
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.5-repository_consistency-{{ checksum "setup.py" }}
-                      - v0.5-repository_consistency
-            - run: pip install --upgrade pip
-            - run: pip install .[all,quality]
-            - save_cache:
-                  key: v0.5-repository_consistency-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
+            - run: python utils/style_doc.py src/transformers docs/source --max_len 119 --check_only
             - run: python utils/check_copies.py
             - run: python utils/check_table.py
             - run: python utils/check_dummies.py
             - run: python utils/check_repo.py
             - run: python utils/check_inits.py
-            - run: python utils/check_config_docstrings.py
-            - run: make deps_table_check_updated
-            - run: python utils/tests_fetcher.py --sanity_check
-            - run: python utils/update_metadata.py --check-only
 
+    check_repository_consistency:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.6
+        resource_class: small
+        parallelism: 1
+        steps:
+            - checkout
+            - run: pip install requests
+            - run: python ./utils/link_tester.py
+
+# TPU JOBS
+    run_examples_tpu:
+        docker:
+            - image: circleci/python:3.6
+        environment:
+            OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
+        resource_class: xlarge
+        parallelism: 1
+        steps:
+            - checkout
+            - go/install
+            - *checkout_ml_testing
+            - gcp-gke/install
+            - gcp-gke/update-kubeconfig-with-credentials:
+                  cluster: $GKE_CLUSTER
+                  perform-login: true
+            - setup_remote_docker
+            - *build_push_docker
+            - *deploy_cluster
+
+    cleanup-gke-jobs:
+        docker:
+            - image: circleci/python:3.6
+        steps:
+            - gcp-gke/install
+            - gcp-gke/update-kubeconfig-with-credentials:
+                  cluster: $GKE_CLUSTER
+                  perform-login: true
+            - *delete_gke_jobs
+
+workflow_filters: &workflow_filters
+    filters:
+        branches:
+            only:
+                - master
 workflows:
     version: 2
-    setup_and_quality:
-        when:
-            not: <<pipeline.parameters.nightly>>
+    build_and_test:
         jobs:
             - check_code_quality
             - check_repository_consistency
-            - fetch_tests
-
-    nightly:
-        when: <<pipeline.parameters.nightly>>
-        jobs:
-            - check_code_quality
-            - check_repository_consistency
-            - fetch_all_tests
+            - run_examples_torch
+            - run_tests_custom_tokenizers
+            - run_tests_torch_and_tf
+            - run_tests_torch_and_flax
+            - run_tests_torch
+            - run_tests_tf
+            - run_tests_flax
+            - run_tests_pipelines_torch
+            - run_tests_pipelines_tf
+            - run_tests_hub
+            - build_doc
+            - deploy_doc: *workflow_filters
+#    tpu_testing_jobs:
+#        triggers:
+#            - schedule:
+#                # Set to run at the first minute of every hour.
+#                cron: "0 8 * * *"
+#                filters:
+#                    branches:
+#                        only:
+#                            - master
+#        jobs:
+#            - cleanup-gke-jobs
+#            - run_examples_tpu

.circleci/create_circleci_config.py

@@ -1,391 +0,0 @@
-# coding=utf-8
-# Copyright 2022 The HuggingFace Inc. team.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import argparse
-import copy
-import os
-from dataclasses import dataclass
-from typing import Any, Dict, List, Optional
-
-import yaml
-
-
-COMMON_ENV_VARIABLES = {"OMP_NUM_THREADS": 1, "TRANSFORMERS_IS_CI": True, "PYTEST_TIMEOUT": 120}
-COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "dist": "loadfile", "s": None}
-DEFAULT_DOCKER_IMAGE = [{"image": "cimg/python:3.7.12"}]
-TORCH_SCATTER_INSTALL = "pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.12.0+cpu.html"
-
-
-@dataclass
-class CircleCIJob:
-    name: str
-    additional_env: Dict[str, Any] = None
-    cache_name: str = None
-    cache_version: str = "0.5"
-    docker_image: List[Dict[str, str]] = None
-    install_steps: List[str] = None
-    marker: Optional[str] = None
-    parallelism: Optional[int] = 1
-    pytest_num_workers: int = 8
-    pytest_options: Dict[str, Any] = None
-    resource_class: Optional[str] = "xlarge"
-    tests_to_run: Optional[List[str]] = None
-    working_directory: str = "~/transformers"
-
-    def __post_init__(self):
-        # Deal with defaults for mutable attributes.
-        if self.additional_env is None:
-            self.additional_env = {}
-        if self.cache_name is None:
-            self.cache_name = self.name
-        if self.docker_image is None:
-            # Let's avoid changing the default list and make a copy.
-            self.docker_image = copy.deepcopy(DEFAULT_DOCKER_IMAGE)
-        if self.install_steps is None:
-            self.install_steps = []
-        if self.pytest_options is None:
-            self.pytest_options = {}
-        if isinstance(self.tests_to_run, str):
-            self.tests_to_run = [self.tests_to_run]
-
-    def to_dict(self):
-        job = {
-            "working_directory": self.working_directory,
-            "docker": self.docker_image,
-            "environment": {**COMMON_ENV_VARIABLES, **self.additional_env},
-        }
-        if self.resource_class is not None:
-            job["resource_class"] = self.resource_class
-        if self.parallelism is not None:
-            job["parallelism"] = self.parallelism
-        steps = [
-            "checkout",
-            {"attach_workspace": {"at": "~/transformers/test_preparation"}},
-            {
-                "restore_cache": {
-                    "keys": [
-                        f"v{self.cache_version}-{self.cache_name}-" + '{{ checksum "setup.py" }}',
-                        f"v{self.cache_version}-{self.cache_name}-",
-                    ]
-                }
-            },
-        ]
-        steps.extend([{"run": l} for l in self.install_steps])
-        steps.append(
-            {
-                "save_cache": {
-                    "key": f"v{self.cache_version}-{self.cache_name}-" + '{{ checksum "setup.py" }}',
-                    "paths": ["~/.cache/pip"],
-                }
-            }
-        )
-
-        all_options = {**COMMON_PYTEST_OPTIONS, **self.pytest_options}
-        pytest_flags = [f"--{key}={value}" if value is not None else f"-{key}" for key, value in all_options.items()]
-        pytest_flags.append(
-            f"--make-reports={self.name}" if "examples" in self.name else f"--make-reports=tests_{self.name}"
-        )
-        test_command = f"python -m pytest -n {self.pytest_num_workers} " + " ".join(pytest_flags)
-        if self.tests_to_run is None:
-            test_command += " << pipeline.parameters.tests_to_run >>"
-        else:
-            test_command += " " + " ".join(self.tests_to_run)
-        if self.marker is not None:
-            test_command += f" -m {self.marker}"
-        test_command += " | tee tests_output.txt"
-        steps.append({"run": {"name": "Run tests", "command": test_command}})
-        steps.append({"store_artifacts": {"path": "~/transformers/tests_output.txt"}})
-        steps.append({"store_artifacts": {"path": "~/transformers/reports"}})
-        job["steps"] = steps
-        return job
-
-    @property
-    def job_name(self):
-        return self.name if "examples" in self.name else f"tests_{self.name}"
-
-
-# JOBS
-torch_and_tf_job = CircleCIJob(
-    "torch_and_tf",
-    additional_env={"RUN_PT_TF_CROSS_TESTS": True},
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng git-lfs",
-        "git lfs install",
-        "pip install --upgrade pip",
-        "pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]",
-        TORCH_SCATTER_INSTALL,
-        "pip install tensorflow_probability",
-        "pip install https://github.com/kpu/kenlm/archive/master.zip",
-        "pip install git+https://github.com/huggingface/accelerate",
-    ],
-    marker="is_pt_tf_cross_test",
-    pytest_options={"rA": None, "durations": 0},
-)
-
-
-torch_and_flax_job = CircleCIJob(
-    "torch_and_flax",
-    additional_env={"RUN_PT_FLAX_CROSS_TESTS": True},
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
-        "pip install --upgrade pip",
-        "pip install .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]",
-        TORCH_SCATTER_INSTALL,
-        "pip install https://github.com/kpu/kenlm/archive/master.zip",
-        "pip install git+https://github.com/huggingface/accelerate",
-    ],
-    marker="is_pt_flax_cross_test",
-    pytest_options={"rA": None, "durations": 0},
-)
-
-
-torch_job = CircleCIJob(
-    "torch",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng time",
-        "pip install --upgrade pip",
-        "pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]",
-        TORCH_SCATTER_INSTALL,
-        "pip install https://github.com/kpu/kenlm/archive/master.zip",
-        "pip install git+https://github.com/huggingface/accelerate",
-    ],
-    pytest_num_workers=3,
-)
-
-
-tf_job = CircleCIJob(
-    "tf",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
-        "pip install --upgrade pip",
-        "pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]",
-        "pip install tensorflow_probability",
-        "pip install https://github.com/kpu/kenlm/archive/master.zip",
-    ],
-    pytest_options={"rA": None},
-)
-
-
-flax_job = CircleCIJob(
-    "flax",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
-        "pip install --upgrade pip",
-        "pip install .[flax,testing,sentencepiece,flax-speech,vision]",
-        "pip install https://github.com/kpu/kenlm/archive/master.zip",
-    ],
-    pytest_options={"rA": None},
-)
-
-
-pipelines_torch_job = CircleCIJob(
-    "pipelines_torch",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
-        "pip install --upgrade pip",
-        "pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]",
-        TORCH_SCATTER_INSTALL,
-        "pip install https://github.com/kpu/kenlm/archive/master.zip",
-    ],
-    pytest_options={"rA": None},
-    tests_to_run="tests/pipelines/"
-)
-
-
-pipelines_tf_job = CircleCIJob(
-    "pipelines_tf",
-    install_steps=[
-        "pip install --upgrade pip",
-        "pip install .[sklearn,tf-cpu,testing,sentencepiece]",
-        "pip install tensorflow_probability",
-    ],
-    pytest_options={"rA": None},
-    tests_to_run="tests/pipelines/"
-)
-
-
-custom_tokenizers_job = CircleCIJob(
-    "custom_tokenizers",
-    additional_env={"RUN_CUSTOM_TOKENIZERS": True},
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y cmake",
-        {
-            "name": "install jumanpp",
-            "command":
-                "wget https://github.com/ku-nlp/jumanpp/releases/download/v2.0.0-rc3/jumanpp-2.0.0-rc3.tar.xz\n"
-                "tar xvf jumanpp-2.0.0-rc3.tar.xz\n"
-                "mkdir jumanpp-2.0.0-rc3/bld\n"
-                "cd jumanpp-2.0.0-rc3/bld\n"
-                "sudo cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local\n"
-                "sudo make install\n",
-        },
-        "pip install --upgrade pip",
-        "pip install .[ja,testing,sentencepiece,jieba,spacy,ftfy,rjieba]",
-        "python -m unidic download",
-    ],
-    parallelism=None,
-    resource_class=None,
-    tests_to_run=[
-        "./tests/models/bert_japanese/test_tokenization_bert_japanese.py",
-        "./tests/models/openai/test_tokenization_openai.py",
-        "./tests/models/clip/test_tokenization_clip.py",
-    ],
-)
-
-
-examples_torch_job = CircleCIJob(
-    "examples_torch",
-    cache_name="torch_examples",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
-        "pip install --upgrade pip",
-        "pip install .[sklearn,torch,sentencepiece,testing,torch-speech]",
-        "pip install -r examples/pytorch/_tests_requirements.txt",
-    ],
-    tests_to_run="./examples/pytorch/",
-)
-
-
-examples_tensorflow_job = CircleCIJob(
-    "examples_tensorflow",
-    cache_name="tensorflow_examples",
-    install_steps=[
-        "pip install --upgrade pip",
-        "pip install .[sklearn,tensorflow,sentencepiece,testing]",
-        "pip install -r examples/tensorflow/_tests_requirements.txt",
-    ],
-    tests_to_run="./examples/tensorflow/",
-)
-
-
-examples_flax_job = CircleCIJob(
-    "examples_flax",
-    cache_name="flax_examples",
-    install_steps=[
-        "pip install --upgrade pip",
-        "pip install .[flax,testing,sentencepiece]",
-        "pip install -r examples/flax/_tests_requirements.txt",
-    ],
-    tests_to_run="./examples/flax/",
-)
-
-
-hub_job = CircleCIJob(
-    "hub",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install git-lfs",
-        'git config --global user.email "ci@dummy.com"',
-        'git config --global user.name "ci"',
-        "pip install --upgrade pip",
-        "pip install .[torch,sentencepiece,testing]",
-    ],
-    marker="is_staging_test",
-    pytest_num_workers=1,
-)
-
-
-onnx_job = CircleCIJob(
-    "onnx",
-    install_steps=[
-        "pip install --upgrade pip",
-        "pip install .[torch,tf,testing,sentencepiece,onnxruntime,vision,rjieba]",
-    ],
-    pytest_options={"k onnx": None},
-    pytest_num_workers=1,
-)
-
-
-layoutlm_job = CircleCIJob(
-    "layoutlmv2_and_v3",
-    install_steps=[
-        "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev",
-        "pip install --upgrade pip",
-        "pip install .[torch,testing,vision]",
-        "pip install torchvision",
-        "pip install 'git+https://github.com/facebookresearch/detectron2.git'",
-        "sudo apt install tesseract-ocr",
-        "pip install pytesseract",
-    ],
-    tests_to_run="tests/models/*layoutlmv*",
-    pytest_num_workers=1,
-    pytest_options={"durations": 100},
-)
-
-
-REGULAR_TESTS = [
-    torch_and_tf_job,
-    torch_and_flax_job,
-    torch_job,
-    tf_job,
-    flax_job,
-    custom_tokenizers_job,
-    hub_job,
-    onnx_job,
-    layoutlm_job,
-]
-EXAMPLES_TESTS = [
-    examples_torch_job,
-    examples_tensorflow_job,
-    examples_flax_job,
-]
-PIPELINE_TESTS = [
-    pipelines_torch_job,
-    pipelines_tf_job,
-]
-
-
-def create_circleci_config(folder=None):
-    if folder is None:
-        folder = os.getcwd()
-    jobs = []
-    all_test_file = os.path.join(folder, "test_list.txt")
-    if os.path.exists(all_test_file):
-        with open(all_test_file) as f:
-            all_test_list = f.read()
-    else:
-        all_test_list = []
-    if len(all_test_list) > 0:
-        jobs.extend(PIPELINE_TESTS)
-
-    test_file = os.path.join(folder, "filtered_test_list.txt")
-    if os.path.exists(test_file):
-        with open(test_file) as f:
-            test_list = f.read()
-    else:
-        test_list = []
-    if len(test_list) > 0:
-        jobs.extend(REGULAR_TESTS)
-
-    example_file = os.path.join(folder, "examples_test_list.txt")
-    if os.path.exists(example_file) and os.path.getsize(example_file) > 0:
-        jobs.extend(EXAMPLES_TESTS)
-
-    if len(jobs) > 0:
-        config = {"version": "2.1"}
-        config["parameters"] = {"tests_to_run": {"type": "string", "default": test_list}}
-        config["jobs"] = {j.job_name: j.to_dict() for j in jobs}
-        config["workflows"] = {"version": 2, "run_tests": {"jobs": [j.job_name for j in jobs]}}
-        with open(os.path.join(folder, "generated_config.yml"), "w") as f:
-            f.write(yaml.dump(config, indent=2, width=1000000, sort_keys=False))
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--fetcher_folder", type=str, default=None, help="Only test that all tests and modules are accounted for."
-    )
-    args = parser.parse_args()
-
-    create_circleci_config(args.fetcher_folder)

.circleci/deploy.sh

@@ -0,0 +1,67 @@
+cd docs
+
+function deploy_doc(){
+	echo "Creating doc at commit $1 and pushing to folder $2"
+	git checkout $1
+	pip install -U ..
+	if [ ! -z "$2" ]
+	then
+		if [ "$2" == "master" ]; then
+		    echo "Pushing master"
+			make clean && make html && scp -r -oStrictHostKeyChecking=no _build/html/* $doc:$dir/$2/
+			cp -r _build/html/_static .
+		elif ssh -oStrictHostKeyChecking=no $doc "[ -d $dir/$2 ]"; then
+			echo "Directory" $2 "already exists"
+			scp -r -oStrictHostKeyChecking=no _static/* $doc:$dir/$2/_static/
+		else
+			echo "Pushing version" $2
+			make clean && make html
+			rm -rf _build/html/_static
+			cp -r _static _build/html
+			scp -r -oStrictHostKeyChecking=no _build/html $doc:$dir/$2
+		fi
+	else
+		echo "Pushing stable"
+		make clean && make html
+		rm -rf _build/html/_static
+		cp -r _static _build/html
+		scp -r -oStrictHostKeyChecking=no _build/html/* $doc:$dir
+	fi
+}
+
+# You can find the commit for each tag on https://github.com/huggingface/transformers/tags
+deploy_doc "master" master
+deploy_doc "b33a385" v1.0.0
+deploy_doc "fe02e45" v1.1.0
+deploy_doc "89fd345" v1.2.0
+deploy_doc "fc9faa8" v2.0.0
+deploy_doc "3ddce1d" v2.1.1
+deploy_doc "3616209" v2.2.0
+deploy_doc "d0f8b9a" v2.3.0
+deploy_doc "6664ea9" v2.4.0
+deploy_doc "fb560dc" v2.5.0
+deploy_doc "b90745c" v2.5.1
+deploy_doc "fbc5bf1" v2.6.0
+deploy_doc "6f5a12a" v2.7.0
+deploy_doc "11c3257" v2.8.0
+deploy_doc "e7cfc1a" v2.9.0
+deploy_doc "7cb203f" v2.9.1
+deploy_doc "10d7239" v2.10.0
+deploy_doc "b42586e" v2.11.0
+deploy_doc "7fb8bdf" v3.0.2
+deploy_doc "4b3ee9c" v3.1.0
+deploy_doc "3ebb1b3" v3.2.0
+deploy_doc "0613f05" v3.3.1
+deploy_doc "eb0e0ce" v3.4.0
+deploy_doc "818878d" v3.5.1
+deploy_doc "c781171" v4.0.1
+deploy_doc "bfa4ccf" v4.1.1
+deploy_doc "7d9a9d0" v4.2.2
+deploy_doc "bae0c79" v4.3.3
+deploy_doc "c988db5" v4.4.0
+deploy_doc "c5d6a28" v4.4.1
+deploy_doc "6bc89ed" v4.4.2
+deploy_doc "4906a29" v4.5.0
+deploy_doc "4bae96e" v4.5.1
+deploy_doc "25dee4a" v4.6.0
+deploy_doc "7a6c9fa"  # v4.7.0 Latest stable release

.gitattributes

@@ -1,4 +1,3 @@
 *.py	eol=lf
 *.rst	eol=lf
-*.md	eol=lf
-*.mdx   eol=lf
+*.md	eol=lf

.github/ISSUE_TEMPLATE/---new-benchmark.md

@@ -0,0 +1,22 @@
+---
+name: "\U0001F5A5 New benchmark"
+about: Benchmark a part of this library and share your results
+title: "[Benchmark]"
+labels: ''
+assignees: ''
+
+---
+
+# 🖥 Benchmarking `transformers`
+
+## Benchmark
+
+Which part of `transformers` did you benchmark?
+
+## Set-up
+
+What did you run your benchmarks on? Please include details, such as: CPU, GPU? If using multiple GPUs, which parallelization did you use?
+
+## Results
+
+Put your results here!

.github/ISSUE_TEMPLATE/--new-model-addition.md

@@ -0,0 +1,20 @@
+---
+name: "\U0001F31F New model addition"
+about: Submit a proposal/request to implement a new Transformer-based model
+title: ''
+labels: New model
+assignees: ''
+
+---
+
+# 🌟 New model addition
+
+## Model description
+
+<!-- Important information -->
+
+## Open source status
+
+* [ ] the model implementation is available: (give details)
+* [ ] the model weights are available: (give details)
+* [ ] who are the authors: (mention them, if possible by @gh-username)

.github/ISSUE_TEMPLATE/bug-report.md

@@ -0,0 +1,94 @@
+---
+name: "\U0001F41B Bug Report"
+about: Submit a bug report to help us improve transformers
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+
+## Environment info
+<!-- You can run the command `transformers-cli env` and copy-and-paste its output below.
+     Don't forget to fill out the missing fields in that output! -->
+
+- `transformers` version:
+- Platform:
+- Python version:
+- PyTorch version (GPU?):
+- Tensorflow version (GPU?):
+- Using GPU in script?:
+- Using distributed or parallel set-up in script?:
+
+### Who can help
+<!-- Your issue will be replied to more quickly if you can figure out the right person to tag with @
+ If you know how to use git blame, that is the easiest way, otherwise, here is a rough guide of **who to tag**.
+ Please tag fewer than 3 people.
+
+Models:
+
+- albert, bert, xlm: @LysandreJik
+- blenderbot, bart, marian, pegasus, encoderdecoder,  t5: @patrickvonplaten, @patil-suraj
+- longformer, reformer, transfoxl, xlnet: @patrickvonplaten
+- fsmt: @stas00
+- funnel: @sgugger
+- gpt2: @patrickvonplaten, @LysandreJik
+- rag: @patrickvonplaten, @lhoestq
+- tensorflow: @Rocketknight1
+
+Library:
+
+- benchmarks: @patrickvonplaten
+- deepspeed: @stas00
+- ray/raytune: @richardliaw, @amogkam
+- text generation: @patrickvonplaten
+- tokenizers: @LysandreJik
+- trainer: @sgugger
+- pipelines: @LysandreJik
+
+Documentation: @sgugger
+
+Model hub:
+
+- for issues with a model report at https://discuss.huggingface.co/ and tag the model's creator.
+
+HF projects:
+
+- datasets: [different repo](https://github.com/huggingface/datasets)
+- rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
+
+Examples:
+
+- maintained examples (not research project or legacy): @sgugger, @patil-suraj
+- research_projects/bert-loses-patience: @JetRunner
+- research_projects/distillation: @VictorSanh
+
+ -->
+
+## Information
+
+Model I am using (Bert, XLNet ...):
+
+The problem arises when using:
+* [ ] the official example scripts: (give details below)
+* [ ] my own modified scripts: (give details below)
+
+The tasks I am working on is:
+* [ ] an official GLUE/SQUaD task: (give the name)
+* [ ] my own task or dataset: (give details below)
+
+## To reproduce
+
+Steps to reproduce the behavior:
+
+1.
+2.
+3.
+
+<!-- If you have code snippets, error messages, stack traces please provide them here as well.
+     Important! Use code tags to correctly format your code. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
+     Do not use screenshots, as they are hard to read and (more importantly) don't allow others to copy-and-paste your code.-->
+
+## Expected behavior
+
+<!-- A clear and concise description of what you would expect to happen. -->

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -1,118 +0,0 @@
-name: "\U0001F41B Bug Report"
-description: Submit a bug report to help us improve transformers
-body:
-  - type: textarea
-    id: system-info
-    attributes:
-      label: System Info
-      description: Please share your system info with us. You can run the command `transformers-cli env` and copy-paste its output below.
-      placeholder: transformers version, platform, python version, ...
-    validations:
-      required: true
-
-  - type: textarea
-    id: who-can-help
-    attributes:
-      label: Who can help?
-      description: |
-        Your issue will be replied to more quickly if you can figure out the right person to tag with @
-        If you know how to use git blame, that is the easiest way, otherwise, here is a rough guide of **who to tag**.
-        Please tag fewer than 3 people.
-        
-        Models:
-
-          - ALBERT, BERT, XLM, DeBERTa, DeBERTa-v2, ELECTRA, MobileBert, SqueezeBert: `@LysandreJik`
-          - T5, Pegasus, EncoderDecoder: `@patrickvonplaten`
-          - Blenderbot, MBART, BART, Marian, Pegasus: `@patil-suraj`
-          - Reformer, TransfoXL, XLNet, FNet: `@patrickvonplaten`
-          - Longformer, BigBird: `@ydshieh`
-          - FSMT: `@stas00`
-          - Funnel: `@sgugger`
-          - GPT-2, GPT: `@patil-suraj`, `@patrickvonplaten`, `@LysandreJik`
-          - RAG, DPR: `@patrickvonplaten`, `@lhoestq`
-          - TensorFlow: `@Rocketknight1`
-          - JAX/Flax: `@patil-suraj`
-          - TAPAS, LayoutLM, LayoutLMv2, LUKE, ViT, BEiT, DEiT, DETR, CANINE: `@NielsRogge`
-          - GPT-Neo, GPT-J, CLIP: `@patil-suraj`
-          - Wav2Vec2, HuBERT, UniSpeech, UniSpeechSAT, SEW, SEW-D: `@patrickvonplaten`, `@anton-l`
-          - SpeechEncoderDecoder, Speech2Text, Speech2Text2: `@sanchit-gandhi`, `@patrickvonplaten`, `@anton-l`
-          
-          If the model isn't in the list, ping `@LysandreJik` who will redirect you to the correct contributor.
-
-        Library:
-          - Benchmarks: `@patrickvonplaten`
-          - Deepspeed: `@stas00`
-          - Ray/raytune: `@richardliaw`, `@amogkam`
-          - Text generation: `@patrickvonplaten`, `@Narsil`, `@gante`
-          - Tokenizers: `@SaulLu`
-          - Trainer: `@sgugger`
-          - Pipelines: `@Narsil`
-          - Speech: `@patrickvonplaten`, `@anton-l`, `@sanchit-gandhi`
-          - Vision: `@NielsRogge`, `@sgugger`
-
-        Documentation: `@sgugger`, `@stevhliu`
-
-        Model hub:
-
-          - for issues with a model, report at https://discuss.huggingface.co/ and tag the model's creator.
-
-        HF projects:
-
-          - datasets: [different repo](https://github.com/huggingface/datasets)
-          - rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
-
-        Examples:
-
-          - maintained examples (not research project or legacy): `@sgugger`, `@patil-suraj`
-
-        For research projetcs, please ping the contributor directly. For example, on the following projects:
-
-          - research_projects/bert-loses-patience: `@JetRunner`
-          - research_projects/distillation: `@VictorSanh`
-      placeholder: "@Username ..."
-
-  - type: checkboxes
-    id: information-scripts-examples
-    attributes:
-      label: Information
-      description: 'The problem arises when using:'
-      options:
-        - label: "The official example scripts"
-        - label: "My own modified scripts"
-
-  - type: checkboxes
-    id: information-tasks
-    attributes:
-      label: Tasks
-      description: "The tasks I am working on are:"
-      options:
-        - label: "An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)"
-        - label: "My own task or dataset (give details below)"
-
-  - type: textarea
-    id: reproduction
-    validations:
-      required: true
-    attributes:
-      label: Reproduction
-      description: |
-        Please provide a code sample that reproduces the problem you ran into. It can be a Colab link or just a code snippet.
-        If you have code snippets, error messages, stack traces please provide them here as well.
-        Important! Use code tags to correctly format your code. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
-        Do not use screenshots, as they are hard to read and (more importantly) don't allow others to copy-and-paste your code.
-
-      placeholder: |
-        Steps to reproduce the behavior:
-          
-          1.
-          2.
-          3.
-          
-
-  - type: textarea
-    id: expected-behavior
-    validations:
-      required: true
-    attributes:
-      label: Expected behavior
-      description: "A clear and concise description of what you would expect to happen."

.github/ISSUE_TEMPLATE/config.yml

@@ -1,12 +0,0 @@
-blank_issues_enabled: true
-version: 2.1
-contact_links:
-  - name: Model checkpoints on the Hugging Face Hub
-    url: https://huggingface.co/models
-    about: Open a Pull request / Discussion related to a specific model checkpoint directly on the Hugging Face Hub
-  - name: Website Related
-    url: https://github.com/huggingface/hub-docs/issues
-    about: Feature requests and bug reports related to the website
-  - name: Forum
-    url: https://discuss.huggingface.co/
-    about: General usage questions and community discussions

.github/ISSUE_TEMPLATE/feature-request.md

@@ -0,0 +1,25 @@
+---
+name: "\U0001F680 Feature request"
+about: Submit a proposal/request for a new transformers feature
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+# 🚀 Feature request
+
+<!-- A clear and concise description of the feature proposal.
+     Please provide a link to the paper and code in case they exist. -->
+
+## Motivation
+
+<!-- Please outline the motivation for the proposal. Is your feature request
+     related to a problem? e.g., I'm always frustrated when [...]. If this is related
+     to another GitHub issue, please link here too. -->
+
+## Your contribution
+
+<!-- Is there any way that you could help, e.g. by submitting a PR?
+     Make sure to read the CONTRIBUTING.MD readme:
+     https://github.com/huggingface/transformers/blob/master/CONTRIBUTING.md -->

.github/ISSUE_TEMPLATE/feature-request.yml

@@ -1,31 +0,0 @@
-name: "\U0001F680 Feature request"
-description: Submit a proposal/request for a new transformers feature
-labels: [ "feature" ]
-body:
-  - type: textarea
-    id: feature-request
-    validations:
-      required: true
-    attributes:
-      label: Feature request
-      description: |
-        A clear and concise description of the feature proposal. Please provide a link to the paper and code in case they exist.
-
-  - type: textarea
-    id: motivation
-    validations:
-      required: true
-    attributes:
-      label: Motivation
-      description: |
-        Please outline the motivation for the proposal. Is your feature request related to a problem? e.g., I'm always frustrated when [...]. If this is related to another GitHub issue, please link here too.
-        
-
-  - type: textarea
-    id: contribution
-    validations:
-      required: true
-    attributes:
-      label: Your contribution
-      description: |
-        Is there any way that you could help, e.g. by submitting a PR? Make sure to read the CONTRIBUTING.MD [readme](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md)

.github/ISSUE_TEMPLATE/migration.md

@@ -0,0 +1,58 @@
+---
+name: "\U0001F4DA Migration from pytorch-pretrained-bert or pytorch-transformers"
+about: Report a problem when migrating from pytorch-pretrained-bert or pytorch-transformers
+  to transformers
+title: ''
+labels: Migration
+assignees: ''
+
+---
+
+# 📚 Migration
+
+## Information
+
+<!-- Important information -->
+
+Model I am using (Bert, XLNet ...):
+
+Language I am using the model on (English, Chinese ...):
+
+The problem arises when using:
+* [ ] the official example scripts: (give details below)
+* [ ] my own modified scripts: (give details below)
+
+The tasks I am working on is:
+* [ ] an official GLUE/SQUaD task: (give the name)
+* [ ] my own task or dataset: (give details below)
+
+## Details
+
+<!-- A clear and concise description of the migration issue.
+    If you have code snippets, please provide it here as well.
+    Important! Use code tags to correctly format your code. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
+    Do not use screenshots, as they are hard to read and (more importantly) don't allow others to copy-and-paste your code.
+    -->
+
+## Environment info
+<!-- You can run the command `python transformers-cli env` and copy-and-paste its output below.
+     Don't forget to fill out the missing fields in that output! -->
+ 
+- `transformers` version:
+- Platform:
+- Python version:
+- PyTorch version (GPU?):
+- Tensorflow version (GPU?):
+- Using GPU in script?:
+- Using distributed or parallel set-up in script?:
+
+<!-- IMPORTANT: which version of the former library do you use? -->
+* `pytorch-transformers` or `pytorch-pretrained-bert` version (or branch):
+
+
+## Checklist
+
+- [ ] I have read the migration guide in the readme.
+ ([pytorch-transformers](https://github.com/huggingface/transformers#migrating-from-pytorch-transformers-to-transformers);
+  [pytorch-pretrained-bert](https://github.com/huggingface/transformers#migrating-from-pytorch-pretrained-bert-to-transformers))
+- [ ] I checked if a related official extension example runs on my machine.

.github/ISSUE_TEMPLATE/migration.yml

@@ -1,72 +0,0 @@
-name: "\U0001F4DA Migration from pytorch-pretrained-bert or pytorch-transformers"
-description: Report a problem when migrating from pytorch-pretrained-bert or pytorch-transformers to transformers
-labels: [ "migration" ]
-body:
-  - type: textarea
-    id: system-info
-    attributes:
-      label: System Info
-      description: Please share your system info with us. You can run the command `transformers-cli env` and copy-paste its output below.
-      render: shell
-      placeholder: transformers version, platform, python version, ...
-    validations:
-      required: true
-
-  - type: checkboxes
-    id: information-scripts-examples
-    attributes:
-      label: Information
-      description: 'The problem arises when using:'
-      options:
-        - label: "The official example scripts"
-        - label: "My own modified scripts"
-
-  - type: checkboxes
-    id: information-tasks
-    attributes:
-      label: Tasks
-      description: "The tasks I am working on are:"
-      options:
-        - label: "An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)"
-        - label: "My own task or dataset (give details below)"
-
-  - type: textarea
-    id: reproduction
-    validations:
-      required: true
-    attributes:
-      label: Reproduction
-      description: |
-        Please provide a code sample that reproduces the problem you ran into. It can be a Colab link or just a code snippet.
-        If you have code snippets, error messages, stack traces please provide them here as well.
-        Important! Use code tags to correctly format your code. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
-        Do not use screenshots, as they are hard to read and (more importantly) don't allow others to copy-and-paste your code.
-
-      placeholder: |
-        Steps to reproduce the behavior:
-          
-          1.
-          2.
-          3.
-          
-
-  - type: textarea
-    id: expected-behavior
-    validations:
-      required: true
-    attributes:
-      label: Expected behavior
-      description: "A clear and concise description of what you would expect to happen."
-      render: shell
-
-  - type: checkboxes
-    id: checklist
-    attributes:
-      label: Checklist
-      options:
-        - label: "I have read the migration guide in the readme.
- ([pytorch-transformers](https://github.com/huggingface/transformers#migrating-from-pytorch-transformers-to-transformers);
-  [pytorch-pretrained-bert](https://github.com/huggingface/transformers#migrating-from-pytorch-pretrained-bert-to-transformers))"
-          required: true
-        - label: "I checked if a related official extension example runs on my machine."
-          required: true

.github/ISSUE_TEMPLATE/new-model-addition.yml

@@ -1,31 +0,0 @@
-name: "\U0001F31F New model addition"
-description: Submit a proposal/request to implement a new model
-labels: [ "New model" ]
-
-body:
-  - type: textarea
-    id: description-request
-    validations:
-      required: true
-    attributes:
-      label: Model description
-      description: |
-        Put any and all important information relative to the model
-
-  - type: checkboxes
-    id: information-tasks
-    attributes:
-      label: Open source status
-      description: |
-          Please note that if the model implementation isn't available or if the weights aren't open-source, we are less likely to implement it in `transformers`.
-      options:
-        - label: "The model implementation is available"
-        - label: "The model weights are available"
-
-  - type: textarea
-    id: additional-info
-    attributes:
-      label: Provide useful links for the implementation
-      description: |
-        Please provide information regarding the implementation, the weights, and the authors.
-        Please mention the authors by @gh-username if you're aware of their usernames.

.github/ISSUE_TEMPLATE/question-help.md

@@ -0,0 +1,26 @@
+---
+name: "❓ Questions & Help"
+about: Post your general questions on the Hugging Face forum: https://discuss.huggingface.co/
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+# ❓ Questions & Help
+
+<!-- The GitHub issue tracker is primarly intended for bugs, feature requests,
+     new models, benchmarks, and migration questions. For all other questions,
+     we direct you to the Hugging Face forum: https://discuss.huggingface.co/ .
+     -->
+
+## Details
+
+<!-- Description of your issue -->
+
+<!-- You should first ask your question on the forum, and only if
+     you didn't get an answer after a few days ask it here on GitHub. -->
+
+**A link to original question on the forum**:
+
+<!-- Your issue will be closed if you don't fill this part. -->

.github/PULL_REQUEST_TEMPLATE.md

@@ -17,13 +17,13 @@ Fixes # (issue)
 
 ## Before submitting
 - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
-- [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
+- [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/master/CONTRIBUTING.md#start-contributing-pull-requests),
       Pull Request section?
 - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link
       to it if that's the case.
 - [ ] Did you make sure to update the documentation with your changes? Here are the
-      [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and
-      [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
+      [documentation guidelines](https://github.com/huggingface/transformers/tree/master/docs), and
+      [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/master/docs#writing-source-documentation).
 - [ ] Did you write any new necessary tests?
 
 

.github/conda/meta.yaml

@@ -25,8 +25,8 @@ requirements:
     - sacremoses
     - regex !=2019.12.17
     - protobuf
-    - tokenizers >=0.11.1,!=0.11.3,<0.13
-    - pyyaml >=5.1
+    - tokenizers >=0.10.1,<0.11.0
+    - pyyaml
   run:
     - python
     - numpy >=1.17
@@ -40,8 +40,8 @@ requirements:
     - sacremoses
     - regex !=2019.12.17
     - protobuf
-    - tokenizers >=0.11.1,!=0.11.3,<0.13
-    - pyyaml >=5.1
+    - tokenizers >=0.10.1,<0.11.0
+    - pyyaml
 
 test:
   imports:

.github/workflows/TROUBLESHOOT.md

@@ -1,9 +0,0 @@
-# Troubleshooting
-
-This is a document explaining how to deal with various issues on github-actions self-hosted CI. The entries may include actually solutions or pointers to Issues that cover those.
-
-## GitHub Actions (self-hosted CI)
-
-* Deepspeed
-
-  - if jit build hangs, clear out `rm -rf ~/.cache/torch_extensions/` reference: https://github.com/huggingface/transformers/pull/12723

.github/workflows/add-model-like.yml

@@ -1,80 +0,0 @@
-name: Add model like runner
-
-on:
-  push:
-    branches:
-      - main
-  pull_request:
-    paths:
-      - "src/**"
-      - "tests/**"
-      - ".github/**"
-    types: [opened, synchronize, reopened]
-
-jobs:
-  run_tests_templates_like:
-    name: "Add new model like template tests"
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v2
-
-      - name: Install dependencies
-        run: |
-          sudo apt -y update && sudo apt install -y libsndfile1-dev
-
-      - name: Load cached virtual environment
-        uses: actions/cache@v2
-        id: cache
-        with:
-          path: ~/venv/
-          key: v4-tests_model_like-${{ hashFiles('setup.py') }}
-
-      - name: Create virtual environment on cache miss
-        if: steps.cache.outputs.cache-hit != 'true'
-        run: |
-          python -m venv ~/venv && . ~/venv/bin/activate
-          pip install --upgrade pip!=21.3
-          pip install -e .[dev]
-
-      - name: Check transformers location
-        # make `transformers` available as package (required since we use `-e` flag) and check it's indeed from the repo.
-        run: |
-          . ~/venv/bin/activate
-          python setup.py develop
-          transformers_install=$(pip list -e | grep transformers)
-          transformers_install_array=($transformers_install)
-          transformers_loc=${transformers_install_array[-1]}
-          transformers_repo_loc=$(pwd .)
-          if [ "$transformers_loc" != "$transformers_repo_loc" ]; then
-              echo "transformers is from $transformers_loc but it shoud be from $transformers_repo_loc/src."
-              echo "A fix is required. Stop testing."
-              exit 1
-          fi
-
-      - name: Create model files
-        run: |
-          . ~/venv/bin/activate
-          transformers-cli add-new-model-like --config_file tests/fixtures/add_distilbert_like_config.json --path_to_repo .
-          make style
-          make fix-copies
-
-      - name: Run all PyTorch modeling test
-        run: |
-          . ~/venv/bin/activate
-          python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_new_models tests/bert_new/test_modeling_bert_new.py
-
-      - name: Run style changes
-        run: |
-          . ~/venv/bin/activate
-          make style && make quality && make repo-consistency
-
-      - name: Failure short reports
-        if: ${{ always() }}
-        run: cat reports/tests_new_models/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: run_all_tests_new_models_test_reports
-          path: reports/tests_new_models

.github/workflows/build-docker-images.yml

@@ -1,236 +0,0 @@
-name: Build docker images (scheduled)
-
-on:
-  push:
-    branches:
-      - docker-image*
-  repository_dispatch:
-  workflow_call:
-    inputs:
-      image_postfix:
-        required: true
-        type: string
-  schedule:
-    - cron: "0 1 * * *"
-
-concurrency:
-  group: docker-images-builds
-  cancel-in-progress: false
-
-jobs:
-  latest-docker:
-    name: "Latest PyTorch + TensorFlow [dev]"
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v3
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-all-latest-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-all-latest-gpu${{ inputs.image_postfix }}
-      # Push CI images still need to be re-built daily
-      -
-        name: Build and push (for Push CI) in a daily basis
-        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
-        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
-        if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-all-latest-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-all-latest-gpu-push-ci
-
-  latest-with-torch-nightly-docker:
-    name: "Nightly PyTorch + Stable TensorFlow"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-all-latest-gpu
-          build-args: |
-            REF=main
-            PYTORCH=pre
-          push: true
-          tags: huggingface/transformers-all-latest-torch-nightly-gpu
-
-  latest-torch-deepspeed-docker:
-    name: "Latest PyTorch + DeepSpeed"
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-latest-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-latest-gpu${{ inputs.image_postfix }}
-      # Push CI images still need to be re-built daily
-      -
-        name: Build and push (for Push CI) in a daily basis
-        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
-        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
-        if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-latest-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
-
-  nightly-torch-deepspeed-docker:
-    name: "Nightly PyTorch + DeepSpeed"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-nightly-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-nightly-gpu
-
-  doc-builder:
-    name: "Doc builder"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-doc-builder
-          push: true
-          tags: huggingface/transformers-doc-builder
-
-  latest-pytorch:
-    name: "Latest PyTorch [dev]"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-pytorch-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-gpu
-
-  latest-tensorflow:
-    name: "Latest TensorFlow [dev]"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-tensorflow-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-tensorflow-gpu

.github/workflows/build-past-ci-docker-images.yml

@@ -1,108 +0,0 @@
-name: Build docker images (Past CI)
-
-on:
-  push:
-    branches:
-      - past-ci-docker-image*
-
-concurrency:
-  group: docker-images-builds
-  cancel-in-progress: false
-
-jobs:
-  past-pytorch-docker:
-    name: "Past PyTorch Docker"
-    strategy:
-      fail-fast: false
-      matrix:
-        version: ["1.11", "1.10", "1.9", "1.8", "1.7", "1.6", "1.5", "1.4"]
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-past-gpu
-          build-args: |
-            REF=main
-            FRAMEWORK=pytorch
-            VERSION=${{ matrix.version }}
-          push: true
-          tags: huggingface/transformers-pytorch-past-${{ matrix.version }}-gpu
-
-  past-tensorflow-docker:
-    name: "Past TensorFlow Docker"
-    strategy:
-      fail-fast: false
-      matrix:
-        version: ["2.8", "2.7", "2.6", "2.5"]
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-past-gpu
-          build-args: |
-            REF=main
-            FRAMEWORK=tensorflow
-            VERSION=${{ matrix.version }}
-          push: true
-          tags: huggingface/transformers-tensorflow-past-${{ matrix.version }}-gpu
-
-  past-tensorflow-docker-2-4:
-    name: "Past TensorFlow Docker"
-    strategy:
-      fail-fast: false
-      matrix:
-        version: ["2.4"]
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v1
-      -
-        name: Check out code
-        uses: actions/checkout@v2
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v1
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v2
-        with:
-          context: ./docker/transformers-past-gpu
-          build-args: |
-            REF=main
-            BASE_DOCKER_IMAGE=nvidia/cuda:11.0.3-cudnn8-devel-ubuntu20.04
-            FRAMEWORK=tensorflow
-            VERSION=${{ matrix.version }}
-          push: true
-          tags: huggingface/transformers-tensorflow-past-${{ matrix.version }}-gpu

.github/workflows/build_documentation.yml

@@ -1,20 +0,0 @@
-name: Build documentation
-
-on:
-  push:
-    branches:
-      - main
-      - doc-builder*
-      - v*-release
-      - use_templates
-
-jobs:
-   build:
-    uses: huggingface/doc-builder/.github/workflows/build_main_documentation.yml@main
-    with:
-      commit_sha: ${{ github.sha }}
-      package: transformers
-      notebook_folder: transformers_doc
-      languages: de en es it pt
-    secrets:
-      token: ${{ secrets.HUGGINGFACE_PUSH }}

.github/workflows/build_pr_documentation.yml

@@ -1,17 +0,0 @@
-name: Build PR Documentation
-
-on:
-  pull_request:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  build:
-    uses: huggingface/doc-builder/.github/workflows/build_pr_documentation.yml@main
-    with:
-      commit_sha: ${{ github.event.pull_request.head.sha }}
-      pr_number: ${{ github.event.number }}
-      package: transformers
-      languages: de en es it pt

.github/workflows/check_runner_status.yml

@@ -1,67 +0,0 @@
-name: Self-hosted runner (check runner status)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  repository_dispatch:
-  schedule:
-    # run per hour
-    - cron: "0 */1 * * *"
-
-env:
-  TRANSFORMERS_IS_CI: yes
-
-jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-latest
-    outputs:
-      offline_runners: ${{ steps.set-offline_runners.outputs.offline_runners }}
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-ci-runner-docker,multi-gpu-ci-runner-docker,single-gpu-scheduled-ci-runner-docker,multi-scheduled-scheduled-ci-runner-docker,single-gpu-doctest-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-      - id: set-offline_runners
-        name: Set output for offline runners
-        if: ${{ always() }}
-        run: |
-          offline_runners=$(python3 -c 'fp = open("offline_runners.txt"); failed = fp.read(); fp.close(); print(failed)')
-          echo "::set-output name=offline_runners::$offline_runners"
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-latest
-    needs: check_runner_status
-    if: ${{ failure() }}
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-
-      - uses: actions/checkout@v2
-      - uses: actions/download-artifact@v2
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_EVENT: runner status check
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          OFFLINE_RUNNERS: ${{ needs.check_runner_status.outputs.offline_runners }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          python utils/notification_service.py

.github/workflows/delete_doc_comment.yml

@@ -1,13 +0,0 @@
-name: Delete dev documentation
-
-on:
-  pull_request:
-    types: [ closed ]
-
-
-jobs:
-  delete:
-    uses: huggingface/doc-builder/.github/workflows/delete_doc_comment.yml@main
-    with:
-      pr_number: ${{ github.event.number }}
-      package: transformers

.github/workflows/doctests.yml

@@ -1,78 +0,0 @@
-name: Doctests
-
-on:
-  push:
-    branches:
-      - doctest*
-  repository_dispatch:
-  schedule:
-    - cron: "0 0 * * *"
-
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  RUN_SLOW: yes
-  OMP_NUM_THREADS: 16
-  MKL_NUM_THREADS: 16
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-
-jobs:
-  run_doctests:
-    runs-on: [self-hosted, doc-tests-gpu]
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - uses: actions/checkout@v2
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: GPU visibility
-        run: |
-          python3 utils/print_env.py
-
-      - name: Prepare files for doctests
-        run: |
-          python3 utils/prepare_for_doc_test.py src docs
-
-      - name: Run doctests
-        run: |
-          python3 -m pytest -v --make-reports doc_tests_gpu --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.mdx"
-
-      - name: Clean files after doctests
-        run: |
-          python3 utils/prepare_for_doc_test.py src docs --remove_new_line
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat reports/doc_tests_gpu/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: doc_tests_gpu_test_reports
-          path: reports/doc_tests_gpu
-
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-latest
-    if: always()
-    needs: [run_doctests]
-    steps:
-      - uses: actions/checkout@v2
-      - uses: actions/download-artifact@v2
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-        run: |
-          pip install slack_sdk
-          python utils/notification_service_doc_tests.py

.github/workflows/github-torch-hub.yml

@@ -0,0 +1,46 @@
+name: Torch hub integration
+
+on:
+  push:
+    branches:
+      - "*"
+
+jobs:
+  torch_hub_integration:
+    runs-on: ubuntu-latest
+    env:
+      # TODO quickfix but may need more investigation
+      ACTIONS_ALLOW_UNSECURE_COMMANDS: True
+    steps:
+    # no checkout necessary here.
+    - name: Extract branch name
+      run: echo "::set-env name=BRANCH::${GITHUB_REF#refs/heads/}"
+    - name: Check branch name
+      run: echo $BRANCH
+    - name: Set up Python
+      uses: actions/setup-python@v1
+      with:
+        python-version: 3.7
+
+    - name: Loading cache
+      uses: actions/cache@v2
+      id: cache
+      with:
+        path: ~/.cache/pip
+        key: v0-torch_hub-${{ hashFiles('setup.py') }}
+
+    - name: Install dependencies
+      run: |
+        pip install --upgrade pip
+        # install torch-hub specific dependencies
+        pip install -e git+https://github.com/huggingface/transformers.git#egg=transformers[torchhub]
+        # no longer needed
+        pip uninstall -y transformers
+
+    #- name: Torch hub list
+    #  run: |
+    #    python -c "import torch; print(torch.hub.list('huggingface/transformers:$BRANCH'))"
+
+    #- name: Torch hub help
+    #  run: |
+    #    python -c "import torch; print(torch.hub.help('huggingface/transformers:$BRANCH', 'modelForSequenceClassification'))"

.github/workflows/model-templates.yml

@@ -1,59 +1,52 @@
 name: Model templates runner
 
 on:
-  repository_dispatch:
-  schedule:
-    - cron: "0 2 * * *"
+  push:
+    branches:
+      - master
+  pull_request:
+    paths:
+      - "src/**"
+      - "tests/**"
+      - ".github/**"
+      - "templates/**"
+    types: [assigned, opened, synchronize, reopened]
 
 jobs:
   run_tests_templates:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v2
+        uses: actions/checkout@v1
 
-      - name: Install dependencies
-        run: |
-          sudo apt -y update && sudo apt install -y libsndfile1-dev
+      - name: Install Python
+        uses: actions/setup-python@v1
+        with:
+          python-version: 3.6
 
-      - name: Load cached virtual environment
+      - name: Loading cache.
         uses: actions/cache@v2
         id: cache
         with:
-          path: ~/venv/
-          key: v4-tests_templates-${{ hashFiles('setup.py') }}
+          path: ~/.cache/pip
+          key: v1.2-tests_templates
+          restore-keys: |
+            v1.2-tests_templates-${{ hashFiles('setup.py') }}
+            v1.2-tests_templates
 
-      - name: Create virtual environment on cache miss
-        if: steps.cache.outputs.cache-hit != 'true'
+      - name: Install dependencies
         run: |
-          python -m venv ~/venv && . ~/venv/bin/activate
-          pip install --upgrade pip!=21.3
-          pip install -e .[dev]
-
-      - name: Check transformers location
-        # make `transformers` available as package (required since we use `-e` flag) and check it's indeed from the repo.
-        run: |
-          . ~/venv/bin/activate
-          python setup.py develop
-          transformer_loc=$(pip show transformers | grep "Location: " | cut -c11-)
-          transformer_repo_loc=$(pwd .)
-          if [ "$transformer_loc" != "$transformer_repo_loc/src" ]; then
-              echo "transformers is from $transformer_loc but it shoud be from $transformer_repo_loc/src."
-              echo "A fix is required. Stop testing."
-              exit 1
-          fi
-
+          pip install --upgrade pip
+          sudo apt -y update && sudo apt install -y libsndfile1-dev
+          pip install .[dev]
       - name: Create model files
         run: |
-          . ~/venv/bin/activate
           transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/encoder-bert-tokenizer.json --path=templates/adding_a_new_model
           transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/pt-encoder-bert-tokenizer.json --path=templates/adding_a_new_model
           transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/standalone.json --path=templates/adding_a_new_model
           transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/tf-encoder-bert-tokenizer.json --path=templates/adding_a_new_model
           transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/tf-seq-2-seq-bart-tokenizer.json --path=templates/adding_a_new_model
           transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/pt-seq-2-seq-bart-tokenizer.json --path=templates/adding_a_new_model
-          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/flax-encoder-bert-tokenizer.json --path=templates/adding_a_new_model
-          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/flax-seq-2-seq-bart-tokenizer.json --path=templates/adding_a_new_model
           make style
           python utils/check_table.py --fix_and_overwrite
           python utils/check_dummies.py --fix_and_overwrite
@@ -61,21 +54,20 @@ jobs:
 
       - name: Run all non-slow tests
         run: |
-          . ~/venv/bin/activate
           python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_templates tests/*template*
 
       - name: Run style changes
         run: |
-          . ~/venv/bin/activate
-          make style && make quality && make repo-consistency
+          git fetch origin master:master
+          make fixup
 
       - name: Failure short reports
         if: ${{ always() }}
-        run: cat reports/tests_templates/failures_short.txt
+        run: cat reports/tests_templates_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
           name: run_all_tests_templates_test_reports
-          path: reports/tests_templates
+          path: reports

.github/workflows/self-nightly-scheduled.yml

@@ -1,286 +0,0 @@
-name: Self-hosted runner (nightly)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  repository_dispatch:
-# Disable temporarily until the test suite can be run under 12 hours.
-#  schedule:
-#    - cron: "0 16 * * *"
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-
-jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-scheduled-ci-runner-docker,multi-gpu-scheduled-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  setup:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - id: set-matrix
-        name: Identify models to test
-        working-directory: /transformers/tests
-        run: |
-          echo "::set-output name=matrix::$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')"
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_tests_multi_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_all_tests_torch_cuda_extensions_gpu:
-    name: Torch CUDA extension tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    needs: setup
-    container:
-      image: huggingface/transformers-pytorch-deepspeed-nightly-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /workspace/transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /workspace/transformers
-        run: |
-          python utils/print_env.py
-
-      - name: Run all tests on GPU
-        working-directory: /workspace/transformers
-        run: |
-          python -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-latest
-    if: always()
-    needs: [
-      check_runner_status,
-      check_runners,
-      setup,
-      run_tests_single_gpu,
-      run_tests_multi_gpu,
-      run_all_tests_torch_cuda_extensions_gpu
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-          echo "Runner status: ${{ needs.check_runners.result }}"
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v2
-      - uses: actions/download-artifact@v2
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
-          CI_EVENT: nightly-build
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"

.github/workflows/self-past-caller.yml

@@ -1,136 +0,0 @@
-name: Self-hosted runner (past-ci-caller)
-
-on:
-  push:
-    branches:
-      - run-past-ci*
-
-jobs:
-  run_past_ci_pytorch_1-11:
-    name: PyTorch 1.11
-    if: always()
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.11"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-10:
-    name: PyTorch 1.10
-    if: always()
-    needs: [run_past_ci_pytorch_1-11]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.10"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-9:
-    name: PyTorch 1.9
-    if: always()
-    needs: [run_past_ci_pytorch_1-10]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.9"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-8:
-    name: PyTorch 1.8
-    if: always()
-    needs: [run_past_ci_pytorch_1-9]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.8"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-7:
-    name: PyTorch 1.7
-    if: always()
-    needs: [run_past_ci_pytorch_1-8]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.7"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-6:
-    name: PyTorch 1.6
-    if: always()
-    needs: [run_past_ci_pytorch_1-7]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.6"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-5:
-    name: PyTorch 1.5
-    if: always()
-    needs: [run_past_ci_pytorch_1-6]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.5"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-4:
-    name: PyTorch 1.4
-    if: always()
-    needs: [run_past_ci_pytorch_1-5]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.4"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-8:
-    name: TensorFlow 2.8
-    if: always()
-    needs: [run_past_ci_pytorch_1-4]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.8"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-7:
-    name: TensorFlow 2.7
-    if: always()
-    needs: [run_past_ci_tensorflow_2-8]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.7"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-6:
-    name: TensorFlow 2.6
-    if: always()
-    needs: [run_past_ci_tensorflow_2-7]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.6"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-5:
-    name: TensorFlow 2.5
-    if: always()
-    needs: [run_past_ci_tensorflow_2-6]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.5"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-4:
-    name: TensorFlow 2.4
-    if: always()
-    needs: [run_past_ci_tensorflow_2-5]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.4"
-    secrets: inherit

.github/workflows/self-past.yml

@@ -1,261 +0,0 @@
-name: Self-hosted runner (past)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  workflow_call:
-    inputs:
-      framework:
-        required: true
-        type: string
-      version:
-        required: true
-        type: string
-      # Use this to control the commit to test against
-      sha:
-        default: 'main'
-        required: false
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-
-jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-past-ci-runner-docker,multi-gpu-past-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  setup:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - id: set-matrix
-        working-directory: /transformers
-        name: Identify models to test
-        run: |
-          cd tests
-          echo "::set-output name=matrix::$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')"
-
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Save job name
-        if: ${{ always() }}
-        shell: bash
-        run: |
-          matrix_folders=${matrix_folders/'models_'/'models/'}
-          job_name="Model tests ($matrix_folders, ${{ matrix.machine_type }})"
-          echo "$job_name"
-          echo "$job_name" > /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/job_name.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_tests_multi_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Save job name
-        if: ${{ always() }}
-        shell: bash
-        run: |
-          matrix_folders=${matrix_folders/'models_'/'models/'}
-          job_name="Model tests ($matrix_folders, ${{ matrix.machine_type }})"
-          echo "$job_name"
-          echo "$job_name" > /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/job_name.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-latest
-    if: always()
-    needs: [check_runner_status, check_runners, setup, run_tests_single_gpu, run_tests_multi_gpu]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-          echo "Runner status: ${{ needs.check_runners.result }}"
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v2
-      - uses: actions/download-artifact@v2
-
-      # Create a directory to store test failure tables in the next step
-      - name: Create directory
-        run: mkdir test_failure_tables
-
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
-          CI_EVENT: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
-
-      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
-      - name: Failure table artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_failure_tables_${{ inputs.framework }}-${{ inputs.version }}
-          path: test_failure_tables

.github/workflows/self-push-caller.yml

@@ -1,54 +0,0 @@
-# Used to trigger self-push CI
-name: Self-hosted runner (push-caller)
-
-on:
-  push:
-    branches:
-      - main
-    paths:
-      - "src/**"
-      - "tests/**"
-      - ".github/**"
-      - "templates/**"
-      - "utils/**"
-
-jobs:
-  check-for-setup:
-      runs-on: ubuntu-latest
-      name: Check if setup was changed
-      outputs:
-        changed: ${{ steps.was_changed.outputs.changed }}
-      steps:
-        - uses: actions/checkout@v3
-          with: 
-            fetch-depth: "2"
-        
-        - name: Get changed files
-          id: changed-files
-          uses: tj-actions/changed-files@v22.2
-        
-        - name: Was setup changed 
-          id: was_changed
-          run: |
-            for file in ${{ steps.changed-files.outputs.all_changed_files }}; do
-              if [ `basename "${file}"` = "setup.py" ]; then
-                echo ::set-output name=changed::"1"
-              fi
-            done
-
-  build-docker-containers:
-    needs: check-for-setup
-    if: (github.event_name == 'push') && (needs.check-for-setup.outputs.changed == '1')
-    uses: ./.github/workflows/build-docker-images.yml
-    with:
-      image_postfix: "-push-ci"
-    secrets: inherit
-
-  run_push_ci:
-    name: Trigger Push CI
-    runs-on: ubuntu-latest
-    if: ${{ always() }}
-    needs: build-docker-containers
-    steps:
-      - name: Trigger push CI via workflow_run
-        run: echo "Trigger push CI via workflow_run"

.github/workflows/self-push.yml

@@ -1,12 +1,9 @@
 name: Self-hosted runner (push)
 
 on:
-  workflow_run:
-    workflows: ["Self-hosted runner (push-caller)"]
-    branches: ["main"]
-    types: [completed]
   push:
     branches:
+      - master
       - ci_*
       - ci-*
     paths:
@@ -14,7 +11,6 @@ on:
       - "tests/**"
       - ".github/**"
       - "templates/**"
-      - "utils/**"
   repository_dispatch:
 
 env:
@@ -22,542 +18,282 @@ env:
   TRANSFORMERS_IS_CI: yes
   OMP_NUM_THREADS: 8
   MKL_NUM_THREADS: 8
-  PYTEST_TIMEOUT: 60
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
 
 jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-latest
+  run_tests_torch_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      - name: Checkout transformers
+      - name: Launcher docker
         uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-ci-runner-docker,multi-gpu-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
-    container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  setup:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
-    container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-      test_map: ${{ steps.set-matrix.outputs.test_map }}
-    steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # `CI_BRANCH_PUSH`: The branch name from the push event
-        # `CI_BRANCH_WORKFLOW_RUN`: The name of the branch on which this workflow is triggered by `workflow_run` event
-        # `CI_BRANCH`: The non-empty branch name from the above two (one and only one of them is empty)
-        # `CI_SHA_PUSH`: The commit SHA from the push event
-        # `CI_SHA_WORKFLOW_RUN`: The commit SHA that triggers this workflow by `workflow_run` event
-        # `CI_SHA`: The non-empty commit SHA from the above two (one and only one of them is empty)
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Fetch the tests to run
-        working-directory: /transformers
-        # TODO: add `git-python` in the docker images
-        run: |
-          pip install --upgrade git-python
-          python3 utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_fetched
-          path: /transformers/test_preparation.txt
-
-      - id: set-matrix
-        name: Organize tests into models
-        working-directory: /transformers
-        # The `keys` is used as GitHub actions matrix for jobs, i.e. `models/bert`, `tokenization`, `pipeline`, etc.
-        # The `test_map` is used to get the actual identified test files under each key.
-        # If no test to run (so no `test_map.json` file), create a dummy map (empty matrix will fail)
-        run: |
-          if [ -f test_map.json ]; then
-              keys=$(python3 -c 'import json; fp = open("test_map.json"); test_map = json.load(fp); fp.close(); d = list(test_map.keys()); print(d)')
-              test_map=$(python3 -c 'import json; fp = open("test_map.json"); test_map = json.load(fp); fp.close(); print(test_map)')
-          else
-              keys=$(python3 -c 'keys = ["dummy"]; print(keys)')
-              test_map=$(python3 -c 'test_map = {"dummy": []}; print(test_map)')
-          fi
-          echo $keys
-          echo $test_map
-          echo "::set-output name=matrix::$keys"
-          echo "::set-output name=test_map::$test_map"
-
-  run_tests_single_gpu:
-    name: Model tests
-    needs: setup
-    # `dummy` means there is no test to run
-    if: contains(fromJson(needs.setup.outputs.matrix), 'dummy') != true
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
-    container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          echo "${{ fromJson(needs.setup.outputs.test_map)[matrix.folders] }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
-      - name: Environment
-        working-directory: /transformers
+      - name: Install dependencies
         run: |
-          python3 utils/print_env.py
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
 
-      - name: Run all non-slow selected tests on GPU
-        working-directory: /transformers
+      - name: Are GPUs recognized by our DL frameworks
         run: |
-          python3 -m pytest -n 2 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} ${{ fromJson(needs.setup.outputs.test_map)[matrix.folders] }}
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all non-slow tests on GPU
+        run: |
+          python -m pytest -n 2 --dist=loadfile --make-reports=tests_torch_gpu tests
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+        if: ${{ always() }}
+        run: cat reports/tests_torch_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+          name: run_all_tests_torch_gpu_test_reports
+          path: reports
 
-  run_tests_multi_gpu:
-    name: Model tests
-    needs: setup
-    # `dummy` means there is no test to run
-    if: contains(fromJson(needs.setup.outputs.matrix), 'dummy') != true
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+  run_tests_tf_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    timeout-minutes: 120
     container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          echo "${{ fromJson(needs.setup.outputs.test_map)[matrix.folders] }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+      - name: Launcher docker
+        uses: actions/checkout@v2
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
-      - name: Environment
-        working-directory: /transformers
+      - name: Install dependencies
         run: |
-          python3 utils/print_env.py
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece]
 
-      - name: Run all non-slow selected tests on GPU
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
+
+      - name: Run all non-slow tests on GPU
+        env:
+          TF_NUM_INTRAOP_THREADS: 8
+          TF_NUM_INTEROP_THREADS: 1
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_tf_gpu_test_reports
+          path: reports
+
+
+  run_tests_torch_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all non-slow tests on GPU
         env:
           MKL_SERVICE_FORCE_INTEL: 1
-        working-directory: /transformers
         run: |
-          python3 -m pytest -n 2 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} ${{ fromJson(needs.setup.outputs.test_map)[matrix.folders] }}
+          python -m pytest -n 2 --dist=loadfile --make-reports=tests_torch_multi_gpu tests
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+        if: ${{ always() }}
+        run: cat reports/tests_torch_multi_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+          name: run_all_tests_torch_multi_gpu_test_reports
+          path: reports
 
-  run_tests_torch_cuda_extensions_single_gpu:
-    name: Torch CUDA extension tests
-    needs: setup
-    if: contains(fromJson(needs.setup.outputs.matrix), 'deepspeed') || contains(fromJson(needs.setup.outputs.matrix), 'extended')
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+  run_tests_tf_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    timeout-minutes: 120
     container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /workspace/transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /workspace/transformers
-        run: |
-          python utils/print_env.py
-
-      - name: Run all non-slow selected tests on GPU
-        working-directory: /workspace/transformers
-        # TODO: Here we pass all tests in the 2 folders for simplicity. It's better to pass only the identified tests.
-        run: |
-          python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
-
-  run_tests_torch_cuda_extensions_multi_gpu:
-    name: Torch CUDA extension tests
-    needs: setup
-    if: contains(fromJson(needs.setup.outputs.matrix), 'deepspeed') || contains(fromJson(needs.setup.outputs.matrix), 'extended')
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
-    container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
+      image: tensorflow/tensorflow:2.4.1-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /workspace/transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+      - name: Launcher docker
+        uses: actions/checkout@v2
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
-      - name: Environment
-        working-directory: /workspace/transformers
+      - name: Install dependencies
         run: |
-          python utils/print_env.py
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece]
 
-      - name: Run all non-slow selected tests on GPU
-        working-directory: /workspace/transformers
-        # TODO: Here we pass all tests in the 2 folders for simplicity. It's better to pass only the identified tests.
+      - name: Are GPUs recognized by our DL frameworks
         run: |
-          python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
+
+      - name: Run all non-slow tests on GPU
+        env:
+          TF_NUM_INTRAOP_THREADS: 8
+          TF_NUM_INTEROP_THREADS: 1
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu tests
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
+        if: ${{ always() }}
+        run: cat reports/tests_tf_multi_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
+          name: run_all_tests_tf_multi_gpu_test_reports
+          path: reports
+
+  run_tests_torch_cuda_extensions_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_gpu_test_reports
+          path: reports
+
+  run_tests_torch_cuda_extensions_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed,fairscale]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_multi_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_multi_gpu_test_reports
+          path: reports
+
 
   send_results:
     name: Send results to webhook
     runs-on: ubuntu-latest
     if: always()
     needs: [
-        check_runner_status,
-        check_runners,
-        setup,
-        run_tests_single_gpu,
-        run_tests_multi_gpu,
-        run_tests_torch_cuda_extensions_single_gpu,
+        run_tests_torch_gpu,
+        run_tests_tf_gpu,
+        run_tests_torch_multi_gpu,
+        run_tests_tf_multi_gpu,
+        run_tests_torch_cuda_extensions_gpu,
         run_tests_torch_cuda_extensions_multi_gpu
     ]
     steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-          echo "Setup status: ${{ needs.setup.result }}"
-          echo "Runner status: ${{ needs.check_runners.result }}"
-
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
       - uses: actions/checkout@v2
-        # To avoid failure when multiple commits are merged into `main` in a short period of time.
-        # Checking out to an old commit beyond the fetch depth will get an error `fatal: reference is not a tree: ...
-        # (Only required for `workflow_run` event, where we get the latest HEAD on `main` instead of the event commit)
-        with:
-          fetch-depth: 20
-
-      - name: Update clone using environment variables
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
 
       - uses: actions/download-artifact@v2
+
       - name: Send message to Slack
         env:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
           CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_EVENT: push
-          CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
-          CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
-          CI_SHA: ${{ env.CI_SHA }}
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup.result }}
 
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
         run: |
           pip install slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
+          python utils/notification_service.py push

.github/workflows/self-scheduled.yml

@@ -1,410 +1,356 @@
 name: Self-hosted runner (scheduled)
 
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
 on:
+  push:
+    branches:
+      - multi_ci_*
   repository_dispatch:
   schedule:
-    - cron: "0 2 * * *"
+    - cron: "0 0 * * *"
 
 env:
   HF_HOME: /mnt/cache
   TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
   RUN_SLOW: yes
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
+  OMP_NUM_THREADS: 16
+  MKL_NUM_THREADS: 16
 
 jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-latest
+  run_all_tests_torch_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      - name: Checkout transformers
+      - name: Launcher docker
         uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-scheduled-ci-runner-docker,multi-gpu-scheduled-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  setup:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - id: set-matrix
-        name: Identify models to test
-        working-directory: /transformers/tests
-        run: |
-          echo "::set-output name=matrix::$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')"
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
+      - name: Install dependencies
         run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
 
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: NVIDIA-SMI
+      - name: Are GPUs recognized by our DL frameworks
         run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
 
       - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_gpu tests
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_tests_multi_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_examples_gpu:
-    name: Examples directory
-    runs-on: [self-hosted, single-gpu-docker]
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
+        run: cat reports/tests_torch_gpu_failures_short.txt
 
       - name: Run examples tests on GPU
-        working-directory: /transformers
+        if: ${{ always() }}
+        env:
+          OMP_NUM_THREADS: 16
+          MKL_NUM_THREADS: 16
+          RUN_SLOW: yes
+          HF_HOME: /mnt/cache
+          TRANSFORMERS_IS_CI: yes
         run: |
           pip install -r examples/pytorch/_tests_requirements.txt
-          python3 -m pytest -v --make-reports=single-gpu_examples_gpu examples/pytorch
+          python -m pytest -n 1 --dist=loadfile --make-reports=examples_torch_gpu examples
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/single-gpu_examples_gpu/failures_short.txt
-
-      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: single-gpu_run_examples_gpu
-          path: /transformers/reports/single-gpu_examples_gpu
-
-  run_pipelines_torch_gpu:
-    name: PyTorch pipelines
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    container:
-      image: huggingface/transformers-pytorch-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
+        run: cat reports/examples_torch_gpu_failures_short.txt
 
       - name: Run all pipeline tests on GPU
-        working-directory: /transformers
+        if: ${{ always() }}
+        env:
+          RUN_PIPELINE_TESTS: yes
         run: |
-          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_tests_torch_pipeline_gpu tests/pipelines
+          python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_gpu tests
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_torch_pipeline_gpu/failures_short.txt
+        if: ${{ always() }}
+        run: cat reports/tests_torch_pipeline_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_tests_torch_pipeline_gpu
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_torch_pipeline_gpu
+          name: run_all_tests_torch_gpu_test_reports
+          path: reports
 
-  run_pipelines_tf_gpu:
-    name: TensorFlow pipelines
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+  run_all_tests_tf_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
-      image: huggingface/transformers-tensorflow-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
+      - name: Launcher docker
+        uses: actions/checkout@v2
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
-      - name: Environment
-        working-directory: /transformers
+      - name: Install dependencies
         run: |
-          python3 utils/print_env.py
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnx,sentencepiece]
 
-      - name: Run all pipeline tests on GPU
-        working-directory: /transformers
+      - name: Are GPUs recognized by our DL frameworks
         run: |
-          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_tests_tf_pipeline_gpu tests/pipelines
-
-      - name: Failure short reports
-        if: ${{ always() }}
-        run: |
-          cat /transformers/reports/${{ matrix.machine_type }}_tests_tf_pipeline_gpu/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: ${{ matrix.machine_type }}_run_tests_tf_pipeline_gpu
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_tf_pipeline_gpu
-
-  run_all_tests_torch_cuda_extensions_gpu:
-    name: Torch CUDA extension tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
-    needs: setup
-    container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /workspace/transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /workspace/transformers
-        run: |
-          python utils/print_env.py
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
 
       - name: Run all tests on GPU
-        working-directory: /workspace/transformers
+        env:
+          TF_NUM_INTEROP_THREADS: 1
+          TF_NUM_INTRAOP_THREADS: 16
         run: |
-          python -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu tests
 
       - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
+        if: ${{ always() }}
+        run: cat reports/tests_tf_gpu_failures_short.txt
+
+      - name: Run all pipeline tests on GPU
+        if: ${{ always() }}
+        env:
+          RUN_PIPELINE_TESTS: yes
+          TF_NUM_INTEROP_THREADS: 1
+          TF_NUM_INTRAOP_THREADS: 16
+        run: |
+          python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_tf_pipeline_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_pipeline_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
+          name: run_all_tests_tf_gpu_test_reports
+          path: reports
+
+  run_all_tests_torch_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        env:
+          MKL_SERVICE_FORCE_INTEL: 1
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_multi_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_multi_gpu_failures_short.txt
+
+      - name: Run all pipeline tests on GPU
+        if: ${{ always() }}
+        env:
+          RUN_PIPELINE_TESTS: yes
+        run: |
+          python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_multi_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_pipeline_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_torch_multi_gpu_test_reports
+          path: reports
+
+  run_all_tests_tf_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnx,sentencepiece]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
+          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
+
+      - name: Run all tests on GPU
+        env:
+          TF_NUM_INTEROP_THREADS: 1
+          TF_NUM_INTRAOP_THREADS: 16
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_multi_gpu_failures_short.txt
+
+      - name: Run all pipeline tests on GPU
+        if: ${{ always() }}
+        env:
+          RUN_PIPELINE_TESTS: yes
+          TF_NUM_INTEROP_THREADS: 1
+          TF_NUM_INTRAOP_THREADS: 16
+        run: |
+          python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_tf_pipeline_multi_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_pipeline_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_tf_multi_gpu_test_reports
+          path: reports
+
+  run_all_tests_torch_cuda_extensions_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_gpu_test_reports
+          path: reports
+
+  run_all_tests_torch_cuda_extensions_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed,fairscale]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_multi_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_multi_gpu_test_reports
+          path: reports
 
   send_results:
     name: Send results to webhook
     runs-on: ubuntu-latest
     if: always()
     needs: [
-      check_runner_status,
-      check_runners,
-      setup,
-      run_tests_single_gpu,
-      run_tests_multi_gpu,
-      run_examples_gpu,
-      run_pipelines_tf_gpu,
-      run_pipelines_torch_gpu,
-      run_all_tests_torch_cuda_extensions_gpu
+        run_all_tests_torch_gpu,
+        run_all_tests_tf_gpu,
+        run_all_tests_torch_multi_gpu,
+        run_all_tests_tf_multi_gpu,
+        run_all_tests_torch_cuda_extensions_gpu,
+        run_all_tests_torch_cuda_extensions_multi_gpu
     ]
     steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-          echo "Runner status: ${{ needs.check_runners.result }}"
-          echo "Setup status: ${{ needs.setup.result }}"
-
       - uses: actions/checkout@v2
+
       - uses: actions/download-artifact@v2
+
       - name: Send message to Slack
         env:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
           CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_EVENT: scheduled
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
+
+
         run: |
           pip install slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
+          python utils/notification_service.py scheduled

.github/workflows/stale.yml

@@ -12,10 +12,10 @@ jobs:
     env:
       GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v2
 
     - name: Setup Python
-      uses: actions/setup-python@v4
+      uses: actions/setup-python@v1
       with:
         python-version: 3.7
 
@@ -24,4 +24,4 @@ jobs:
         pip install PyGithub
     - name: Close stale issues
       run: |
-        python scripts/stale.py
+        python scripts/stale.py

.github/workflows/update_metdata.yml

@@ -1,40 +0,0 @@
-name: Update Transformers metadata
-
-on:
-  push:
-    branches:
-      - main
-      - update_transformers_metadata
-
-jobs:
-  build_and_package:
-    runs-on: ubuntu-latest
-    defaults:
-      run:
-        shell: bash -l {0}
-
-    steps:
-      - uses: actions/checkout@v2
-
-      - name: Load cached virtual environment
-        uses: actions/cache@v2
-        id: cache
-        with:
-          path: ~/venv/
-          key: v3-metadata-${{ hashFiles('setup.py') }}
-
-      - name: Create virtual environment on cache miss
-        if: steps.cache.outputs.cache-hit != 'true'
-        run: |
-          python -m venv ~/venv && . ~/venv/bin/activate
-          pip install --upgrade pip
-
-      - name: Setup environment
-        run: |
-          . ~/venv/bin/activate
-          pip install git+https://github.com/huggingface/transformers#egg=transformers[dev]
-
-      - name: Update metadata
-        run: |
-          . ~/venv/bin/activate
-          python utils/update_metadata.py --token ${{ secrets.SYLVAIN_HF_TOKEN }} --commit_sha ${{ github.sha }}

.gitignore

@@ -160,7 +160,4 @@ tags
 .pre-commit*
 
 # .lock
-*.lock
-
-# DS_Store (MacOS)
-.DS_Store
+*.lock

CITATION.cff

@@ -1,82 +0,0 @@
-cff-version: "1.2.0"
-date-released: 2020-10
-message: "If you use this software, please cite it using these metadata."
-title: "Transformers: State-of-the-Art Natural Language Processing"
-url: "https://github.com/huggingface/transformers"
-authors: 
-  - family-names: Wolf
-    given-names: Thomas
-  - family-names: Debut
-    given-names: Lysandre
-  - family-names: Sanh
-    given-names: Victor
-  - family-names: Chaumond
-    given-names: Julien
-  - family-names: Delangue
-    given-names: Clement
-  - family-names: Moi
-    given-names: Anthony
-  - family-names: Cistac
-    given-names: Perric
-  - family-names: Ma
-    given-names: Clara
-  - family-names: Jernite
-    given-names: Yacine
-  - family-names: Plu
-    given-names: Julien
-  - family-names: Xu
-    given-names: Canwen
-  - family-names: "Le Scao"
-    given-names: Teven
-  - family-names: Gugger
-    given-names: Sylvain
-  - family-names: Drame
-    given-names: Mariama
-  - family-names: Lhoest
-    given-names: Quentin
-  - family-names: Rush
-    given-names: "Alexander M."
-preferred-citation:
-  type: conference-paper
-  authors:
-  - family-names: Wolf
-    given-names: Thomas
-  - family-names: Debut
-    given-names: Lysandre
-  - family-names: Sanh
-    given-names: Victor
-  - family-names: Chaumond
-    given-names: Julien
-  - family-names: Delangue
-    given-names: Clement
-  - family-names: Moi
-    given-names: Anthony
-  - family-names: Cistac
-    given-names: Perric
-  - family-names: Ma
-    given-names: Clara
-  - family-names: Jernite
-    given-names: Yacine
-  - family-names: Plu
-    given-names: Julien
-  - family-names: Xu
-    given-names: Canwen
-  - family-names: "Le Scao"
-    given-names: Teven
-  - family-names: Gugger
-    given-names: Sylvain
-  - family-names: Drame
-    given-names: Mariama
-  - family-names: Lhoest
-    given-names: Quentin
-  - family-names: Rush
-    given-names: "Alexander M."
-  booktitle: "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations"
-  month: 10
-  start: 38
-  end: 45
-  title: "Transformers: State-of-the-Art Natural Language Processing"
-  year: 2020
-  publisher: "Association for Computational Linguistics"
-  url: "https://www.aclweb.org/anthology/2020.emnlp-demos.6"
-  address: "Online"

CONTRIBUTING.md

@@ -26,7 +26,7 @@ on the awesome projects it made possible, shout out on Twitter every time it has
 helped you, or simply star the repo to say "thank you".
 
 Whichever way you choose to contribute, please be mindful to respect our
-[code of conduct](https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md).
+[code of conduct](https://github.com/huggingface/transformers/blob/master/CODE_OF_CONDUCT.md).
 
 ## You can contribute in so many ways!
 
@@ -92,7 +92,7 @@ If you are willing to contribute the model yourself, let us know so we can best
 guide you.
 
 We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them
-in the [`templates`](https://github.com/huggingface/transformers/tree/main/templates) folder.
+in the [`templates`](https://github.com/huggingface/transformers/tree/master/templates) folder.
 
 ### Do you want a new feature (that is not a model)?
 
@@ -114,7 +114,7 @@ If your issue is well written we're already 80% of the way there by the time you
 post it.
 
 We have added **templates** to guide you in the process of adding a new example script for training or testing the
-models in the library. You can find them in the [`templates`](https://github.com/huggingface/transformers/tree/main/templates)
+models in the library. You can find them in the [`templates`](https://github.com/huggingface/transformers/tree/master/templates)
 folder.
 
 ## Start contributing! (Pull Requests)
@@ -124,11 +124,11 @@ issues to make sure that nobody is already working on the same thing. If you are
 unsure, it is always a good idea to open an issue to get some feedback.
 
 You will need basic `git` proficiency to be able to contribute to
-🤗 Transformers. `git` is not the easiest tool to use but it has the greatest
+`transformers`. `git` is not the easiest tool to use but it has the greatest
 manual. Type `git --help` in a shell and enjoy. If you prefer books, [Pro
 Git](https://git-scm.com/book/en/v2) is a very good reference.
 
-Follow these steps to start contributing ([supported Python versions](https://github.com/huggingface/transformers/blob/main/setup.py#L426)):
+Follow these steps to start contributing:
 
 1. Fork the [repository](https://github.com/huggingface/transformers) by
    clicking on the 'Fork' button on the repository's page. This creates a copy of the code
@@ -148,7 +148,7 @@ Follow these steps to start contributing ([supported Python versions](https://gi
    $ git checkout -b a-descriptive-name-for-my-changes
    ```
 
-   **Do not** work on the `main` branch.
+   **Do not** work on the `master` branch.
 
 4. Set up a development environment by running the following command in a virtual environment:
 
@@ -171,38 +171,38 @@ Follow these steps to start contributing ([supported Python versions](https://gi
 
    If you have already cloned that repo, you might need to `git pull` to get the most recent changes in the `datasets`
    library.
-   
-   Depending on your OS, you might need to install some external libraries, as well, if the `pip` installation fails.
-   
-   For macOS, you will likely need [MeCab](https://taku910.github.io/mecab/), which can be installed from Homebrew:
-   
-   ```bash
-   brew install mecab
-   ```
 
 5. Develop the features on your branch.
 
    As you work on the features, you should make sure that the test suite
-   passes. You should run the tests impacted by your changes like this:
-
-   ```bash
-   $ pytest tests/<TEST_TO_RUN>.py
-   ```
-
-   You can also run the full suite with the following command, but it takes
-   a beefy machine to produce a result in a decent amount of time now that
-   Transformers has grown a lot. Here is the command for it:
+   passes:
 
    ```bash
    $ make test
    ```
 
-   For more information about tests, check out the
-   [dedicated documentation](https://huggingface.co/docs/transformers/testing)
+   Note, that this command uses `-n auto` pytest flag, therefore, it will start as many parallel `pytest` processes as the number of your computer's CPU-cores, and if you have lots of those and a few GPUs and not a great amount of RAM, it's likely to overload your computer. Therefore, to run the test suite, you may want to consider using this command instead:
 
-   🤗 Transformers relies on `black` and `isort` to format its source code
-   consistently. After you make changes, apply automatic style corrections and code verifications
-   that can't be automated in one go with:
+   ```bash
+   $ python -m pytest -n 3 --dist=loadfile -s -v ./tests/
+   ```
+
+   Adjust the value of `-n` to fit the load your hardware can support.
+
+   `transformers` relies on `black` and `isort` to format its source code
+   consistently. After you make changes, format them with:
+
+   ```bash
+   $ make style
+   ```
+
+   `transformers` also uses `flake8` and a few custom scripts to check for coding mistakes. Quality
+   control runs in CI, however you can also run the same checks with:
+
+   ```bash
+   $ make quality
+   ```
+   You can do the automatic style corrections and code verifications that can't be automated in one go:
 
    ```bash
    $ make fixup
@@ -210,55 +210,16 @@ Follow these steps to start contributing ([supported Python versions](https://gi
 
    This target is also optimized to only work with files modified by the PR you're working on.
 
-   If you prefer to run the checks one after the other, the following command apply the
-   style corrections:
-
-   ```bash
-   $ make style
-   ```
-
-   🤗 Transformers also uses `flake8` and a few custom scripts to check for coding mistakes. Quality
-   control runs in CI, however you can also run the same checks with:
-
-   ```bash
-   $ make quality
-   ```
-
-   Finally we have a lot of scripts that check we didn't forget to update
-   some files when adding a new model, that you can run with
-
-   ```bash
-   $ make repo-consistency
-   ```
-
-   To learn more about those checks and how to fix any issue with them, check out the
-   [documentation](https://huggingface.co/docs/transformers/pr_checks)
-
    If you're modifying documents under `docs/source`, make sure to validate that
    they can still be built. This check also runs in CI. To run a local check
-   make sure you have installed the documentation builder requirements. First you will need to clone the
-   repository containing our tools to build the documentation:
-   
-   ```bash
-   $ pip install git+https://github.com/huggingface/doc-builder
-   ```
-
-   Then, make sure you have all the dependencies to be able to build the doc with:
-   
-   ```bash
-   $ pip install ".[docs]"
-   ```
-
-   Finally run the following command from the root of the repository:
+   make sure you have installed the documentation builder requirements, by
+   running `pip install .[tf,torch,docs]` once from the root of this repository
+   and then run:
 
    ```bash
-   $ doc-builder build transformers docs/source/ --build_dir ~/tmp/test-build
+   $ make docs
    ```
 
-   This will build the documentation in the `~/tmp/test-build` folder where you can inspect the generated
-   Markdown files with your favorite editor. You won't be able to see the final rendering on the website
-   before your PR is merged, we are actively working on adding a tool for this.
-
    Once you're happy with your changes, add changed files using `git add` and
    make a commit with `git commit` to record your changes locally:
 
@@ -275,7 +236,7 @@ Follow these steps to start contributing ([supported Python versions](https://gi
 
    ```bash
    $ git fetch upstream
-   $ git rebase upstream/main
+   $ git rebase upstream/master
    ```
 
    Push the changes to your account using:
@@ -312,21 +273,14 @@ Follow these steps to start contributing ([supported Python versions](https://gi
    - If you are adding a new tokenizer, write tests, and make sure
      `RUN_SLOW=1 python -m pytest tests/test_tokenization_{your_model_name}.py` passes.
    CircleCI does not run the slow tests, but github actions does every night!
-6. All public methods must have informative docstrings that work nicely with sphinx. See `modeling_bert.py` for an
+6. All public methods must have informative docstrings that work nicely with sphinx. See `modeling_ctrl.py` for an
    example.
-7. Due to the rapidly growing repository, it is important to make sure that no files that would significantly weigh down the repository are added. This includes images, videos and other non-text files. We prefer to leverage a hf.co hosted `dataset` like
-   the ones hosted on [`hf-internal-testing`](https://huggingface.co/hf-internal-testing) in which to place these files and reference 
-   them by URL. We recommend putting them in the following dataset: [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images).
-   If an external contribution, feel free to add the images to your PR and ask a Hugging Face member to migrate your images
-   to this dataset.
-
-See more about the checks run on a pull request in our [PR guide](pr_checks)
 
 ### Tests
 
 An extensive test suite is included to test the library behavior and several examples. Library tests can be found in
-the [tests folder](https://github.com/huggingface/transformers/tree/main/tests) and examples tests in the
-[examples folder](https://github.com/huggingface/transformers/tree/main/examples).
+the [tests folder](https://github.com/huggingface/transformers/tree/master/tests) and examples tests in the
+[examples folder](https://github.com/huggingface/transformers/tree/master/examples).
 
 We like `pytest` and `pytest-xdist` because it's faster. From the root of the
 repository, here's how to run tests with `pytest` for the library:
@@ -372,11 +326,12 @@ $ python -m unittest discover -s examples -t examples -v
 
 ### Style guide
 
-For documentation strings, 🤗 Transformers follows the [google style](https://google.github.io/styleguide/pyguide.html).
-Check our [documentation writing guide](https://github.com/huggingface/transformers/tree/main/docs#writing-documentation---specification)
+For documentation strings, `transformers` follows the [google style](https://google.github.io/styleguide/pyguide.html).
+Check our [documentation writing guide](https://github.com/huggingface/transformers/tree/master/docs#writing-documentation---specification)
 for more information.
 
-**This guide was heavily inspired by the awesome [scikit-learn guide to contributing](https://github.com/scikit-learn/scikit-learn/blob/main/CONTRIBUTING.md).**
+#### This guide was heavily inspired by the awesome [scikit-learn guide to contributing](https://github.com/scikit-learn/scikit-learn/blob/master/CONTRIBUTING.md)
+
 
 ### Develop on Windows
 
@@ -393,15 +348,15 @@ One way one can run the make command on Window is to pass by MSYS2:
 
 You can now use `make` from any terminal (Powershell, cmd.exe, etc) 🎉
 
-### Syncing forked main with upstream (HuggingFace) main
+### Syncing forked master with upstream (HuggingFace) master
 
-To avoid pinging the upstream repository which adds reference notes to each upstream PR and sends unnecessary notifications to the developers involved in these PRs,
-when syncing the main branch of a forked repository, please, follow these steps:
-1. When possible, avoid syncing with the upstream using a branch and PR on the forked repository. Instead merge directly into the forked main.
+To avoid pinging the upstream repository which adds reference notes to each upstream PR and sends unnessary notifications to the developers involved in these PRs,
+when syncing the master branch of a forked repository, please, follow these steps:
+1. When possible, avoid syncing with the upstream using a branch and PR on the forked repository. Instead merge directly into the forked master.
 2. If a PR is absolutely necessary, use the following steps after checking out your branch:
 ```
 $ git checkout -b your-branch-for-syncing
-$ git pull --squash --no-commit upstream main
+$ git pull --squash --no-commit upstream master
 $ git commit -m '<your message without GitHub references>'
 $ git push --set-upstream origin your-branch-for-syncing
 ```

ISSUES.md

@@ -71,8 +71,8 @@ You are not required to read the following guidelines before opening an issue. H
      File "/transformers/src/transformers/__init__.py", line 34, in <module>
        from . import dependency_versions_check
      File "/transformers/src/transformers/dependency_versions_check.py", line 34, in <module>
-       from .utils import is_tokenizers_available
-     File "/transformers/src/transformers/utils/import_utils.py", line 40, in <module>
+       from .file_utils import is_tokenizers_available
+     File "/transformers/src/transformers/file_utils.py", line 40, in <module>
        from tqdm.auto import tqdm
     ModuleNotFoundError: No module named 'tqdm.auto'
     ```
@@ -124,8 +124,8 @@ You are not required to read the following guidelines before opening an issue. H
      File "/transformers/src/transformers/__init__.py", line 34, in <module>
        from . import dependency_versions_check
      File "/transformers/src/transformers/dependency_versions_check.py", line 34, in <module>
-       from .utils import is_tokenizers_available
-     File "/transformers/src/transformers/utils/import_utils.py", line 40, in <module>
+       from .file_utils import is_tokenizers_available
+     File "/transformers/src/transformers/file_utils.py", line 40, in <module>
        from tqdm.auto import tqdm
    ModuleNotFoundError: No module named 'tqdm.auto'
    ```
@@ -205,7 +205,7 @@ You are not required to read the following guidelines before opening an issue. H
 
    If you really tried to make a short reproducible code but couldn't figure it out, it might be that having a traceback will give the developer enough information to know what's going on. But if it is not enough and we can't reproduce the problem, we can't really solve it.
 
-   Do not despair if you can't figure it out from the beginning, just share what you can and perhaps someone else will be able to help you at the forums.
+   Do not dispair if you can't figure it out from the begining, just share what you can and perhaps someone else will be able to help you at the forums.
 
    If your setup involves any custom datasets, the best way to help us reproduce the problem is to create a [Google Colab notebook](https://colab.research.google.com/) that demonstrates the issue and once you verify that the issue still exists, include a link to that notebook in the Issue. Just make sure that you don't copy and paste the location bar url of the open notebook - as this is private and we won't be able to open it. Instead, you need to click on `Share` in the right upper corner of the notebook, select `Get Link` and then copy and paste the public link it will give to you.
 

Makefile

@@ -1,4 +1,4 @@
-.PHONY: deps_table_update modified_only_fixup extra_style_checks quality style fixup fix-copies test test-examples
+.PHONY: deps_table_update modified_only_fixup extra_quality_checks quality style fixup fix-copies test test-examples docs
 
 # make sure to test the local checkout in scripts and not the pre-installed one (don't use quotes!)
 export PYTHONPATH = src
@@ -9,7 +9,7 @@ modified_only_fixup:
 	$(eval modified_py_files := $(shell python utils/get_modified_files.py $(check_dirs)))
 	@if test -n "$(modified_py_files)"; then \
 		echo "Checking/fixing $(modified_py_files)"; \
-		black --preview $(modified_py_files); \
+		black $(modified_py_files); \
 		isort $(modified_py_files); \
 		flake8 $(modified_py_files); \
 	else \
@@ -21,58 +21,44 @@ modified_only_fixup:
 deps_table_update:
 	@python setup.py deps_table_update
 
-deps_table_check_updated:
-	@md5sum src/transformers/dependency_versions_table.py > md5sum.saved
-	@python setup.py deps_table_update
-	@md5sum -c --quiet md5sum.saved || (printf "\nError: the version dependency table is outdated.\nPlease run 'make fixup' or 'make style' and commit the changes.\n\n" && exit 1)
-	@rm md5sum.saved
-
 # autogenerating code
 
 autogenerate_code: deps_table_update
+	python utils/class_mapping_update.py
 
-# Check that the repo is in a good state
+# Check that source code meets quality standards
 
-repo-consistency:
+extra_quality_checks:
 	python utils/check_copies.py
 	python utils/check_table.py
 	python utils/check_dummies.py
 	python utils/check_repo.py
 	python utils/check_inits.py
-	python utils/check_config_docstrings.py
-	python utils/tests_fetcher.py --sanity_check
-	python utils/update_metadata.py --check-only
 
 # this target runs checks on all files
-
 quality:
-	black --check --preview $(check_dirs)
+	black --check $(check_dirs)
 	isort --check-only $(check_dirs)
 	python utils/custom_init_isort.py --check_only
-	python utils/sort_auto_mappings.py --check_only
 	flake8 $(check_dirs)
-	doc-builder style src/transformers docs/source --max_len 119 --check_only --path_to_docs docs/source
-	python utils/check_doc_toc.py
+	${MAKE} extra_quality_checks
 
 # Format source code automatically and check is there are any problems left that need manual fixing
 
 extra_style_checks:
 	python utils/custom_init_isort.py
-	python utils/sort_auto_mappings.py
-	doc-builder style src/transformers docs/source --max_len 119 --path_to_docs docs/source
-	python utils/check_doc_toc.py --fix_and_overwrite
+	python utils/style_doc.py src/transformers docs/source --max_len 119
 
 # this target runs checks on all files and potentially modifies some of them
-
 style:
-	black --preview $(check_dirs)
+	black $(check_dirs)
 	isort $(check_dirs)
 	${MAKE} autogenerate_code
 	${MAKE} extra_style_checks
 
 # Super fast fix and check target that only works on relevant modified files since the branch was made
 
-fixup: modified_only_fixup extra_style_checks autogenerate_code repo-consistency
+fixup: modified_only_fixup extra_style_checks autogenerate_code extra_quality_checks
 
 # Make marked copies of snippets of codes conform to the original
 
@@ -97,6 +83,11 @@ test-sagemaker: # install sagemaker dependencies in advance with pip install .[s
 	TEST_SAGEMAKER=True python -m pytest -n auto  -s -v ./tests/sagemaker
 
 
+# Check that docs can build
+
+docs:
+	cd docs && make html SPHINXOPTS="-W -j 4"
+
 # Release stuff
 
 pre-release:

README.md

@@ -16,55 +16,37 @@ limitations under the License.
 
 <p align="center">
     <br>
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
+    <img src="https://raw.githubusercontent.com/huggingface/transformers/master/docs/source/imgs/transformers_logo_name.png" width="400"/>
     <br>
 <p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/master">
     </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+    <a href="https://github.com/huggingface/transformers/blob/master/LICENSE">
         <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
     </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    <a href="https://huggingface.co/transformers/index.html">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/transformers/index.html.svg?down_color=red&down_message=offline&up_message=online">
     </a>
     <a href="https://github.com/huggingface/transformers/releases">
         <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
     </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+    <a href="https://github.com/huggingface/transformers/blob/master/CODE_OF_CONDUCT.md">
         <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
     </a>
     <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
 </p>
 
-<h4 align="center">
-    <p>
-        <b>English</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> 
-    <p>
-</h4>
-
 <h3 align="center">
-    <p>State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow</p>
+    <p>State-of-the-art Natural Language Processing for Jax, PyTorch and TensorFlow</p>
 </h3>
 
 <h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
+    <a href="https://hf.co/course"><img src="https://raw.githubusercontent.com/huggingface/transformers/master/docs/source/imgs/course_banner.png"></a>
 </h3>
 
-🤗 Transformers provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio.
-
-These models can be applied on:
-
-* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, text generation, in over 100 languages.
-* 🖼️ Images, for tasks like image classification, object detection, and segmentation.
-* 🗣️ Audio, for tasks like speech recognition and audio classification.
-
-Transformer models can also perform tasks on **several modalities combined**, such as table question answering, optical character recognition, information extraction from scanned documents, video classification, and visual question answering.
+🤗 Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, question answering, summarization, translation, text generation and more in over 100 languages. Its aim is to make cutting-edge NLP easier to use for everyone.
 
 🤗 Transformers provides APIs to quickly download and use those pretrained models on a given text, fine-tune them on your own datasets and then share them with the community on our [model hub](https://huggingface.co/models). At the same time, each python module defining an architecture is fully standalone and can be modified to enable quick research experiments.
 
@@ -75,8 +57,6 @@ Transformer models can also perform tasks on **several modalities combined**, su
 You can test most of our models directly on their pages from the [model hub](https://huggingface.co/models). We also offer [private model hosting, versioning, & an inference API](https://huggingface.co/pricing) for public and private models.
 
 Here are a few examples:
-
- In Natural Language Processing:
 - [Masked word completion with BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
 - [Name Entity Recognition with Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
 - [Text generation with GPT-2](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
@@ -85,30 +65,17 @@ Here are a few examples:
 - [Question answering with DistilBERT](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
 - [Translation with T5](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
 
-In Computer Vision:
-- [Image classification with ViT](https://huggingface.co/google/vit-base-patch16-224)
-- [Object Detection with DETR](https://huggingface.co/facebook/detr-resnet-50)
-- [Semantic Segmentation with SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
-- [Panoptic Segmentation with DETR](https://huggingface.co/facebook/detr-resnet-50-panoptic)
-
-In Audio:
-- [Automatic Speech Recognition with Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base-960h)
-- [Keyword Spotting with Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
-
-In Multimodal tasks:
-- [Visual Question Answering with ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
-
 **[Write With Transformer](https://transformer.huggingface.co)**, built by the Hugging Face team, is the official demo of this repo’s text generation capabilities.
 
 ## If you are looking for custom support from the Hugging Face team
 
 <a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
 </a><br>
 
 ## Quick tour
 
-To immediately use a model on a given input (text, image, audio, ...), we provide the `pipeline` API. Pipelines group together a pretrained model with the preprocessing that was used during that model's training. Here is how to quickly use a pipeline to classify positive versus negative texts:
+To immediately use a model on a given text, we provide the `pipeline` API. Pipelines group together a pretrained model with the preprocessing that was used during that model's training. Here is how to quickly use a pipeline to classify positive versus negative texts:
 
 ```python
 >>> from transformers import pipeline
@@ -121,48 +88,24 @@ To immediately use a model on a given input (text, image, audio, ...), we provid
 
 The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here the answer is "positive" with a confidence of 99.97%.
 
-Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in computer vision and speech. For example, we can easily extract detected objects in an image:
+Many NLP tasks have a pre-trained `pipeline` ready to go. For example, we can easily extract question answers given context:
 
 ``` python
->>> import requests
->>> from PIL import Image
 >>> from transformers import pipeline
 
-# Download an image with cute cats
->>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
->>> image_data = requests.get(url, stream=True).raw
->>> image = Image.open(image_data)
+# Allocate a pipeline for question-answering
+>>> question_answerer = pipeline('question-answering')
+>>> question_answerer({
+...     'question': 'What is the name of the repository ?',
+...     'context': 'Pipeline has been included in the huggingface/transformers repository'
+... })
+{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
 
-# Allocate a pipeline for object detection
->>> object_detector = pipeline('object-detection')
->>> object_detector(image)
-[{'score': 0.9982201457023621,
-  'label': 'remote',
-  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
- {'score': 0.9960021376609802,
-  'label': 'remote',
-  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
- {'score': 0.9954745173454285,
-  'label': 'couch',
-  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
- {'score': 0.9988006353378296,
-  'label': 'cat',
-  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
- {'score': 0.9986783862113953,
-  'label': 'cat',
-  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
 ```
 
-Here we get a list of objects detected in the image, with a box surrounding the object and a confidence score. Here is the original image on the right, with the predictions displayed on the left:
+In addition to the answer, the pretrained model used here returned its confidence score, along with the start position and end position of the answer in the tokenized sentence. You can learn more about the tasks supported by the `pipeline` API in [this tutorial](https://huggingface.co/transformers/task_summary.html).
 
-<h3 align="center">
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
-</h3>
-
-You can learn more about the tasks supported by the `pipeline` API in [this tutorial](https://huggingface.co/docs/transformers/task_summary).
-
-In addition to `pipeline`, to download and use any of the pretrained models on your given task, all it takes is three lines of code. Here is the PyTorch version:
+To download and use any of the pretrained models on your given task, all it takes is three lines of code. Here is the PyTorch version:
 ```python
 >>> from transformers import AutoTokenizer, AutoModel
 
@@ -172,7 +115,6 @@ In addition to `pipeline`, to download and use any of the pretrained models on y
 >>> inputs = tokenizer("Hello world!", return_tensors="pt")
 >>> outputs = model(**inputs)
 ```
-
 And here is the equivalent code for TensorFlow:
 ```python
 >>> from transformers import AutoTokenizer, TFAutoModel
@@ -186,12 +128,12 @@ And here is the equivalent code for TensorFlow:
 
 The tokenizer is responsible for all the preprocessing the pretrained model expects, and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
 
-The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use as usual. [This tutorial](https://huggingface.co/docs/transformers/training) explains how to integrate such a model into a classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune on a new dataset.
+The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use normally. [This tutorial](https://huggingface.co/transformers/training.html) explains how to integrate such a model into a classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune on a new dataset.
 
 ## Why should I use transformers?
 
 1. Easy-to-use state-of-the-art models:
-    - High performance on natural language understanding & generation, computer vision, and audio tasks.
+    - High performance on NLU and NLG tasks.
     - Low barrier to entry for educators and practitioners.
     - Few user-facing abstractions with just three classes to learn.
     - A unified API for using all our pretrained models.
@@ -199,11 +141,11 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 1. Lower compute costs, smaller carbon footprint:
     - Researchers can share trained models instead of always retraining.
     - Practitioners can reduce compute time and production costs.
-    - Dozens of architectures with over 60,000 pretrained models across all modalities.
+    - Dozens of architectures with over 2,000 pretrained models, some in more than 100 languages.
 
 1. Choose the right framework for every part of a model's lifetime:
     - Train state-of-the-art models in 3 lines of code.
-    - Move a single model between TF2.0/PyTorch/JAX frameworks at will.
+    - Move a single model between TF2.0/PyTorch frameworks at will.
     - Seamlessly pick the right framework for training, evaluation and production.
 
 1. Easily customize a model or an example to your needs:
@@ -214,8 +156,8 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 ## Why shouldn't I use transformers?
 
 - This library is not a modular toolbox of building blocks for neural nets. The code in the model files is not refactored with additional abstractions on purpose, so that researchers can quickly iterate on each of the models without diving into additional abstractions/files.
-- The training API is not intended to work on any model but is optimized to work with the models provided by the library. For generic machine learning loops, you should use another library (possibly, [Accelerate](https://huggingface.co/docs/accelerate)).
-- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/main/examples) are just that: examples. It is expected that they won't work out-of-the box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
+- The training API is not intended to work on any model but is optimized to work with the models provided by the library. For generic machine learning loops, you should use another library.
+- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/master/examples) are just that: examples. It is expected that they won't work out-of-the box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
 
 ## Installation
 
@@ -228,7 +170,7 @@ You should install 🤗 Transformers in a [virtual environment](https://docs.pyt
 First, create a virtual environment with the version of Python you're going to use and activate it.
 
 Then, you will need to install at least one of Flax, PyTorch or TensorFlow.
-Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax](https://github.com/google/flax#quick-install) and [Jax](https://github.com/google/jax#installation) installation pages regarding the specific installation command for your platform.
+Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax installation page](https://github.com/google/flax#quick-install) regarding the specific install command for your platform.
 
 When one of those backends has been installed, 🤗 Transformers can be installed using pip as follows:
 
@@ -236,7 +178,7 @@ When one of those backends has been installed, 🤗 Transformers can be installe
 pip install transformers
 ```
 
-If you'd like to play with the examples or need the bleeding edge of the code and can't wait for a new release, you must [install the library from source](https://huggingface.co/docs/transformers/installation#installing-from-source).
+If you'd like to play with the examples or need the bleeding edge of the code and can't wait for a new release, you must [install the library from source](https://huggingface.co/transformers/installation.html#installing-from-source).
 
 ### With conda
 
@@ -250,181 +192,96 @@ conda install -c huggingface transformers
 
 Follow the installation pages of Flax, PyTorch or TensorFlow to see how to install them with conda.
 
-> **_NOTE:_**  On Windows, you may be prompted to activate Developer Mode in order to benefit from caching. If this is not an option for you, please let us know in [this issue](https://github.com/huggingface/huggingface_hub/issues/1062).
-
 ## Model architectures
 
 **[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co) where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
 
 Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
 
-🤗 Transformers currently provides the following architectures (see [here](https://huggingface.co/docs/transformers/model_summary) for a high-level summary of each them):
+🤗 Transformers currently provides the following architectures (see [here](https://huggingface.co/transformers/model_summary.html) for a high-level summary of each them):
 
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2** was released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](https://huggingface.co/docs/transformers/main/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)**  (from HuggingFace).
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+1. **[ALBERT](https://huggingface.co/transformers/model_doc/albert.html)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
+1. **[BART](https://huggingface.co/transformers/model_doc/bart.html)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
+1. **[BARThez](https://huggingface.co/transformers/model_doc/barthez.html)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
+1. **[BERT](https://huggingface.co/transformers/model_doc/bert.html)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+1. **[BERT For Sequence Generation](https://huggingface.co/transformers/model_doc/bertgeneration.html)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BigBird-RoBERTa](https://huggingface.co/transformers/model_doc/bigbird.html)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-Pegasus](https://huggingface.co/transformers/model_doc/bigbird_pegasus.html)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[Blenderbot](https://huggingface.co/transformers/model_doc/blenderbot.html)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BlenderbotSmall](https://huggingface.co/transformers/model_doc/blenderbot_small.html)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BORT](https://huggingface.co/transformers/model_doc/bort.html)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[ByT5](https://huggingface.co/transformers/model_doc/byt5.html)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
+1. **[CamemBERT](https://huggingface.co/transformers/model_doc/camembert.html)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+1. **[CLIP](https://huggingface.co/transformers/model_doc/clip.html)** from (OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
+1. **[ConvBERT](https://huggingface.co/transformers/model_doc/convbert.html)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[CPM](https://huggingface.co/transformers/model_doc/cpm.html)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+1. **[CTRL](https://huggingface.co/transformers/model_doc/ctrl.html)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+1. **[DeBERTa](https://huggingface.co/transformers/model_doc/deberta.html)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa-v2](https://huggingface.co/transformers/model_doc/deberta_v2.html)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeiT](https://huggingface.co/transformers/model_doc/deit.html)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DETR](https://huggingface.co/transformers/model_doc/detr.html)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
+1. **[DialoGPT](https://huggingface.co/transformers/model_doc/dialogpt.html)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
+1. **[DistilBERT](https://huggingface.co/transformers/model_doc/distilbert.html)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/master/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) and a German version of DistilBERT.
+1. **[DPR](https://huggingface.co/transformers/model_doc/dpr.html)** (from Facebook) released with the paper [Dense Passage Retrieval
+for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon
+Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+1. **[ELECTRA](https://huggingface.co/transformers/model_doc/electra.html)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+1. **[FlauBERT](https://huggingface.co/transformers/model_doc/flaubert.html)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
+1. **[Funnel Transformer](https://huggingface.co/transformers/model_doc/funnel.html)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
+1. **[GPT](https://huggingface.co/transformers/model_doc/gpt.html)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT-2](https://huggingface.co/transformers/model_doc/gpt2.html)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT Neo](https://huggingface.co/transformers/model_doc/gpt_neo.html)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[Hubert](https://huggingface.co/transformers/model_doc/hubert.html)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
+1. **[I-BERT](https://huggingface.co/transformers/model_doc/ibert.html)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer
+1. **[LayoutLM](https://huggingface.co/transformers/model_doc/layoutlm.html)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LED](https://huggingface.co/transformers/model_doc/led.html)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[Longformer](https://huggingface.co/transformers/model_doc/longformer.html)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LUKE](https://huggingface.co/transformers/model_doc/luke.html)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+1. **[LXMERT](https://huggingface.co/transformers/model_doc/lxmert.html)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[M2M100](https://huggingface.co/transformers/model_doc/m2m_100.html)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
+1. **[MarianMT](https://huggingface.co/transformers/model_doc/marian.html)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
+1. **[MBart](https://huggingface.co/transformers/model_doc/mbart.html)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
+1. **[MBart-50](https://huggingface.co/transformers/model_doc/mbart.html)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
+1. **[Megatron-BERT](https://huggingface.co/transformers/model_doc/megatron_bert.html)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[Megatron-GPT2](https://huggingface.co/transformers/model_doc/megatron_gpt2.html)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[MPNet](https://huggingface.co/transformers/model_doc/mpnet.html)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
+1. **[MT5](https://huggingface.co/transformers/model_doc/mt5.html)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
+1. **[Pegasus](https://huggingface.co/transformers/model_doc/pegasus.html)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777)> by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[ProphetNet](https://huggingface.co/transformers/model_doc/prophetnet.html)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[Reformer](https://huggingface.co/transformers/model_doc/reformer.html)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
+1. **[RoBERTa](https://huggingface.co/transformers/model_doc/roberta.html)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoFormer](https://huggingface.co/transformers/model_doc/roformer.html)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+1. **[SpeechToTextTransformer](https://huggingface.co/transformers/model_doc/speech_to_text.html)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
+1. **[SqueezeBert](https://huggingface.co/transformers/model_doc/squeezebert.html)** released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
+1. **[T5](https://huggingface.co/transformers/model_doc/t5.html)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[TAPAS](https://huggingface.co/transformers/model_doc/tapas.html)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+1. **[Transformer-XL](https://huggingface.co/transformers/model_doc/transformerxl.html)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
+1. **[Vision Transformer (ViT)](https://huggingface.co/transformers/model_doc/vit.html)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+1. **[VisualBERT](https://huggingface.co/transformers/model_doc/visual_bert.html)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+1. **[Wav2Vec2](https://huggingface.co/transformers/model_doc/wav2vec2.html)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
+1. **[XLM](https://huggingface.co/transformers/model_doc/xlm.html)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLM-ProphetNet](https://huggingface.co/transformers/model_doc/xlmprophetnet.html)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[XLM-RoBERTa](https://huggingface.co/transformers/model_doc/xlmroberta.html)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+1. **[XLNet](https://huggingface.co/transformers/model_doc/xlnet.html)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+1. **[XLSR-Wav2Vec2](https://huggingface.co/transformers/model_doc/xlsr_wav2vec2.html)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
 1. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
-To check if each model has an implementation in Flax, PyTorch or TensorFlow, or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/docs/transformers/index#supported-frameworks).
+To check if each model has an implementation in Flax, PyTorch or TensorFlow, or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/transformers/index.html#supported-frameworks).
 
-These implementations have been tested on several datasets (see the example scripts) and should match the performance of the original implementations. You can find more details on performance in the Examples section of the [documentation](https://github.com/huggingface/transformers/tree/main/examples).
+These implementations have been tested on several datasets (see the example scripts) and should match the performance of the original implementations. You can find more details on performance in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).
 
 
 ## Learn more
 
 | Section | Description |
 |-|-|
-| [Documentation](https://huggingface.co/docs/transformers/) | Full API documentation and tutorials |
-| [Task summary](https://huggingface.co/docs/transformers/task_summary) | Tasks supported by 🤗 Transformers |
-| [Preprocessing tutorial](https://huggingface.co/docs/transformers/preprocessing) | Using the `Tokenizer` class to prepare data for the models |
-| [Training and fine-tuning](https://huggingface.co/docs/transformers/training) | Using the models provided by 🤗 Transformers in a PyTorch/TensorFlow training loop and the `Trainer` API |
-| [Quick tour: Fine-tuning/usage scripts](https://github.com/huggingface/transformers/tree/main/examples) | Example scripts for fine-tuning models on a wide range of tasks |
-| [Model sharing and uploading](https://huggingface.co/docs/transformers/model_sharing) | Upload and share your fine-tuned models with the community |
-| [Migration](https://huggingface.co/docs/transformers/migration) | Migrate to 🤗 Transformers from `pytorch-transformers` or `pytorch-pretrained-bert` |
+| [Documentation](https://huggingface.co/transformers/) | Full API documentation and tutorials |
+| [Task summary](https://huggingface.co/transformers/task_summary.html) | Tasks supported by 🤗 Transformers |
+| [Preprocessing tutorial](https://huggingface.co/transformers/preprocessing.html) | Using the `Tokenizer` class to prepare data for the models |
+| [Training and fine-tuning](https://huggingface.co/transformers/training.html) | Using the models provided by 🤗 Transformers in a PyTorch/TensorFlow training loop and the `Trainer` API |
+| [Quick tour: Fine-tuning/usage scripts](https://github.com/huggingface/transformers/tree/master/examples) | Example scripts for fine-tuning models on a wide range of tasks |
+| [Model sharing and uploading](https://huggingface.co/transformers/model_sharing.html) | Upload and share your fine-tuned models with the community |
+| [Migration](https://huggingface.co/transformers/migration.html) | Migrate to 🤗 Transformers from `pytorch-transformers` or `pytorch-pretrained-bert` |
 
 ## Citation
 

README_es.md

@@ -1,444 +0,0 @@
-<!---
-Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
--->
-
-<p align="center">
-    <br>
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
-    <br>
-<p>
-<p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
-        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
-    </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
-    </a>
-    <a href="https://github.com/huggingface/transformers/releases">
-        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
-        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
-    </a>
-    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
-</p>
-
-<h4 align="center">
-    <p>
-        <a href="https://github.com/huggingface/transformers/">English</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <b>Español</b> 
-    <p>
-</h4>
-
-<h3 align="center">
-    <p>Lo último de Machine Learning para JAX, PyTorch y TensorFlow</p>
-</h3>
-
-<h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
-</h3>
-
-🤗 Transformers aporta miles de modelos preentrenados Para realizar tareas en diferentes modalidades como texto, vision, y audio. 
-
-Estos modelos pueden ser aplicados en:
-
-* 📝 Texto, Para tareas como clasificación de texto, extracción de información, responder preguntas, resumir, traducir, generación de texto, en más de 100 idiomas. 
-* 🖼️ Imágenes, para tareas como clasificación de imágenes, detección the objetos, y segmentación. 
-* 🗣️ Audio, para tareas como reconocimiento de voz y clasificación de audio. 
-
-Los modelos de Transformer también pueden realizar tareas en **muchas modalidades combinadas**, como responder pregunstas, reconocimiento de carácteres ópticos,extracción de información de documentos escaneados, clasificación de video, y respuesta de preguntas visuales.
-
-🤗 Transformers aporta APIs para descargar rápidamente y usar estos modelos preentrenados en un texto dado, afinarlos en tus propios sets de datos y compartirlos con la comunidad en nuestro [centro de modelos](https://huggingface.co/models). Al mismo tiempo, cada módulo de Python que define una arquitectura es completamente independiente y se puede modificar para permitir experimentos de investigación rápidos.
-
-🤗 Transformers está respaldado por las tres bibliotecas de deep learning más populares — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) y [TensorFlow](https://www.tensorflow.org/) — con una perfecta integración entre ellos. Es sencillo entrenar sus modelos con uno antes de cargarlos para la inferencia con el otro.
-
-## Demostraciones en línea
-
-Puedes probar la mayoría de nuestros modelos directamente en sus páginas desde el [centro de modelos](https://huggingface.co/models). También ofrecemos [alojamiento de modelos privados, control de versiones y una API de inferencia](https://huggingface.co/pricing) para modelos públicos y privados.
-
-Aquí hay algunos ejemplos:
-
- En procesamiento del lenguaje natural:
-- [Terminación de palabras enmascaradas con BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [Reconocimiento del nombre de la entidad con Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [Generación de texto con GPT-2](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
-- [Inferencia del lenguaje natural con RoBERTa](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
-- [Resumen con BART](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [Responder a preguntas con DistilBERT](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [Traducción con T5](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-En visión de ordenador:
-- [Clasificación de imágenes con ViT](https://huggingface.co/google/vit-base-patch16-224)
-- [Detección de objetos con DETR](https://huggingface.co/facebook/detr-resnet-50)
-- [Segmentación semántica con SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
-- [Segmentación panóptica con DETR](https://huggingface.co/facebook/detr-resnet-50-panoptic)
-
-En Audio:
-- [Reconocimiento de voz automático con Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base-960h)
-- [Detección de palabras clave con Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
-
-En tareas multimodales:
-- [Respuesta visual a preguntas con ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
-
-**[Escribe con Transformer](https://transformer.huggingface.co)**, construido por el equipo de Hugging Face, es la demostración oficial de las capacidades de generación de texto de este repositorio.
-
-## Si está buscando soporte personalizado del equipo de Hugging Face
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-## Tour rápido
-
-Para usar inmediatamente un modelo en una entrada determinada (texto, imagen, audio, ...), proporcionamos la API de `pipeline`. Los pipelines agrupan un modelo previamente entrenado con el preprocesamiento que se usó durante el entrenamiento de ese modelo. Aquí se explica cómo usar rápidamente un pipeline para clasificar textos positivos frente a negativos:
-
-```python
->>> from transformers import pipeline
-
-# Allocate a pipeline for sentiment-analysis
->>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to introduce pipeline to the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
-```
-
-La segunda línea de código descarga y almacena en caché el modelo previamente entrenado que usa la canalización, mientras que la tercera lo evalúa en el texto dado. Aquí la respuesta es "positiva" con una confianza del 99,97%.
-
-Muchas tareas tienen un `pipeline` preentrenado listo para funcionar, en NLP pero también en visión por ordenador y habla. Por ejemplo, podemos extraer fácilmente los objetos detectados en una imagen:
-
-``` python
->>> import requests
->>> from PIL import Image
->>> from transformers import pipeline
-
-# Download an image with cute cats
->>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
->>> image_data = requests.get(url, stream=True).raw
->>> image = Image.open(image_data)
-
-# Allocate a pipeline for object detection
->>> object_detector = pipeline('object_detection')
->>> object_detector(image)
-[{'score': 0.9982201457023621,
-  'label': 'remote',
-  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
- {'score': 0.9960021376609802,
-  'label': 'remote',
-  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
- {'score': 0.9954745173454285,
-  'label': 'couch',
-  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
- {'score': 0.9988006353378296,
-  'label': 'cat',
-  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
- {'score': 0.9986783862113953,
-  'label': 'cat',
-  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
-```
-
-Aquí obtenemos una lista de objetos detectados en la imagen, con un cuadro que rodea el objeto y una puntuación de confianza. Aquí está la imagen original a la derecha, con las predicciones mostradas a la izquierda:
-
-<h3 align="center">
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
-</h3>
-
-Puedes obtener más información sobre las tareas admitidas por la API de `pipeline` en [este tutorial](https://huggingface.co/docs/transformers/task_summary).
-
-Además de `pipeline`, para descargar y usar cualquiera de los modelos previamente entrenados en su tarea dada, todo lo que necesita son tres líneas de código. Aquí está la versión de PyTorch:
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = AutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
-```
-
-Y aquí está el código equivalente para TensorFlow:
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
-```
-
-El tokenizador es responsable de todo el preprocesamiento que espera el modelo preentrenado y se puede llamar directamente en una sola cadena (como en los ejemplos anteriores) o en una lista. Dará como resultado un diccionario que puedes usar en el código descendente o simplemente pasarlo directamente a su modelo usando el operador de desempaquetado de argumento **.
-
-El modelo en si es un [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) normal o un [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (dependiendo De tu backend) que puedes usar de forma habitual. [Este tutorial](https://huggingface.co/docs/transformers/training) explica cómo integrar un modelo de este tipo en un ciclo de entrenamiento PyTorch o TensorFlow clásico, o como usar nuestra API `Trainer` para ajustar rápidamente un nuevo conjunto de datos.
-
-## ¿Por qué debo usar transformers?
-
-1. Modelos de última generación fáciles de usar:
-    - Alto rendimiento en comprensión y generación de lenguaje natural, visión artificial y tareas de audio.
-    - Baja barrera de entrada para educadores y profesionales.
-    - Pocas abstracciones de cara al usuario con solo tres clases para aprender.
-    - Una API unificada para usar todos nuestros modelos preentrenados.
-
-1. Menores costes de cómputo, menor huella de carbono:
-    - Los investigadores pueden compartir modelos entrenados en lugar de siempre volver a entrenar.
-    - Los profesionales pueden reducir el tiempo de cómputo y los costos de producción.
-    - Docenas de arquitecturas con más de 60 000 modelos preentrenados en todas las modalidades.
-
-1. Elija el marco adecuado para cada parte de la vida útil de un modelo:
-    - Entrene modelos de última generación en 3 líneas de código.
-    - Mueva un solo modelo entre los marcos TF2.0/PyTorch/JAX a voluntad.
-    - Elija sin problemas el marco adecuado para la formación, la evaluación y la producción.
-
-1. Personalice fácilmente un modelo o un ejemplo según sus necesidades:
-    - Proporcionamos ejemplos de cada arquitectura para reproducir los resultados publicados por sus autores originales..
-    - Los internos del modelo están expuestos lo más consistentemente posible..
-    - Los archivos modelo se pueden usar independientemente de la biblioteca para experimentos rápidos.
-
-## ¿Por qué no debería usar transformers?
-
-- Esta biblioteca no es una caja de herramientas modular de bloques de construcción para redes neuronales. El código en los archivos del modelo no se refactoriza con abstracciones adicionales a propósito, de modo que los investigadores puedan iterar rápidamente en cada uno de los modelos sin sumergirse en abstracciones/archivos adicionales.
-- La API de entrenamiento no está diseñada para funcionar en ningún modelo, pero está optimizada para funcionar con los modelos proporcionados por la biblioteca. Para bucles genéricos de aprendizaje automático, debe usar otra biblioteca (posiblemente, [Accelerate](https://huggingface.co/docs/accelerate)).
-- Si bien nos esforzamos por presentar tantos casos de uso como sea posible, los scripts en nuestra [carpeta de ejemplos](https://github.com/huggingface/transformers/tree/main/examples) son solo eso: ejemplos. Se espera que no funcionen de forma inmediata en su problema específico y que deba cambiar algunas líneas de código para adaptarlas a sus necesidades.
-
-## Instalación
-
-### Con pip
-
-Este repositorio está probado en Python 3.6+, Flax 0.3.2+, PyTorch 1.3.1+ y TensorFlow 2.3+.
-
-Deberías instalar 🤗 Transformers en un [ambiente virtual](https://docs.python.org/3/library/venv.html). Si no estas familiarizado con los entornos virtuales de Python, consulta la [guía de usuario](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
-
-Primero, crea un entorno virtual con la versión de Python que vas a usar y actívalo.
-
-Luego, deberás instalar al menos uno de Flax, PyTorch o TensorFlow.
-Por favor, ve a la [página de instalación de TensorFlow](https://www.tensorflow.org/install/), [página de instalación de PyTorch](https://pytorch.org/get-started/locally/#start-locally) y/o las páginas de instalación de [Flax](https://github.com/google/flax#quick-install) y [Jax](https://github.com/google/jax#installation) con respecto al comando de instalación específico para tu plataforma.
-
-Cuando se ha instalado uno de esos backends, los 🤗 Transformers se pueden instalar usando pip de la siguiente manera:
-
-```bash
-pip install transformers
-```
-
-Si deseas jugar con los ejemplos o necesitas la última versión del código y no puedes esperar a una nueva versión, tienes que [instalar la librería de la fuente](https://huggingface.co/docs/transformers/installation#installing-from-source).
-
-### Con conda
-
-Desde la versión v4.0.0 de Transformers, ahora tenemos un canal conda: `huggingface`.
-
-🤗 Transformers se puede instalar usando conda de la siguiente manera:
-
-```shell script
-conda install -c huggingface transformers
-```
-
-Sigue las páginas de instalación de Flax, PyTorch o TensorFlow para ver cómo instalarlos con conda.
-
-> **_NOTA:_**  En Windows, es posible que se le pida que active el modo de desarrollador para beneficiarse del almacenamiento en caché. Si esta no es una opción para usted, háganoslo saber en [esta issue](https://github.com/huggingface/huggingface_hub/issues/1062).
-
-## Arquitecturas modelo
-
-**[Todos los puntos de control del modelo](https://huggingface.co/models)** aportados por 🤗 Transformers están perfectamente integrados desde huggingface.co [Centro de modelos](https://huggingface.co) donde son subidos directamente por los [usuarios](https://huggingface.co/users) y [organizaciones](https://huggingface.co/organizations).
-
-Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 Transformers actualmente proporciona las siguientes arquitecturas (ver [aquí](https://huggingface.co/docs/transformers/model_summary) para un resumen de alto nivel de cada uno de ellas.):
-
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2** was released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](https://huggingface.co/docs/transformers/main/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)**  (from HuggingFace).
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-1. ¿Quieres aportar un nuevo modelo? Hemos agregado una **guía detallada y plantillas** para guiarte en el proceso de agregar un nuevo modelo. Puedes encontrarlos en la carpeta de [`templates`](./templates) del repositorio. Asegúrate de revisar las [pautas de contribución](./CONTRIBUTING.md) y comunícate con los mantenedores o abra un problema para recopilar comentarios antes de comenzar su PR.
-
-Para comprobar si cada modelo tiene una implementación en Flax, PyTorch o TensorFlow, o tiene un tokenizador asociado respaldado por la librería 🤗 Tokenizers , ve a [esta tabla](https://huggingface.co/docs/transformers/index#supported-frameworks).
-
-Estas implementaciones se han probado en varios conjuntos de datos (consulte los scripts de ejemplo) y deberían coincidir con el rendimiento de las implementaciones originales. Puede encontrar más detalles sobre el rendimiento en la sección Examples de la [documentación](https://github.com/huggingface/transformers/tree/main/examples).
-
-
-## Aprender más
-
-| Sección | Descripción |
-|-|-|
-| [Documentación](https://huggingface.co/docs/transformers/) | Toda la documentación de la API y tutoriales |
-| [Resumen de tareas](https://huggingface.co/docs/transformers/task_summary) | Tareas soportadas 🤗 Transformers |
-| [Tutorial de preprocesAmiento](https://huggingface.co/docs/transformers/preprocessing) | Usando la clase `Tokenizer` para preparar datos para los modelos |
-| [Entrenamiento y puesta a punto](https://huggingface.co/docs/transformers/training) | Usando los modelos aportados por 🤗 Transformers en un bucle de entreno de PyTorch/TensorFlow y la API de `Trainer` |
-| [Recorrido rápido: secuencias de comandos de ajuste/uso](https://github.com/huggingface/transformers/tree/main/examples) | Scripts de ejemplo para ajustar modelos en una amplia gama de tareas |
-| [Compartir y subir modelos](https://huggingface.co/docs/transformers/model_sharing) | Carga y comparte tus modelos perfeccionados con la comunidad |
-| [Migración](https://huggingface.co/docs/transformers/migration) | Migra a 🤗 Transformers desde `pytorch-transformers` o `pytorch-pretrained-bert` |
-
-## Citación
-
-Ahora nosotros tenemos un [papel](https://www.aclweb.org/anthology/2020.emnlp-demos.6/) que puedes citar para la librería de  🤗 Transformers:
-```bibtex
-@inproceedings{wolf-etal-2020-transformers,
-    title = "Transformers: State-of-the-Art Natural Language Processing",
-    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
-    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
-    month = oct,
-    year = "2020",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
-    pages = "38--45"
-}
-```

README_ko.md

@@ -1,393 +0,0 @@
-<!---
-Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
--->
-
-<p align="center">
-    <br>
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
-    <br>
-<p>
-<p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
-        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
-    </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
-    </a>
-    <a href="https://github.com/huggingface/transformers/releases">
-        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
-        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
-    </a>
-    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
-</p>
-
-<h4 align="center">
-    <p>
-        <a href="https://github.com/huggingface/transformers/">English</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <b>한국어</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> 
-    <p>
-</h4>
-
-<h3 align="center">
-    <p> Jax, Pytorch, TensorFlow를 위한 최첨단 자연어처리</p>
-</h3>
-
-<h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
-</h3>
-
-🤗 Transformers는 분류, 정보 추출, 질문 답변, 요약, 번역, 문장 생성 등을 100개 이상의 언어로 수행할 수 있는 수천개의 사전학습된 모델을 제공합니다. 우리의 목표는 모두가 최첨단의 NLP 기술을 쉽게 사용하는 것입니다.
-
-🤗 Transformers는 이러한 사전학습 모델을 빠르게 다운로드해 특정 텍스트에 사용하고, 원하는 데이터로 fine-tuning해 커뮤니티나 우리의 [모델 허브](https://huggingface.co/models)에 공유할 수 있도록 API를 제공합니다. 또한, 모델 구조를 정의하는 각 파이썬 모듈은 완전히 독립적이여서 연구 실험을 위해 손쉽게 수정할 수 있습니다.
-
-🤗 Transformers는 가장 유명한 3개의 딥러닝 라이브러리를 지원합니다. 이들은 서로 완벽히 연동됩니다 — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/), [TensorFlow](https://www.tensorflow.org/). 간단하게 이 라이브러리 중 하나로 모델을 학습하고, 또 다른 라이브러리로 추론을 위해 모델을 불러올 수 있습니다.
-
-## 온라인 데모
-
-대부분의 모델을 [모델 허브](https://huggingface.co/models) 페이지에서 바로 테스트해볼 수 있습니다. 공개 및 비공개 모델을 위한 [비공개 모델 호스팅, 버전 관리, 추론 API](https://huggingface.co/pricing)도 제공합니다.
-
-예시:
-- [BERT로 마스킹된 단어 완성하기](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [Electra를 이용한 개체명 인식](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [GPT-2로 텍스트 생성하기](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
-- [RoBERTa로 자연어 추론하기](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
-- [BART를 이용한 요약](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [DistilBERT를 이용한 질문 답변](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [T5로 번역하기](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-**[Transformer와 글쓰기](https://transformer.huggingface.co)** 는 이 저장소의 텍스트 생성 능력에 관한 Hugging Face 팀의 공식 데모입니다.
-
-## Hugging Face 팀의 커스텀 지원을 원한다면
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-## 퀵 투어
-
-원하는 텍스트에 바로 모델을 사용할 수 있도록, 우리는 `pipeline` API를 제공합니다. Pipeline은 사전학습 모델과 그 모델을 학습할 때 적용한 전처리 방식을 하나로 합칩니다. 다음은 긍정적인 텍스트와 부정적인 텍스트를 분류하기 위해 pipeline을 사용한 간단한 예시입니다:
-
-```python
->>> from transformers import pipeline
-
-# Allocate a pipeline for sentiment-analysis
->>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to introduce pipeline to the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
-```
-
-코드의 두번째 줄은 pipeline이 사용하는 사전학습 모델을 다운로드하고 캐시로 저장합니다. 세번째 줄에선 그 모델이 주어진 텍스트를 평가합니다. 여기서 모델은 99.97%의 확률로 텍스트가 긍정적이라고 평가했습니다.
-
-많은 NLP 과제들을 `pipeline`으로 바로 수행할 수 있습니다. 예를 들어, 질문과 문맥이 주어지면 손쉽게 답변을 추출할 수 있습니다:
-
-``` python
->>> from transformers import pipeline
-
-# Allocate a pipeline for question-answering
->>> question_answerer = pipeline('question-answering')
->>> question_answerer({
-...     'question': 'What is the name of the repository ?',
-...     'context': 'Pipeline has been included in the huggingface/transformers repository'
-... })
-{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
-
-```
-
-답변뿐만 아니라, 여기에 사용된 사전학습 모델은 확신도와 토크나이즈된 문장 속 답변의 시작점, 끝점까지 반환합니다. [이 튜토리얼](https://huggingface.co/docs/transformers/task_summary)에서 `pipeline` API가 지원하는 다양한 과제를 확인할 수 있습니다.
-
-코드 3줄로 원하는 과제에 맞게 사전학습 모델을 다운로드 받고 사용할 수 있습니다. 다음은 PyTorch 버전입니다:
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = AutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
-```
-다음은 TensorFlow 버전입니다:
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
-```
-
-토크나이저는 사전학습 모델의 모든 전처리를 책임집니다. 그리고 (위의 예시처럼) 1개의 스트링이나 리스트도 처리할 수 있습니다. 토크나이저는 딕셔너리를 반환하는데, 이는 다운스트림 코드에 사용하거나 언패킹 연산자 ** 를 이용해 모델에 바로 전달할 수도 있습니다.
-
-모델 자체는 일반적으로 사용되는 [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)나 [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)입니다. [이 튜토리얼](https://huggingface.co/transformers/training.html)은 이러한 모델을 표준적인 PyTorch나 TensorFlow 학습 과정에서 사용하는 방법, 또는 새로운 데이터로 fine-tune하기 위해 `Trainer` API를 사용하는 방법을 설명해줍니다.
-
-## 왜 transformers를 사용해야 할까요?
-
-1. 손쉽게 사용할 수 있는 최첨단 모델:
-    - NLU와 NLG 과제에서 뛰어난 성능을 보입니다.
-    - 교육자 실무자에게 진입 장벽이 낮습니다.
-    - 3개의 클래스만 배우면 바로 사용할 수 있습니다.
-    - 하나의 API로 모든 사전학습 모델을 사용할 수 있습니다.
-
-1. 더 적은 계산 비용, 더 적은 탄소 발자국:
-    - 연구자들은 모델을 계속 다시 학습시키는 대신 학습된 모델을 공유할 수 있습니다.
-    - 실무자들은 학습에 필요한 시간과 비용을 절약할 수 있습니다.
-    - 수십개의 모델 구조, 2,000개 이상의 사전학습 모델, 100개 이상의 언어로 학습된 모델 등.
-
-1. 모델의 각 생애주기에 적합한 프레임워크:
-    - 코드 3줄로 최첨단 모델을 학습하세요.
-    - 자유롭게 모델을 TF2.0나 PyTorch 프레임워크로 변환하세요.
-    - 학습, 평가, 공개 등 각 단계에 맞는 프레임워크를 원하는대로 선택하세요.
-
-1. 필요한 대로 모델이나 예시를 커스터마이즈하세요:
-    - 우리는 저자가 공개한 결과를 재현하기 위해 각 모델 구조의 예시를 제공합니다.
-    - 모델 내부 구조는 가능한 일관적으로 공개되어 있습니다.
-    - 빠른 실험을 위해 모델 파일은 라이브러리와 독립적으로 사용될 수 있습니다.
-
-## 왜 transformers를 사용하지 말아야 할까요?
-
-- 이 라이브러리는 신경망 블록을 만들기 위한 모듈이 아닙니다. 연구자들이 여러 파일을 살펴보지 않고 바로 각 모델을 사용할 수 있도록, 모델 파일 코드의 추상화 수준을 적정하게 유지했습니다.
-- 학습 API는 모든 모델에 적용할 수 있도록 만들어지진 않았지만, 라이브러리가 제공하는 모델들에 적용할 수 있도록 최적화되었습니다. 일반적인 머신 러닝을 위해선, 다른 라이브러리를 사용하세요.
-- 가능한 많은 사용 예시를 보여드리고 싶어서, [예시 폴더](https://github.com/huggingface/transformers/tree/main/examples)의 스크립트를 준비했습니다. 이 스크립트들을 수정 없이 특정한 문제에 바로 적용하지 못할 수 있습니다. 필요에 맞게 일부 코드를 수정해야 할 수 있습니다.
-
-## 설치
-
-### pip로 설치하기
-
-이 저장소는 Python 3.6+, Flax 0.3.2+, PyTorch 1.3.1+, TensorFlow 2.3+에서 테스트 되었습니다.
-
-[가상 환경](https://docs.python.org/3/library/venv.html)에 🤗 Transformers를 설치하세요. Python 가상 환경에 익숙하지 않다면, [사용자 가이드](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)를 확인하세요.
-
-우선, 사용할 Python 버전으로 가상 환경을 만들고 실행하세요.
-
-그 다음, Flax, PyTorch, TensorFlow 중 적어도 하나는 설치해야 합니다.
-플랫폼에 맞는 설치 명령어를 확인하기 위해 [TensorFlow 설치 페이지](https://www.tensorflow.org/install/), [PyTorch 설치 페이지](https://pytorch.org/get-started/locally/#start-locally), [Flax 설치 페이지](https://github.com/google/flax#quick-install)를 확인하세요.
-
-이들 중 적어도 하나가 설치되었다면, 🤗 Transformers는 다음과 같이 pip을 이용해 설치할 수 있습니다:
-
-```bash
-pip install transformers
-```
-
-예시들을 체험해보고 싶거나, 최최최첨단 코드를 원하거나, 새로운 버전이 나올 때까지 기다릴 수 없다면 [라이브러리를 소스에서 바로 설치](https://huggingface.co/docs/transformers/installation#installing-from-source)하셔야 합니다.
-
-### conda로 설치하기
-
-Transformers 버전 v4.0.0부터, conda 채널이 생겼습니다: `huggingface`.
-
-🤗 Transformers는 다음과 같이 conda로 설치할 수 있습니다:
-
-```shell script
-conda install -c huggingface transformers
-```
-
-Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는 방법을 확인하세요.
-
-## 모델 구조
-
-**🤗 Transformers가 제공하는 [모든 모델 체크포인트](https://huggingface.co/models)** 는 huggingface.co [모델 허브](https://huggingface.co)에 완벽히 연동되어 있습니다. [개인](https://huggingface.co/users)과 [기관](https://huggingface.co/organizations)이 모델 허브에 직접 업로드할 수 있습니다.
-
-현재 사용 가능한 모델 체크포인트의 개수: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 Transformers는 다음 모델들을 제공합니다 (각 모델의 요약은 [여기](https://huggingface.co/docs/transformers/model_summary)서 확인하세요):
-
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/distillation) and a German version of DistilBERT.
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER) released with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2** was released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](https://huggingface.co/docs/transformers/main/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI) released with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-1. 새로운 모델을 올리고 싶나요? 우리가 **상세한 가이드와 템플릿** 으로 새로운 모델을 올리도록 도와드릴게요. 가이드와 템플릿은 이 저장소의 [`templates`](./templates) 폴더에서 확인하실 수 있습니다. [컨트리뷰션 가이드라인](./CONTRIBUTING.md)을 꼭 확인해주시고, PR을 올리기 전에 메인테이너에게 연락하거나 이슈를 오픈해 피드백을 받으시길 바랍니다.
-
-각 모델이 Flax, PyTorch, TensorFlow으로 구현되었는지 또는 🤗 Tokenizers 라이브러리가 지원하는 토크나이저를 사용하는지 확인하려면, [이 표](https://huggingface.co/docs/transformers/index#supported-frameworks)를 확인하세요.
-
-이 구현은 여러 데이터로 검증되었고 (예시 스크립트를 참고하세요) 오리지널 구현의 성능과 같아야 합니다. [도큐먼트](https://huggingface.co/docs/transformers/examples)의 Examples 섹션에서 성능에 대한 자세한 설명을 확인할 수 있습니다.
-
-## 더 알아보기
-
-| 섹션 | 설명 |
-|-|-|
-| [도큐먼트](https://huggingface.co/transformers/) | 전체 API 도큐먼트와 튜토리얼 |
-| [과제 요약](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers가 지원하는 과제들 |
-| [전처리 튜토리얼](https://huggingface.co/docs/transformers/preprocessing) | `Tokenizer` 클래스를 이용해 모델을 위한 데이터 준비하기 |
-| [학습과 fine-tuning](https://huggingface.co/docs/transformers/training) | 🤗 Transformers가 제공하는 모델 PyTorch/TensorFlow 학습 과정과 `Trainer` API에서 사용하기 |
-| [퀵 투어: Fine-tuning/사용 스크립트](https://github.com/huggingface/transformers/tree/main/examples) | 다양한 과제에서 모델 fine-tuning하는 예시 스크립트 |
-| [모델 공유 및 업로드](https://huggingface.co/docs/transformers/model_sharing) | 커뮤니티에 fine-tune된 모델을 업로드 및 공유하기 |
-| [마이그레이션](https://huggingface.co/docs/transformers/migration) | `pytorch-transformers`나 `pytorch-pretrained-bert`에서 🤗 Transformers로 이동하기|
-
-## 인용
-
-🤗 Transformers 라이브러리를 인용하고 싶다면, 이 [논문](https://www.aclweb.org/anthology/2020.emnlp-demos.6/)을 인용해 주세요:
-```bibtex
-@inproceedings{wolf-etal-2020-transformers,
-    title = "Transformers: State-of-the-Art Natural Language Processing",
-    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
-    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
-    month = oct,
-    year = "2020",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
-    pages = "38--45"
-}
-```

README_zh-hans.md

@@ -1,418 +0,0 @@
-<!---
-Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
--->
-
-<!---
-A useful guide for English-Chinese translation of Hugging Face documentation
-- Add space around English words and numbers when they appear between Chinese characters. E.g., 共 100 多种语言; 使用 transformers 库。
-- Use square quotes, e.g.,「引用」
-
-Dictionary
-
-Hugging Face: 抱抱脸
-token: 词符（并用括号标注原英文）
-tokenize: 词符化（并用括号标注原英文）
-tokenizer: 词符化器（并用括号标注原英文）
-transformer: transformer（不翻译）
-pipeline: 流水线
-API: API (不翻译）
-inference: 推理
-Trainer: 训练器。当作为类名出现时不翻译。
-pretrained/pretrain: 预训练
-finetune: 微调
-community: 社区
-example: 当特指仓库中 example 目录时翻译为「用例」
-Python data structures (e.g., list, set, dict): 翻译为列表，集合，词典，并用括号标注原英文
-NLP/Natural Language Processing: 以 NLP 出现时不翻译，以 Natural Language Processing 出现时翻译为自然语言处理
-checkpoint: 检查点
--->
-
-<p align="center">
-    <br>
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
-    <br>
-<p>
-<p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
-        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
-    </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
-    </a>
-    <a href="https://github.com/huggingface/transformers/releases">
-        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
-        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
-    </a>
-    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
-</p>
-
-<h4 align="center">
-    <p>
-        <a href="https://github.com/huggingface/transformers/">English</a> |
-        <b>简体中文</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> 
-    <p>
-</h4>
-
-<h3 align="center">
-    <p>为 Jax、PyTorch 和 TensorFlow 打造的先进的自然语言处理</p>
-</h3>
-
-<h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
-</h3>
-
-🤗 Transformers 提供了数以千计的预训练模型，支持 100 多种语言的文本分类、信息抽取、问答、摘要、翻译、文本生成。它的宗旨让最先进的 NLP 技术人人易用。
-
-🤗 Transformers 提供了便于快速下载和使用的API，让你可以把预训练模型用在给定文本、在你的数据集上微调然后通过 [model hub](https://huggingface.co/models) 与社区共享。同时，每个定义的 Python 模块均完全独立，方便修改和快速研究实验。
-
-🤗 Transformers 支持三个最热门的深度学习库： [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — 并与之无缝整合。你可以直接使用一个框架训练你的模型然后用另一个加载和推理。
-
-## 在线演示
-
-你可以直接在模型页面上测试大多数 [model hub](https://huggingface.co/models) 上的模型。 我们也提供了 [私有模型托管、模型版本管理以及推理API](https://huggingface.co/pricing)。
-
-这里是一些例子：
-- [用 BERT 做掩码填词](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [用 Electra 做命名实体识别](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [用 GPT-2 做文本生成](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
-- [用 RoBERTa 做自然语言推理](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
-- [用 BART 做文本摘要](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [用 DistilBERT 做问答](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [用 T5 做翻译](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-**[Write With Transformer](https://transformer.huggingface.co)**，由抱抱脸团队打造，是一个文本生成的官方 demo。
-
-## 如果你在寻找由抱抱脸团队提供的定制化支持服务
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-## 快速上手
-
-我们为快速使用模型提供了 `pipeline` （流水线）API。流水线聚合了预训练模型和对应的文本预处理。下面是一个快速使用流水线去判断正负面情绪的例子：
-
-```python
->>> from transformers import pipeline
-
-# 使用情绪分析流水线
->>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to introduce pipeline to the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
-```
-
-第二行代码下载并缓存了流水线使用的预训练模型，而第三行代码则在给定的文本上进行了评估。这里的答案“正面” (positive) 具有 99 的置信度。
-
-许多的 NLP 任务都有开箱即用的预训练流水线。比如说，我们可以轻松的从给定文本中抽取问题答案：
-
-``` python
->>> from transformers import pipeline
-
-# 使用问答流水线
->>> question_answerer = pipeline('question-answering')
->>> question_answerer({
-...     'question': 'What is the name of the repository ?',
-...     'context': 'Pipeline has been included in the huggingface/transformers repository'
-... })
-{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
-
-```
-
-除了给出答案，预训练模型还给出了对应的置信度分数、答案在词符化 (tokenized) 后的文本中开始和结束的位置。你可以从[这个教程](https://huggingface.co/docs/transformers/task_summary)了解更多流水线API支持的任务。
-
-要在你的任务上下载和使用任意预训练模型也很简单，只需三行代码。这里是 PyTorch 版的示例：
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = AutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
-```
-这里是等效的 TensorFlow 代码：
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
-```
-
-词符化器 (tokenizer) 为所有的预训练模型提供了预处理，并可以直接对单个字符串进行调用（比如上面的例子）或对列表 (list) 调用。它会输出一个你可以在下游代码里使用或直接通过 `**` 解包表达式传给模型的词典 (dict)。
-
-模型本身是一个常规的 [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) 或 [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)（取决于你的后端），可以常规方式使用。 [这个教程](https://huggingface.co/transformers/training.html)解释了如何将这样的模型整合到经典的 PyTorch 或 TensorFlow 训练循环中，或是如何使用我们的 `Trainer` 训练器）API 来在一个新的数据集上快速微调。
-
-## 为什么要用 transformers？
-
-1. 便于使用的先进模型：
-    - NLU 和 NLG 上表现优越
-    - 对教学和实践友好且低门槛
-    - 高级抽象，只需了解三个类
-    - 对所有模型统一的API
-
-1. 更低计算开销，更少的碳排放：
-    - 研究人员可以分享已训练的模型而非每次从头开始训练
-    - 工程师可以减少计算用时和生产环境开销
-    - 数十种模型架构、两千多个预训练模型、100多种语言支持
-
-1. 对于模型生命周期的每一个部分都面面俱到：
-    - 训练先进的模型，只需 3 行代码
-    - 模型在不同深度学习框架间任意转移，随你心意
-    - 为训练、评估和生产选择最适合的框架，衔接无缝
-
-1. 为你的需求轻松定制专属模型和用例：
-    - 我们为每种模型架构提供了多个用例来复现原论文结果
-    - 模型内部结构保持透明一致
-    - 模型文件可单独使用，方便魔改和快速实验
-
-## 什么情况下我不该用 transformers？
-
-- 本库并不是模块化的神经网络工具箱。模型文件中的代码特意呈若璞玉，未经额外抽象封装，以便研究人员快速迭代魔改而不致溺于抽象和文件跳转之中。
-- `Trainer` API 并非兼容任何模型，只为本库之模型优化。若是在寻找适用于通用机器学习的训练循环实现，请另觅他库。
-- 尽管我们已尽力而为，[examples 目录](https://github.com/huggingface/transformers/tree/main/examples)中的脚本也仅为用例而已。对于你的特定问题，它们并不一定开箱即用，可能需要改几行代码以适之。
-
-## 安装
-
-### 使用 pip
-
-这个仓库已在 Python 3.6+、Flax 0.3.2+、PyTorch 1.3.1+ 和 TensorFlow 2.3+ 下经过测试。
-
-你可以在[虚拟环境](https://docs.python.org/3/library/venv.html)中安装 🤗 Transformers。如果你还不熟悉 Python 的虚拟环境，请阅此[用户说明](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)。
-
-首先，用你打算使用的版本的 Python 创建一个虚拟环境并激活。
-
-然后，你需要安装 Flax、PyTorch 或 TensorFlow 其中之一。关于在你使用的平台上安装这些框架，请参阅 [TensorFlow 安装页](https://www.tensorflow.org/install/), [PyTorch 安装页](https://pytorch.org/get-started/locally/#start-locally) 或 [Flax 安装页](https://github.com/google/flax#quick-install)。
-
-当这些后端之一安装成功后， 🤗 Transformers 可依此安装：
-
-```bash
-pip install transformers
-```
-
-如果你想要试试用例或者想在正式发布前使用最新的开发中代码，你得[从源代码安装](https://huggingface.co/docs/transformers/installation#installing-from-source)。
-
-### 使用 conda
-
-自 Transformers 4.0.0 版始，我们有了一个 conda 频道： `huggingface`。
-
-🤗 Transformers 可以通过 conda 依此安装：
-
-```shell script
-conda install -c huggingface transformers
-```
-
-要通过 conda 安装 Flax、PyTorch 或 TensorFlow 其中之一，请参阅它们各自安装页的说明。
-
-## 模型架构
-
-🤗 Transformers 支持的[**所有的模型检查点**](https://huggingface.co/models)由[用户](https://huggingface.co/users)和[组织](https://huggingface.co/organizations)上传，均与 huggingface.co [model hub](https://huggingface.co) 无缝整合。
-
-目前的检查点数量： ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 Transformers 目前支持如下的架构（模型概述请阅[这里](https://huggingface.co/docs/transformers/model_summary)）：
-
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (来自 Google Research and the Toyota Technological Institute at Chicago) 伴随论文 [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), 由 Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut 发布。
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (来自 Facebook) 伴随论文 [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) 由 Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer 发布。
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (来自 École polytechnique) 伴随论文 [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) 由 Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis 发布。
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (来自 VinAI Research) 伴随论文 [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) 由 Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen 发布。
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (来自 Microsoft) 伴随论文 [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) 由 Hangbo Bao, Li Dong, Furu Wei 发布。
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (来自 Google) 伴随论文 [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) 由 Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova 发布。
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (来自 Google) 伴随论文 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 由 Sascha Rothe, Shashi Narayan, Aliaksei Severyn 发布。
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (来自 VinAI Research) 伴随论文 [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) 由 Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen 发布。
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (来自 Google Research) 伴随论文 [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) 由 Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed 发布。
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (来自 Google Research) 伴随论文 [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) 由 Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed 发布。
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (来自 Facebook) 伴随论文 [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) 由 Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston 发布。
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (来自 Facebook) 伴随论文 [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) 由 Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston 发布。
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (来自 Alexa) 伴随论文 [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) 由 Adrian de Wynter and Daniel J. Perry 发布。
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (来自 Google Research) 伴随论文 [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) 由 Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel 发布。
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (来自 Inria/Facebook/Sorbonne) 伴随论文 [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) 由 Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot 发布。
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (来自 Google Research) 伴随论文 [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) 由 Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting 发布。
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (来自 OpenAI) 伴随论文 [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) 由 Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever 发布。
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (来自 Salesforce) 伴随论文 [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) 由 Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong 发布。
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (来自 Microsoft Research Asia) 伴随论文 [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) 由 Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang 发布。
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (来自 YituTech) 伴随论文 [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) 由 Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan 发布。
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (来自 Facebook AI) 伴随论文 [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) 由 Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie 发布。
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (来自 Tsinghua University) 伴随论文 [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) 由 Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun 发布。
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (来自 Salesforce) 伴随论文 [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) 由 Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher 发布。
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (来自 Microsoft) 伴随论文 [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) 由 Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang 发布。
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (来自 Facebook) 伴随论文 [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) 由 Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli 发布。
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (来自 Microsoft) 伴随论文 [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) 由 Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen 发布。
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (来自 Microsoft) 伴随论文 [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) 由 Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen 发布。
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (来自 Berkeley/Facebook/Google) 伴随论文 [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) 由 Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch 发布。
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (来自 SenseTime Research) 伴随论文 [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) 由 Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai 发布。
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (来自 Facebook) 伴随论文 [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) 由 Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou 发布。
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (来自 Facebook) 伴随论文 [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) 由 Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko 发布。
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (来自 Microsoft Research) 伴随论文 [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) 由 Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan 发布。
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (来自 HuggingFace), 伴随论文 [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) 由 Victor Sanh, Lysandre Debut and Thomas Wolf 发布。 同样的方法也应用于压缩 GPT-2 到 [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/distillation), RoBERTa 到 [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/distillation), Multilingual BERT 到 [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/distillation) 和德语版 DistilBERT。
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (来自 Microsoft Research) 伴随论文 [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) 由 Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei 发布。
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (来自 NAVER) 伴随论文 [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) 由 Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park 发布。
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (来自 Facebook) 伴随论文 [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) 由 Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih 发布。
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (来自 Intel Labs) 伴随论文 [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) 由 René Ranftl, Alexey Bochkovskiy, Vladlen Koltun 发布。
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (来自 Google Research/Stanford University) 伴随论文 [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) 由 Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning 发布。
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (来自 Google Research) 伴随论文 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 由 Sascha Rothe, Shashi Narayan, Aliaksei Severyn 发布。
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (来自 Baidu) 伴随论文 [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu 发布。
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2** was released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (来自 CNRS) 伴随论文 [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) 由 Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab 发布。
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (来自 Facebook AI) 伴随论文 [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) 由 Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela 发布。
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (来自 Google Research) 伴随论文 [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) 由 James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon 发布。
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (来自 CMU/Google Brain) 伴随论文 [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) 由 Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le 发布。
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (来自 KAIST) 伴随论文 [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) 由 Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim 发布。
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (来自 OpenAI) 伴随论文 [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) 由 Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever 发布。
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (来自 EleutherAI) 随仓库 [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) 发布。作者为 Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy 发布。
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (来自 ABEJA) 由 Shinya Otani, Takayoshi Makabe, Anuj Arora, Kyo Hattori。
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (来自 OpenAI) 伴随论文 [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) 由 Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever** 发布。
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (来自 EleutherAI) 伴随论文 [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) 由 Ben Wang and Aran Komatsuzaki 发布。
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (来自 UCSD, NVIDIA) 伴随论文 [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) 由 Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang 发布。
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (来自 Facebook) 伴随论文 [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) 由 Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed 发布。
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (来自 Berkeley) 伴随论文 [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) 由 Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer 发布。
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (来自 OpenAI) 伴随论文 [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) 由 Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever 发布。
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) 由 Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou 发布。
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) 由 Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou 发布。
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) 由 Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei 发布。
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (来自 Microsoft Research Asia) 伴随论文 [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) 由 Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei 发布。
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (来自 AllenAI) 伴随论文 [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) 由 Iz Beltagy, Matthew E. Peters, Arman Cohan 发布。
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (来自 Meta AI) 伴随论文 [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) 由 Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze 发布。
-1. **[LiLT](https://huggingface.co/docs/transformers/main/model_doc/lilt)** (来自 South China University of Technology) 伴随论文 [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) 由 Jiapeng Wang, Lianwen Jin, Kai Ding 发布。
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (来自 AllenAI) 伴随论文 [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) 由 Iz Beltagy, Matthew E. Peters, Arman Cohan 发布。
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (来自 Google AI) released 伴随论文 [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) 由 Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang 发布。
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (来自 Studio Ousia) 伴随论文 [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) 由 Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto 发布。
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (来自 UNC Chapel Hill) 伴随论文 [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) 由 Hao Tan and Mohit Bansal 发布。
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (来自 Facebook) 伴随论文 [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) 由 Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert 发布。
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (来自 Facebook) 伴随论文 [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) 由 Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin 发布。
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** 用 [OPUS](http://opus.nlpl.eu/) 数据训练的机器翻译模型由 Jörg Tiedemann 发布。[Marian Framework](https://marian-nmt.github.io/) 由微软翻译团队开发。
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (来自 Microsoft Research Asia) 伴随论文 [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) 由 Junlong Li, Yiheng Xu, Lei Cui, Furu Wei 发布。
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov  >>>>>>> Fix rebase
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (来自 Facebook) 伴随论文 [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) 由 Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer 发布。
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (来自 Facebook) 伴随论文 [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) 由 Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan 发布。
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (来自 NVIDIA) 伴随论文 [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) 由 Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro 发布。
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (来自 NVIDIA) 伴随论文 [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) 由 Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro 发布。
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (来自 Studio Ousia) 伴随论文 [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) 由 Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka 发布。
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (来自 CMU/Google Brain) 伴随论文 [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) 由 Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou 发布。
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (来自 Apple) 伴随论文 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 由 Sachin Mehta and Mohammad Rastegari 发布。
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (来自 Microsoft Research) 伴随论文 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 由 Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 发布。
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (来自 Google AI) 伴随论文 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 由 Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 发布。
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (来自华为诺亚方舟实验室) 伴随论文 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 由 Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 发布。
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (来自 the University of Wisconsin - Madison) 伴随论文 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 由 Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 发布。
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (来自 Meta AI) 伴随论文 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 由 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 发布。
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (来自 Google AI) 伴随论文 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 由 Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 发布。
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (来自 Google) 伴随论文 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 由 Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 发布。
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (来自 Google) 伴随论文 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 由 Jason Phang, Yao Zhao, Peter J. Liu 发布。
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (来自 Deepmind) 伴随论文 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 由 Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira 发布。
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (来自 VinAI Research) 伴随论文 [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) 由 Dat Quoc Nguyen and Anh Tuan Nguyen 发布。
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (来自 UCLA NLP) 伴随论文 [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) 由 Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang 发布。
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (来自 Sea AI Labs) 伴随论文 [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) 由 Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng 发布。
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (来自 Microsoft Research) 伴随论文 [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) 由 Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou 发布。
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (来自 NVIDIA) 伴随论文 [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) 由 Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius 发布。
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (来自 Facebook) 伴随论文 [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) 由 Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela 发布。
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (来自 Google Research) 伴随论文 [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) 由 Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang 发布。
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (来自 Google Research) 伴随论文 [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) 由 Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya 发布。
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (来自 Google Research) 伴随论文 [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) 由 Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder 发布。
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (来自 Facebook), 伴随论文 [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) 由 Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov 发布。
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (来自 ZhuiyiTechnology), 伴随论文 [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) 由 Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu 发布。
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (来自 NVIDIA) 伴随论文 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 由 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 发布。
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (来自 Facebook), 伴随论文 [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) 由 Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino 发布。
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (来自 Facebook) 伴随论文 [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) 由 Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau 发布。
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (来自 Tel Aviv University) 伴随论文 [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) 由 Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy 发布。
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (来自 Berkeley) 伴随论文 [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) 由 Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer 发布。
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (来自 Microsoft) 伴随论文 [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) 由 Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo 发布。
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (来自 Microsoft) 伴随论文 [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) 由 Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo 发布。
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (来自 Google AI) 伴随论文 [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) 由 Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu 发布。
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (来自 Google AI) 伴随论文 [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) 由 Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu 发布。
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (来自 Google AI) 伴随论文 [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) 由 Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos 发布。
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (来自 Microsoft Research) 伴随论文 [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) 由 Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou 发布。
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (来自 Google/CMU) 伴随论文 [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) 由 Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov 发布。
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (来自 Microsoft) 伴随论文 [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) 由 Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei 发布。
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (来自 Microsoft Research) 伴随论文 [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) 由 Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang 发布。
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (来自 Microsoft Research) 伴随论文 [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) 由 Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu 发布。
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (来自 Tsinghua University and Nankai University) 伴随论文 [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) 由 Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu 发布。
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (来自 Multimedia Computing Group, Nanjing University) 伴随论文 [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) 由 Zhan Tong, Yibing Song, Jue Wang, Limin Wang 发布。
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (来自 NAVER AI Lab/Kakao Enterprise/Kakao Brain) 伴随论文 [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) 由 Wonjae Kim, Bokyung Son, Ildoo Kim 发布。
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (来自 Google AI) 伴随论文 [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) 由 Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby 发布。
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (来自 UCLA NLP) 伴随论文 [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) 由 Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang 发布。
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (来自 Meta AI) 伴随论文 [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) 由 Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick 发布。
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (来自 Meta AI) 伴随论文 [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas 发布.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (来自 Facebook AI) 伴随论文 [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) 由 Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli 发布。
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (来自 Facebook AI) 伴随论文 [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) 由 Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino 发布。
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (来自 Facebook AI) 伴随论文 [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) 由 Qiantong Xu, Alexei Baevski, Michael Auli 发布。
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (来自 OpenAI) 伴随论文 [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) 由 Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever 发布。
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (来自 Microsoft Research) 伴随论文 [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) 由 Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling 发布。
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (来自 Facebook) 伴随论文 [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) 由 Guillaume Lample and Alexis Conneau 发布。
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (来自 Microsoft Research) 伴随论文 [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) 由 Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou 发布。
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (来自 Facebook AI), 伴随论文 [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) 由 Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov 发布。
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (来自 Facebook AI) 伴随论文 [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) 由 Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau 发布。
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (来自 Google/CMU) 伴随论文 [XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) 由 Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le 发布。
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (来自 Facebook AI) 伴随论文 [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) 由 Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli 发布。
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (来自 Facebook AI) 伴随论文 [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) 由 Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli 发布。
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (来自 Huazhong University of Science & Technology) 伴随论文 [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) 由 Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu 发布。
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (来自 the University of Wisconsin - Madison) 伴随论文 [You Only Sample (Almost) 由 Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh 发布。
-1. 想要贡献新的模型？我们这里有一份**详细指引和模板**来引导你添加新的模型。你可以在 [`templates`](./templates) 目录中找到他们。记得查看 [贡献指南](./CONTRIBUTING.md) 并在开始写 PR 前联系维护人员或开一个新的 issue 来获得反馈。
-
-要检查某个模型是否已有 Flax、PyTorch 或 TensorFlow 的实现，或其是否在 🤗 Tokenizers 库中有对应词符化器（tokenizer），敬请参阅[此表](https://huggingface.co/docs/transformers/index#supported-frameworks)。
-
-这些实现均已于多个数据集测试（请参看用例脚本）并应于原版实现表现相当。你可以在用例文档的[此节](https://huggingface.co/docs/transformers/examples)中了解表现的细节。
-
-
-## 了解更多
-
-| 章节 | 描述 |
-|-|-|
-| [文档](https://huggingface.co/transformers/) | 完整的 API 文档和教程 |
-| [任务总结](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers 支持的任务 |
-| [预处理教程](https://huggingface.co/docs/transformers/preprocessing) | 使用 `Tokenizer` 来为模型准备数据 |
-| [训练和微调](https://huggingface.co/docs/transformers/training) | 在 PyTorch/TensorFlow 的训练循环或 `Trainer` API 中使用 🤗 Transformers 提供的模型 |
-| [快速上手：微调和用例脚本](https://github.com/huggingface/transformers/tree/main/examples) | 为各种任务提供的用例脚本 |
-| [模型分享和上传](https://huggingface.co/docs/transformers/model_sharing) | 和社区上传和分享你微调的模型 |
-| [迁移](https://huggingface.co/docs/transformers/migration) | 从 `pytorch-transformers` 或 `pytorch-pretrained-bert` 迁移到 🤗 Transformers |
-
-## 引用
-
-我们已将此库的[论文](https://www.aclweb.org/anthology/2020.emnlp-demos.6/)正式发表，如果你使用了 🤗 Transformers 库，请引用:
-```bibtex
-@inproceedings{wolf-etal-2020-transformers,
-    title = "Transformers: State-of-the-Art Natural Language Processing",
-    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
-    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
-    month = oct,
-    year = "2020",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
-    pages = "38--45"
-}
-```

README_zh-hant.md

@@ -1,430 +0,0 @@
-<!---
-Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
--->
-
-<!---
-A useful guide for English-Traditional Chinese translation of Hugging Face documentation
-- Add space around English words and numbers when they appear between Chinese characters. E.g., 共 100 多種語言; 使用 transformers 函式庫。
-- Use square quotes, e.g.,「引用」
-- Some of terms in the file can be found at National Academy for Educational Research (https://terms.naer.edu.tw/), an official website providing bilingual translations between English and Traditional Chinese.
-
-Dictionary
-
-API: API (不翻譯）
-add: 加入
-checkpoint: 檢查點
-code: 程式碼
-community: 社群
-confidence: 信賴度
-dataset: 資料集
-documentation: 文件
-example: 基本翻譯為「範例」，或依語意翻為「例子」
-finetune: 微調
-Hugging Face: Hugging Face（不翻譯）
-implementation: 實作
-inference: 推論
-library: 函式庫
-module: 模組
-NLP/Natural Language Processing: 以 NLP 出現時不翻譯，以 Natural Language Processing 出現時翻譯為自然語言處理
-online demos: 線上Demo
-pipeline: pipeline（不翻譯）
-pretrained/pretrain: 預訓練
-Python data structures (e.g., list, set, dict): 翻譯為串列，集合，字典，並用括號標註原英文
-repository: repository（不翻譯）
-summary: 概覽
-token-: token-（不翻譯）
-Trainer: Trainer（不翻譯）
-transformer: transformer（不翻譯）
-tutorial: 教學
-user: 使用者
--->
-
-<p align="center">
-    <br>
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
-    <br>
-<p>
-<p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
-        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
-    </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
-    </a>
-    <a href="https://github.com/huggingface/transformers/releases">
-        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
-        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
-    </a>
-    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
-</p>
-
-<h4 align="center">
-    <p>
-        <a href="https://github.com/huggingface/transformers/">English</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <b>繁體中文</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> 
-    <p>
-</h4>
-
-<h3 align="center">
-    <p>為 Jax、PyTorch 以及 TensorFlow 打造的先進自然語言處理函式庫</p>
-</h3>
-
-<h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
-</h3>
-
-🤗 Transformers 提供了數以千計的預訓練模型，支援 100 多種語言的文本分類、資訊擷取、問答、摘要、翻譯、文本生成。它的宗旨是讓最先進的 NLP 技術人人易用。
-
-🤗 Transformers 提供了便於快速下載和使用的API，讓你可以將預訓練模型用在給定文本、在你的資料集上微調然後經由 [model hub](https://huggingface.co/models) 與社群共享。同時，每個定義的 Python 模組架構均完全獨立，方便修改和快速研究實驗。
-
-🤗 Transformers 支援三個最熱門的深度學習函式庫： [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) 以及 [TensorFlow](https://www.tensorflow.org/) — 並與之完美整合。你可以直接使用其中一個框架訓練你的模型，然後用另一個載入和推論。
-
-## 線上Demo
-
-你可以直接在 [model hub](https://huggingface.co/models) 上測試大多數的模型。我們也提供了 [私有模型託管、模型版本管理以及推論API](https://huggingface.co/pricing)。
-
-這裡是一些範例：
-- [用 BERT 做遮蓋填詞](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [用 Electra 做專有名詞辨識](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [用 GPT-2 做文本生成](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
-- [用 RoBERTa 做自然語言推論](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
-- [用 BART 做文本摘要](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [用 DistilBERT 做問答](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [用 T5 做翻譯](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-**[Write With Transformer](https://transformer.huggingface.co)**，由 Hugging Face 團隊所打造，是一個文本生成的官方 demo。
-
-## 如果你在尋找由 Hugging Face 團隊所提供的客製化支援服務
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-## 快速上手
-
-我們為快速使用模型提供了 `pipeline` API。 Pipeline 包含了預訓練模型和對應的文本預處理。下面是一個快速使用 pipeline 去判斷正負面情緒的例子：
-
-```python
->>> from transformers import pipeline
-
-# 使用情緒分析 pipeline
->>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to introduce pipeline to the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
-```
-
-第二行程式碼下載並快取 pipeline 使用的預訓練模型，而第三行程式碼則在給定的文本上進行了評估。這裡的答案“正面” (positive) 具有 99.97% 的信賴度。
-
-許多的 NLP 任務都有隨選即用的預訓練 `pipeline`。例如，我們可以輕鬆地從給定文本中擷取問題答案：
-
-``` python
->>> from transformers import pipeline
-
-# 使用問答 pipeline
->>> question_answerer = pipeline('question-answering')
->>> question_answerer({
-...     'question': 'What is the name of the repository ?',
-...     'context': 'Pipeline has been included in the huggingface/transformers repository'
-... })
-{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
-
-```
-
-除了提供問題解答，預訓練模型還提供了對應的信賴度分數以及解答在 tokenized 後的文本中開始和結束的位置。你可以從[這個教學](https://huggingface.co/docs/transformers/task_summary)了解更多 `pipeline` API支援的任務。
-
-要在你的任務中下載和使用任何預訓練模型很簡單，只需三行程式碼。這裡是 PyTorch 版的範例：
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = AutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
-```
-這裡是對應的 TensorFlow 程式碼：
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
-```
-
-Tokenizer 為所有的預訓練模型提供了預處理，並可以直接轉換單一字串（比如上面的例子）或串列 (list)。它會輸出一個的字典 (dict) 讓你可以在下游程式碼裡使用或直接藉由 `**` 運算式傳給模型。
-
-模型本身是一個常規的 [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) 或 [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)（取決於你的後端），可依常規方式使用。 [這個教學](https://huggingface.co/transformers/training.html)解釋了如何將這樣的模型整合到一般的 PyTorch 或 TensorFlow 訓練迴圈中，或是如何使用我們的 `Trainer` API 在一個新的資料集上快速進行微調。
-
-## 為什麼要用 transformers？
-
-1. 便於使用的先進模型：
-    - NLU 和 NLG 上性能卓越
-    - 對教學和實作友好且低門檻
-    - 高度抽象，使用者只須學習 3 個類別
-    - 對所有模型使用的制式化API
-
-1. 更低的運算成本，更少的碳排放：
-    - 研究人員可以分享已訓練的模型而非每次從頭開始訓練
-    - 工程師可以減少計算時間以及生產成本
-    - 數十種模型架構、兩千多個預訓練模型、100多種語言支援
-
-1. 對於模型生命週期的每一個部分都面面俱到：
-    - 訓練先進的模型，只需 3 行程式碼
-    - 模型可以在不同深度學習框架之間任意轉換
-    - 為訓練、評估和生產選擇最適合的框架，並完美銜接
-
-1. 為你的需求輕鬆客製化專屬模型和範例：
-    - 我們為每種模型架構提供了多個範例來重現原論文結果
-    - 一致的模型內部架構
-    - 模型檔案可單獨使用，便於修改和快速實驗
-
-## 什麼情況下我不該用 transformers？
-
-- 本函式庫並不是模組化的神經網絡工具箱。模型文件中的程式碼並未做額外的抽象封裝，以便研究人員快速地翻閱及修改程式碼，而不會深陷複雜的類別包裝之中。
-- `Trainer` API 並非相容任何模型，它只為本函式庫中的模型最佳化。對於一般的機器學習用途，請使用其他函式庫。
-- 儘管我們已盡力而為，[examples 目錄](https://github.com/huggingface/transformers/tree/main/examples)中的腳本也僅為範例而已。對於特定問題，它們並不一定隨選即用，可能需要修改幾行程式碼以符合需求。
-
-## 安裝
-
-### 使用 pip
-
-這個 Repository 已在 Python 3.6+、Flax 0.3.2+、PyTorch 1.3.1+ 和 TensorFlow 2.3+ 下經過測試。
-
-你可以在[虛擬環境](https://docs.python.org/3/library/venv.html)中安裝 🤗 Transformers。如果你還不熟悉 Python 的虛擬環境，請閱此[使用者指引](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)。
-
-首先，用你打算使用的版本的 Python 創建一個虛擬環境並進入。
-
-然後，你需要安裝 Flax、PyTorch 或 TensorFlow 其中之一。對於該如何在你使用的平台上安裝這些框架，請參閱 [TensorFlow 安裝頁面](https://www.tensorflow.org/install/), [PyTorch 安裝頁面](https://pytorch.org/get-started/locally/#start-locally) 或 [Flax 安裝頁面](https://github.com/google/flax#quick-install)。
-
-當其中一個後端安裝成功後，🤗 Transformers 可依此安裝：
-
-```bash
-pip install transformers
-```
-
-如果你想要試試範例或者想在正式發布前使用最新開發中的程式碼，你必須[從原始碼安裝](https://huggingface.co/docs/transformers/installation#installing-from-source)。
-
-### 使用 conda
-
-自 Transformers 4.0.0 版始，我們有了一個 conda channel： `huggingface`。
-
-🤗 Transformers 可以藉由 conda 依此安裝：
-
-```shell script
-conda install -c huggingface transformers
-```
-
-要藉由 conda 安裝 Flax、PyTorch 或 TensorFlow 其中之一，請參閱它們各自安裝頁面的說明。
-
-## 模型架構
-
-**🤗 Transformers 支援的[所有的模型檢查點](https://huggingface.co/models)**，由[使用者](https://huggingface.co/users)和[組織](https://huggingface.co/organizations)上傳，均與 huggingface.co [model hub](https://huggingface.co) 完美結合。
-
-目前的檢查點數量： ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 Transformers 目前支援以下的架構（模型概覽請參閱[這裡](https://huggingface.co/docs/transformers/model_summary)）：
-
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/distillation) and a German version of DistilBERT.
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER) released with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2** was released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released with the paper [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](https://huggingface.co/docs/transformers/main/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook) released with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University) released with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released with the paper [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft) released with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI) released with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-1. 想要貢獻新的模型？我們這裡有一份**詳細指引和模板**來引導你加入新的模型。你可以在 [`templates`](./templates) 目錄中找到它們。記得查看[貢獻指引](./CONTRIBUTING.md)並在開始寫 PR 前聯繫維護人員或開一個新的 issue 來獲得 feedbacks。
-
-要檢查某個模型是否已有 Flax、PyTorch 或 TensorFlow 的實作，或其是否在🤗 Tokenizers 函式庫中有對應的 tokenizer，敬請參閱[此表](https://huggingface.co/docs/transformers/index#supported-frameworks)。
-
-這些實作均已於多個資料集測試（請參閱範例腳本）並應與原版實作表現相當。你可以在範例文件的[此節](https://huggingface.co/docs/transformers/examples)中了解實作的細節。
-
-
-## 了解更多
-
-| 章節 | 描述 |
-|-|-|
-| [文件](https://huggingface.co/transformers/) | 完整的 API 文件和教學 |
-| [任務概覽](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers 支援的任務 |
-| [預處理教學](https://huggingface.co/docs/transformers/preprocessing) | 使用 `Tokenizer` 來為模型準備資料 |
-| [訓練和微調](https://huggingface.co/docs/transformers/training) | 使用 PyTorch/TensorFlow 的內建的訓練方式或於 `Trainer` API 中使用 🤗 Transformers 提供的模型 |
-| [快速上手：微調和範例腳本](https://github.com/huggingface/transformers/tree/main/examples) | 為各種任務提供的範例腳本 |
-| [模型分享和上傳](https://huggingface.co/docs/transformers/model_sharing) | 上傳並與社群分享你微調的模型 |
-| [遷移](https://huggingface.co/docs/transformers/migration) | 從 `pytorch-transformers` 或 `pytorch-pretrained-bert` 遷移到 🤗 Transformers |
-
-## 引用
-
-我們已將此函式庫的[論文](https://www.aclweb.org/anthology/2020.emnlp-demos.6/)正式發表。如果你使用了 🤗 Transformers 函式庫，可以引用：
-```bibtex
-@inproceedings{wolf-etal-2020-transformers,
-    title = "Transformers: State-of-the-Art Natural Language Processing",
-    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
-    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
-    month = oct,
-    year = "2020",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
-    pages = "38--45"
-}
-```

docker/transformers-all-latest-gpu/Dockerfile

@@ -1,56 +0,0 @@
-FROM nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-# Use login shell to read variables from `~/.profile` (to pass dynamic created variables between RUN commands)
-SHELL ["sh", "-lc"]
-
-# The following `ARG` are mainly used to specify the versions explicitly & directly in this docker file, and not meant
-# to be used as arguments for docker build (so far).
-
-ARG PYTORCH='1.12.1'
-# (not always a valid torch version)
-ARG INTEL_TORCH_EXT='1.11.0'
-# Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu113'
-
-RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs
-RUN git lfs install
-RUN python3 -m pip install --no-cache-dir --upgrade pip
-
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev,onnxruntime]
-
-# TODO: Handle these in a python utility script
-RUN [ ${#PYTORCH} -gt 0 -a "$PYTORCH" != "pre" ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; echo "export VERSION='$VERSION'" >> ~/.profile
-RUN echo torch=$VERSION
-# `torchvision` and `torchaudio` should be installed along with `torch`, especially for nightly build.
-# Currently, let's just use their latest releases (when `torch` is installed with a release version)
-# TODO: We might need to specify proper versions that work with a specific torch version (especially for past CI).
-RUN [ "$PYTORCH" != "pre" ] && python3 -m pip install --no-cache-dir -U $VERSION torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA || python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/$CUDA
-
-RUN python3 -m pip install --no-cache-dir -U tensorflow
-RUN python3 -m pip install --no-cache-dir -U tensorflow_probability
-RUN python3 -m pip uninstall -y flax jax
-
-# Use installed torch version for `torch-scatter` to avid to deal with PYTORCH='pre'.
-# If torch is nightly version, the link is likely to be invalid, but the installation falls back to the latest torch-scatter
-RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+$CUDA.html
-RUN python3 -m pip install --no-cache-dir intel_extension_for_pytorch==$INTEL_TORCH_EXT+cpu -f https://software.intel.com/ipex-whl-stable
-
-RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
-RUN python3 -m pip install -U "itsdangerous<2.1.0"
-
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-# Add bitsandbytes for mixed int8 testing
-RUN python3 -m pip install --no-cache-dir bitsandbytes
-
-RUN python3 -m pip install --no-cache-dir decord
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop

docker/transformers-cpu/Dockerfile

@@ -23,4 +23,4 @@ COPY . transformers/
 RUN cd transformers/ && \
     python3 -m pip install --no-cache-dir .
 
-CMD ["/bin/bash"]
+CMD ["/bin/bash"]

docker/transformers-doc-builder/Dockerfile

@@ -1,20 +0,0 @@
-FROM python:3.8
-LABEL maintainer="Hugging Face"
-
-RUN apt update
-RUN git clone https://github.com/huggingface/transformers
-
-RUN python3 -m pip install --no-cache-dir --upgrade pip && python3 -m pip install --no-cache-dir git+https://github.com/huggingface/doc-builder ./transformers[dev]
-RUN apt-get -y update && apt-get install -y libsndfile1-dev && apt install -y tesseract-ocr
-
-# Torch needs to be installed before deepspeed
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed]
-
-RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python -c "from torch import version; print(version.__version__.split('+')[0])")+cpu.html
-RUN python3 -m pip install --no-cache-dir torchvision git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
-RUN python3 -m pip install --no-cache-dir pytorch-quantization --extra-index-url https://pypi.ngc.nvidia.com
-RUN python3 -m pip install -U "itsdangerous<2.1.0"
-
-# Test if the image could successfully build the doc. before publishing the image
-RUN doc-builder build transformers transformers/docs/source/en --build_dir doc-build-dev --notebook_dir notebooks/transformers_doc --clean
-RUN rm -rf doc-build-dev

docker/transformers-past-gpu/Dockerfile

@@ -1,43 +0,0 @@
-ARG BASE_DOCKER_IMAGE="nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04"
-FROM $BASE_DOCKER_IMAGE
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-# Use login shell to read variables from `~/.profile` (to pass dynamic created variables between RUN commands)
-SHELL ["sh", "-lc"]
-
-RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs
-RUN git lfs install
-RUN python3 -m pip install --no-cache-dir --upgrade pip
-
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev,onnxruntime]
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
-
-ARG FRAMEWORK
-ARG VERSION
-
-# Remove all frameworks
-# (`accelerate` requires `torch`, and this causes import issues for TF-only testing)
-RUN python3 -m pip uninstall -y torch torchvision torchaudio accelerate tensorflow jax flax
-
-# Get the libraries and their versions to install, and write installation command to `~/.profile`.
-RUN python3 ./transformers/utils/past_ci_versions.py --framework $FRAMEWORK --version $VERSION
-
-# Install the target framework
-RUN echo "INSTALL_CMD = $INSTALL_CMD"
-RUN $INSTALL_CMD
-
-# Having installation problems for torch-scatter with torch <= 1.6. Disable so we have the same set of tests.
-# (This part will be removed once the logic of using `past_ci_versions.py` is used in other Dockerfile files.)
-# # Use installed torch version for `torch-scatter`.
-# # (The env. variable $CUDA is defined in `past_ci_versions.py`)
-# RUN [ "$FRAMEWORK" = "pytorch" ] && python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+$CUDA.html || echo "torch-scatter not to be installed"
-
-RUN python3 -m pip install -U "itsdangerous<2.1.0"

docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile

@@ -1,35 +0,0 @@
-FROM nvcr.io/nvidia/pytorch:21.03-py3
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-ARG PYTORCH='1.12.1'
-# Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu113'
-
-RUN apt -y update
-RUN apt install -y libaio-dev
-RUN python3 -m pip install --no-cache-dir --upgrade pip
-
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-
-# Install latest release PyTorch
-# (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)
-# (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
-RUN python3 -m pip install --no-cache-dir -U torch==$PYTORCH torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA
-
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
-
-# Pre-build **latest** DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
-RUN python3 -m pip uninstall -y deepspeed
-# This has to be run (again) inside the GPU VMs running the tests.
-# The installation works here, but some tests fail, if we don't pre-build deepspeed again in the VMs running the tests.
-# TODO: Find out why test fail.
-RUN DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
-
-RUN python3 -c "from deepspeed.launcher.runner import main"

docker/transformers-pytorch-deepspeed-nightly-gpu/Dockerfile

@@ -1,53 +0,0 @@
-FROM nvcr.io/nvidia/pytorch:21.03-py3
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-# Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu113'
-
-RUN apt -y update
-RUN apt install -y libaio-dev
-RUN python3 -m pip install --no-cache-dir --upgrade pip
-
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-
-# Install **nightly** release PyTorch (flag `--pre`)
-# (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)
-# (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
-RUN python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/$CUDA
-
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
-
-# Pre-build **nightly** release of DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
-RUN python3 -m pip uninstall -y deepspeed
-# This has to be run inside the GPU VMs running the tests. (So far, it fails here due to GPU checks during compilation.)
-# Issue: https://github.com/microsoft/DeepSpeed/issues/2010
-# RUN git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build && \
-#    DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
-
-# For `torchdynamo` tests
-# (see https://github.com/huggingface/transformers/pull/17765)
-RUN git clone https://github.com/pytorch/functorch
-RUN python3 -m pip install --no-cache-dir ./functorch[aot]
-RUN cd functorch && python3 setup.py develop
-
-RUN git clone https://github.com/pytorch/torchdynamo
-RUN python3 -m pip install -r ./torchdynamo/requirements.txt
-RUN cd torchdynamo && python3 setup.py develop
-
-# install TensorRT
-RUN python3 -m pip install --no-cache-dir -U nvidia-pyindex
-RUN python3 -m pip install --no-cache-dir -U nvidia-tensorrt==8.2.4.2
-
-# install torch_tensorrt (fx path)
-RUN git clone https://github.com/pytorch/TensorRT.git
-RUN cd TensorRT/py && python3 setup.py install --fx-only
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
-
-# Disable for now as deepspeed is not installed above. To be enabled once the issue is fixed.
-# RUN python3 -c "from deepspeed.launcher.runner import main"

docker/transformers-pytorch-gpu/Dockerfile

@@ -1,31 +1,30 @@
-FROM nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04
+FROM nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04
 LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
 
-ARG DEBIAN_FRONTEND=noninteractive
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
 
-RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg
-RUN python3 -m pip install --no-cache-dir --upgrade pip
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    mkl \
+    torch
 
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing]
+RUN git clone https://github.com/NVIDIA/apex
+RUN cd apex && \
+    python3 setup.py install && \
+    pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
 
-# If set to nothing, will install the latest version
-ARG PYTORCH='1.12.1'
-ARG TORCH_VISION=''
-ARG TORCH_AUDIO=''
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
 
-RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
-RUN [ ${#TORCH_VISION} -gt 0 ] && VERSION='torchvision=='TORCH_VISION'.*' ||  VERSION='torchvision'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
-RUN [ ${#TORCH_AUDIO} -gt 0 ] && VERSION='torchaudio=='TORCH_AUDIO'.*' ||  VERSION='torchaudio'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
-
-RUN python3 -m pip uninstall -y tensorflow flax
-
-RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cu113.html
-RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
-RUN python3 -m pip install -U "itsdangerous<2.1.0"
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
+CMD ["/bin/bash"]

docker/transformers-pytorch-tpu/Dockerfile

@@ -1,7 +1,7 @@
 FROM google/cloud-sdk:slim
 
 # Build args.
-ARG GITHUB_REF=refs/heads/main
+ARG GITHUB_REF=refs/heads/master
 
 # TODO: This Dockerfile installs pytorch/xla 3.6 wheels. There are also 3.7
 # wheels available; see below.

docker/transformers-tensorflow-cpu/Dockerfile

@@ -22,4 +22,4 @@ COPY . transformers/
 RUN cd transformers/ && \
     python3 -m pip install --no-cache-dir .
 
-CMD ["/bin/bash"]
+CMD ["/bin/bash"]

docker/transformers-tensorflow-gpu/Dockerfile

@@ -1,23 +1,25 @@
-FROM nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04
+FROM nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04
 LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
 
-ARG DEBIAN_FRONTEND=noninteractive
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
 
-RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg
-RUN python3 -m pip install --no-cache-dir --upgrade pip
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    mkl \
+    tensorflow
 
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-tensorflow,testing]
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
 
-# If set to nothing, will install the latest version
-ARG TENSORFLOW=''
-
-RUN [ ${#TENSORFLOW} -gt 0 ] && VERSION='tensorflow=='$TENSORFLOW'.*' ||  VERSION='tensorflow'; python3 -m pip install --no-cache-dir -U $VERSION
-RUN python3 -m pip uninstall -y torch flax
-RUN python3 -m pip install -U "itsdangerous<2.1.0"
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
+CMD ["/bin/bash"]

docs/Makefile

@@ -0,0 +1,19 @@
+# Minimal makefile for Sphinx documentation
+#
+
+# You can set these variables from the command line.
+SPHINXOPTS    =
+SPHINXBUILD   = sphinx-build
+SOURCEDIR     = source
+BUILDDIR      = _build
+
+# Put it first so that "make" without argument is like "make help".
+help:
+	@$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+
+.PHONY: help Makefile
+
+# Catch-all target: route all unknown targets to Sphinx using the new
+# "make mode" option.  $(O) is meant as a shortcut for $(SPHINXOPTS).
+%: Makefile
+	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)

docs/README.md

@@ -16,130 +16,114 @@ limitations under the License.
 
 # Generating the documentation
 
-To generate the documentation, you first have to build it. Several packages are necessary to build the doc, 
+To generate the documentation, you first have to build it. Several packages are necessary to build the doc,
 you can install them with the following command, at the root of the code repository:
 
 ```bash
 pip install -e ".[docs]"
 ```
 
-Then you need to install our special tool that builds the documentation:
-
-```bash
-pip install git+https://github.com/huggingface/doc-builder
-```
-
 ---
 **NOTE**
 
 You only need to generate the documentation to inspect it locally (if you're planning changes and want to
-check how they look before committing for instance). You don't have to commit the built documentation.
+check how they look like before committing for instance). You don't have to commit the built documentation.
 
 ---
 
+## Packages installed
+
+Here's an overview of all the packages installed. If you ran the previous command installing all packages from
+`requirements.txt`, you do not need to run the following commands.
+
+Building it requires the package `sphinx` that you can
+install using:
+
+```bash
+pip install -U sphinx
+```
+
+You would also need the custom installed [theme](https://github.com/readthedocs/sphinx_rtd_theme) by
+[Read The Docs](https://readthedocs.org/). You can install it using the following command:
+
+```bash
+pip install sphinx_rtd_theme
+```
+
+The third necessary package is the `recommonmark` package to accept Markdown as well as Restructured text:
+
+```bash
+pip install recommonmark
+```
+
 ## Building the documentation
 
-Once you have setup the `doc-builder` and additional packages, you can generate the documentation by 
-typing the following command:
+Once you have setup `sphinx`, you can build the documentation by running the following command in the `/docs` folder:
 
 ```bash
-doc-builder build transformers docs/source/en/ --build_dir ~/tmp/test-build
+make html
 ```
 
-You can adapt the `--build_dir` to set any temporary folder that you prefer. This command will create it and generate
-the MDX files that will be rendered as the documentation on the main website. You can inspect them in your favorite
-Markdown editor.
-
-## Previewing the documentation
-
-To preview the docs, first install the `watchdog` module with:
-
-```bash
-pip install watchdog
-```
-
-Then run the following command:
-
-```bash
-doc-builder preview {package_name} {path_to_docs}
-```
-
-For example:
-
-```bash
-doc-builder preview transformers docs/source/en/
-```
-
-The docs will be viewable at [http://localhost:3000](http://localhost:3000). You can also preview the docs once you have opened a PR. You will see a bot add a comment to a link where the documentation with your changes lives.
+A folder called ``_build/html`` should have been created. You can now open the file ``_build/html/index.html`` in your
+browser.
 
 ---
 **NOTE**
 
-The `preview` command only works with existing doc files. When you add a completely new file, you need to update `_toctree.yml` & restart `preview` command (`ctrl-c` to stop it & call `doc-builder preview ...` again).
+If you are adding/removing elements from the toc-tree or from any structural item, it is recommended to clean the build
+directory before rebuilding. Run the following command to clean and build:
+
+```bash
+make clean && make html
+```
 
 ---
 
-## Adding a new element to the navigation bar
+It should build the static app that will be available under `/docs/_build/html`
 
-Accepted files are Markdown (.md or .mdx).
+## Adding a new element to the tree (toc-tree)
 
-Create a file with its extension and put it in the source directory. You can then link it to the toc-tree by putting
-the filename without the extension in the [`_toctree.yml`](https://github.com/huggingface/transformers/blob/main/docs/source/_toctree.yml) file.
+Accepted files are reStructuredText (.rst) and Markdown (.md). Create a file with its extension and put it
+in the source directory. You can then link it to the toc-tree by putting the filename without the extension.
 
-## Renaming section headers and moving sections
+## Preview the documentation in a pull request
 
-It helps to keep the old links working when renaming the section header and/or moving sections from one document to another. This is because the old links are likely to be used in Issues, Forums, and Social media and it'd make for a much more superior user experience if users reading those months later could still easily navigate to the originally intended information.
-
-Therefore we simply keep a little map of moved sections at the end of the document where the original section was. The key is to preserve the original anchor.
-
-So if you renamed a section from: "Section A" to "Section B", then you can add at the end of the file:
-
-```
-Sections that were moved:
-
-[ <a href="#section-b">Section A</a><a id="section-a"></a> ]
-```
-and of course, if you moved it to another file, then:
-
-```
-Sections that were moved:
-
-[ <a href="../new-file#section-b">Section A</a><a id="section-a"></a> ]
-```
-
-Use the relative style to link to the new file so that the versioned docs continue to work.
-
-For an example of a rich moved section set please see the very end of [the Trainer doc](https://github.com/huggingface/transformers/blob/main/docs/source/en/main_classes/trainer.mdx).
+Once you have made your pull request, you can check what the documentation will look like after it's merged by
+following these steps:
 
+- Look at the checks at the bottom of the conversation page of your PR (you may need to click on "show all checks" to
+  expand them).
+- Click on "details" next to the `ci/circleci: build_doc` check.
+- In the new window, click on the "Artifacts" tab.
+- Locate the file "docs/_build/html/index.html" (or any specific page you want to check) and click on it to get a
+  preview.
 
 ## Writing Documentation - Specification
 
 The `huggingface/transformers` documentation follows the
-[Google documentation](https://sphinxcontrib-napoleon.readthedocs.io/en/latest/example_google.html) style for docstrings,
-although we can write them directly in Markdown.
+[Google documentation](https://sphinxcontrib-napoleon.readthedocs.io/en/latest/example_google.html) style. It is
+mostly written in ReStructuredText
+([Sphinx simple documentation](https://www.sphinx-doc.org/en/master/usage/restructuredtext/index.html),
+[Sourceforge complete documentation](https://docutils.sourceforge.io/docs/ref/rst/restructuredtext.html)).
+
 
 ### Adding a new tutorial
 
 Adding a new tutorial or section is done in two steps:
 
 - Add a new file under `./source`. This file can either be ReStructuredText (.rst) or Markdown (.md).
-- Link that file in `./source/_toctree.yml` on the correct toc-tree.
+- Link that file in `./source/index.rst` on the correct toc-tree.
 
 Make sure to put your new file under the proper section. It's unlikely to go in the first section (*Get Started*), so
-depending on the intended targets (beginners, more advanced users, or researchers) it should go in sections two, three, or
+depending on the intended targets (beginners, more advanced users or researchers) it should go in section two, three or
 four.
 
-### Translating
-
-When translating, refer to the guide at [./TRANSLATING.md](https://github.com/huggingface/transformers/blob/main/docs/TRANSLATING.md).
-
-
 ### Adding a new model
 
 When adding a new model:
 
-- Create a file `xxx.mdx` or under `./source/model_doc` (don't hesitate to copy an existing file as template).
-- Link that file in `./source/_toctree.yml`.
+- Create a file `xxx.rst` under `./source/model_doc` (don't hesitate to copy an existing file as template).
+- Link that file in `./source/index.rst` on the `model_doc` toc-tree.
 - Write a short overview of the model:
     - Overview with paper & authors
     - Paper abstract
@@ -153,82 +137,64 @@ When adding a new model:
     - PyTorch head models
     - TensorFlow base model
     - TensorFlow head models
-    - Flax base model
-    - Flax head models
-
-These classes should be added using our Markdown syntax. Usually as follows:
 
+These classes should be added using the RST syntax. Usually as follows:
 ```
-## XXXConfig
+XXXConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-[[autodoc]] XXXConfig
+.. autoclass:: transformers.XXXConfig
+    :members:
 ```
 
 This will include every public method of the configuration that is documented. If for some reason you wish for a method
 not to be displayed in the documentation, you can do so by specifying which methods should be in the docs:
 
 ```
-## XXXTokenizer
+XXXTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-[[autodoc]] XXXTokenizer
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - save_vocabulary
-```
+.. autoclass:: transformers.XXXTokenizer
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
 
-If you just want to add a method that is not documented (for instance magic methods like `__call__` are not documented
-by default) you can put the list of methods to add in a list that contains `all`:
-
-```
-## XXXTokenizer
-
-[[autodoc]] XXXTokenizer
-    - all
-    - __call__
 ```
 
 ### Writing source documentation
 
-Values that should be put in `code` should either be surrounded by backticks: \`like so\`. Note that argument names
-and objects like True, None, or any strings should usually be put in `code`.
+Values that should be put in `code` should either be surrounded by double backticks: \`\`like so\`\` or be written as
+an object using the :obj: syntax: :obj:\`like so\`. Note that argument names and objects like True, None or any strings
+should usually be put in `code`.
 
-When mentioning a class, function, or method, it is recommended to use our syntax for internal links so that our tool
-adds a link to its documentation with this syntax: \[\`XXXClass\`\] or \[\`function\`\]. This requires the class or 
-function to be in the main package.
+When mentionning a class, it is recommended to use the :class: syntax as the mentioned class will be automatically
+linked by Sphinx: :class:\`~transformers.XXXClass\`
 
-If you want to create a link to some internal class or function, you need to
-provide its path. For instance: \[\`utils.ModelOutput\`\]. This will be converted into a link with
-`utils.ModelOutput` in the description. To get rid of the path and only keep the name of the object you are
-linking to in the description, add a ~: \[\`~utils.ModelOutput\`\] will generate a link with `ModelOutput` in the description.
+When mentioning a function, it is recommended to use the :func: syntax as the mentioned function will be automatically
+linked by Sphinx: :func:\`~transformers.function\`.
 
-The same works for methods so you can either use \[\`XXXClass.method\`\] or \[~\`XXXClass.method\`\].
+When mentioning a method, it is recommended to use the :meth: syntax as the mentioned method will be automatically
+linked by Sphinx: :meth:\`~transformers.XXXClass.method\`.
+
+Links should be done as so (note the double underscore at the end): \`text for the link <./local-link-or-global-link#loc>\`__
 
 #### Defining arguments in a method
 
-Arguments should be defined with the `Args:` (or `Arguments:` or `Parameters:`) prefix, followed by a line return and
-an indentation. The argument should be followed by its type, with its shape if it is a tensor, a colon, and its
-description:
-
-```
-    Args:
-        n_layers (`int`): The number of layers of the model.
-```
-
-If the description is too long to fit in one line, another indentation is necessary before writing the description
-after the argument.
+Arguments should be defined with the `Args:` prefix, followed by a line return and an indentation.
+The argument should be followed by its type, with its shape if it is a tensor, and a line return.
+Another indentation is necessary before writing the description of the argument.
 
 Here's an example showcasing everything so far:
 
 ```
     Args:
-        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`):
             Indices of input sequence tokens in the vocabulary.
 
-            Indices can be obtained using [`AlbertTokenizer`]. See [`~PreTrainedTokenizer.encode`] and
-            [`~PreTrainedTokenizer.__call__`] for details.
+            Indices can be obtained using :class:`~transformers.AlbertTokenizer`.
+            See :meth:`~transformers.PreTrainedTokenizer.encode` and
+            :meth:`~transformers.PreTrainedTokenizer.__call__` for details.
 
-            [What are input IDs?](../glossary#input-ids)
+            `What are input IDs? <../glossary.html#input-ids>`__
 ```
 
 For optional arguments or arguments with defaults we follow the following syntax: imagine we have a function with the
@@ -242,190 +208,93 @@ then its documentation should look like this:
 
 ```
     Args:
-        x (`str`, *optional*):
+        x (:obj:`str`, `optional`):
             This argument controls ...
-        a (`float`, *optional*, defaults to 1):
+        a (:obj:`float`, `optional`, defaults to 1):
             This argument is used to ...
 ```
 
-Note that we always omit the "defaults to \`None\`" when None is the default for any argument. Also note that even
+Note that we always omit the "defaults to :obj:\`None\`" when None is the default for any argument. Also note that even
 if the first line describing your argument type and its default gets long, you can't break it on several lines. You can
 however write as many lines as you want in the indented description (see the example above with `input_ids`).
 
 #### Writing a multi-line code block
 
-Multi-line code blocks can be useful for displaying examples. They are done between two lines of three backticks as usual in Markdown:
+Multi-line code blocks can be useful for displaying examples. They are done like so:
 
+```
+Example::
 
-````
+    # first line of code
+    # second line
+    # etc
 ```
-# first line of code
-# second line
-# etc
-```
-````
+
+The `Example` string at the beginning can be replaced by anything as long as there are two semicolons following it.
 
 We follow the [doctest](https://docs.python.org/3/library/doctest.html) syntax for the examples to automatically test
-the results to stay consistent with the library.
+the results stay consistent with the library.
 
 #### Writing a return block
 
-The return block should be introduced with the `Returns:` prefix, followed by a line return and an indentation.
+Arguments should be defined with the `Args:` prefix, followed by a line return and an indentation.
 The first line should be the type of the return, followed by a line return. No need to indent further for the elements
 building the return.
 
-Here's an example of a single value return:
+Here's an example for tuple return, comprising several objects:
 
 ```
     Returns:
-        `List[int]`: A list of integers in the range [0, 1] --- 1 for a special token, 0 for a sequence token.
+        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.BertConfig`) and inputs:
+        loss (`optional`, returned when ``masked_lm_labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
+            Total loss as the sum of the masked language modeling loss and the next sequence prediction (classification) loss.
+        prediction_scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`)
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
 ```
 
-Here's an example of a tuple return, comprising several objects:
+Here's an example for a single value return:
 
 ```
     Returns:
-        `tuple(torch.FloatTensor)` comprising various elements depending on the configuration ([`BertConfig`]) and inputs:
-        - ** loss** (*optional*, returned when `masked_lm_labels` is provided) `torch.FloatTensor` of shape `(1,)` --
-          Total loss is the sum of the masked language modeling loss and the next sequence prediction (classification) loss.
-        - **prediction_scores** (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`) --
-          Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        :obj:`List[int]`: A list of integers in the range [0, 1] --- 1 for a special token, 0 for a sequence token.
 ```
 
-#### Adding an image
+#### Adding a new section
 
-Due to the rapidly growing repository, it is important to make sure that no files that would significantly weigh down the repository are added. This includes images, videos, and other non-text files. We prefer to leverage a hf.co hosted `dataset` like
-the ones hosted on [`hf-internal-testing`](https://huggingface.co/hf-internal-testing) in which to place these files and reference
-them by URL. We recommend putting them in the following dataset: [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images).
-If an external contribution, feel free to add the images to your PR and ask a Hugging Face member to migrate your images
-to this dataset.
-
-## Styling the docstring
-
-We have an automatic script running with the `make style` comment that will make sure that:
-- the docstrings fully take advantage of the line width
-- all code examples are formatted using black, like the code of the Transformers library
-
-This script may have some weird failures if you made a syntax mistake or if you uncover a bug. Therefore, it's
-recommended to commit your changes before running `make style`, so you can revert the changes done by that script
-easily.
-
-# Testing documentation examples
-
-Good documentation often comes with an example of how a specific function or class should be used. 
-Each model class should contain at least one example showcasing
-how to use this model class in inference. *E.g.* the class [Wav2Vec2ForCTC](https://huggingface.co/docs/transformers/model_doc/wav2vec2#transformers.Wav2Vec2ForCTC) 
-includes an example of how to transcribe speech to text in the 
-[docstring of its forward function](https://huggingface.co/docs/transformers/model_doc/wav2vec2#transformers.Wav2Vec2ForCTC.forward).
-
-## Writing documentation examples
-
-The syntax for Example docstrings can look as follows:
+In ReST section headers are designated as such with the help of a line of underlying characters, e.g.,:
 
 ```
-    Example:
+Section 1
+^^^^^^^^^^^^^^^^^^
 
-    ```python
-    >>> from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
-    >>> from datasets import load_dataset
-    >>> import torch
-
-    >>> dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
-    >>> dataset = dataset.sort("id")
-    >>> sampling_rate = dataset.features["audio"].sampling_rate
-
-    >>> processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
-    >>> model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
-
-    >>> # audio file is decoded on the fly
-    >>> inputs = processor(dataset[0]["audio"]["array"], sampling_rate=sampling_rate, return_tensors="pt")
-    >>> with torch.no_grad():
-    ...     logits = model(**inputs).logits
-    >>> predicted_ids = torch.argmax(logits, dim=-1)
-
-    >>> # transcribe speech
-    >>> transcription = processor.batch_decode(predicted_ids)
-    >>> transcription[0]
-    'MISTER QUILTER IS THE APOSTLE OF THE MIDDLE CLASSES AND WE ARE GLAD TO WELCOME HIS GOSPEL'
-    ```
+Sub-section 1
+~~~~~~~~~~~~~~~~~~
 ```
 
-The docstring should give a minimal, clear example of how the respective model 
-is to be used in inference and also include the expected (ideally sensible)
-output.
-Often, readers will try out the example before even going through the function 
-or class definitions. Therefore it is of utmost importance that the example 
-works as expected.
+ReST allows the use of any characters to designate different section levels, as long as they are used consistently within the same document. For details see [sections doc](https://www.sphinx-doc.org/en/master/usage/restructuredtext/basics.html#sections). Because there is no standard different documents often end up using different characters for the same levels which makes it very difficult to know which character to use when creating a new section.
 
-## Docstring testing
+Specifically, if when running `make docs` you get an error like:
+```
+docs/source/main_classes/trainer.rst:127:Title level inconsistent:
+```
+you picked an inconsistent character for some of the levels.
 
-To do so each example should be included in the doctests. 
-We use pytests' [doctest integration](https://docs.pytest.org/doctest.html) to verify that all of our examples run correctly. 
-For Transformers, the doctests are run on a daily basis via GitHub Actions as can be 
-seen [here](https://github.com/huggingface/transformers/actions/workflows/doctests.yml).
+But how do you know which characters you must use for an already existing level or when adding a new level?
 
-To include your example in the daily doctests, you need to add the filename that
-contains the example docstring to the [documentation_tests.txt](../utils/documentation_tests.txt).
-
-### For Python files
-
-You will first need to run the following command (from the root of the repository) to prepare the doc file (doc-testing needs to add additional lines that we don't include in the doc source files):
-
-```bash
-python utils/prepare_for_doc_test.py src docs
+You can use this helper script:
+```
+perl -ne '/^(.)\1{100,}/ && do { $h{$1}=++$c if !$h{$1} }; END { %h = reverse %h ; print "$_ $h{$_}\n" for sort keys %h}' docs/source/main_classes/trainer.rst
+1 -
+2 ~
+3 ^
+4 =
+5 "
 ```
 
-If you work on a specific python module, say `modeling_wav2vec2.py`, you can run the command as follows (to avoid the unnecessary temporary changes in irrelevant files):
+This tells you which characters have already been assigned for each level.
 
-```bash
-python utils/prepare_for_doc_test.py src/transformers/utils/doc.py src/transformers/models/wav2vec2/modeling_wav2vec2.py
-```
-(`utils/doc.py` should always be included)
+So using this particular example's output -- if your current section's header uses `=` as its underline character, you now know you're at level 4, and if you want to add a sub-section header you know you want `"` as it'd level 5.
 
-Then you can run all the tests in the docstrings of a given file with the following command, here is how we test the modeling file of Wav2Vec2 for instance:
+If you needed to add yet another sub-level, then pick a character that is not used already. That is you must pick a character that is not in the output of that script.
 
-```bash
-pytest --doctest-modules src/transformers/models/wav2vec2/modeling_wav2vec2.py -sv --doctest-continue-on-failure
-```
-
-If you want to isolate a specific docstring, just add `::` after the file name then type the whole path of the function/class/method whose docstring you want to test. For instance, here is how to just test the forward method of `Wav2Vec2ForCTC`:
-
-```bash
-pytest --doctest-modules src/transformers/models/wav2vec2/modeling_wav2vec2.py::transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC.forward -sv --doctest-continue-on-failure
-```
-
-Once you're done, you can run the following command (still from the root of the repository) to undo the changes made by the first command before committing:
-
-```bash
-python utils/prepare_for_doc_test.py src docs --remove_new_line
-```
-
-### For Markdown files
-
-You will first need to run the following command (from the root of the repository) to prepare the doc file (doc-testing needs to add additional lines that we don't include in the doc source files):
-
-```bash
-python utils/prepare_for_doc_test.py src docs
-```
-
-Then you can test locally a given file with this command (here testing the quicktour):
-
-```bash
-pytest --doctest-modules docs/source/quicktour.mdx -sv --doctest-continue-on-failure --doctest-glob="*.mdx"
-```
-
-Once you're done, you can run the following command (still from the root of the repository) to undo the changes made by the first command before committing:
-
-```bash
-python utils/prepare_for_doc_test.py src docs --remove_new_line
-```
-
-### Writing doctests
-
-Here are a few tips to help you debug the doctests and make them pass:
-
-- The outputs of the code need to match the expected output **exactly**, so make sure you have the same outputs. In particular doctest will see a difference between single quotes and double quotes, or a missing parenthesis. The only exceptions to that rule are:
-  * whitespace: one give whitespace (space, tabulation, new line) is equivalent to any number of whitespace, so you can add new lines where there are spaces to make your output more readable.
-  * numerical values: you should never put more than 4 or 5 digits to expected results as different setups or library versions might get you slightly different results. `doctest` is configured to ignore any difference lower than the precision to which you wrote (so 1e-4 if you write 4 digits).
-- Don't leave a block of code that is very long to execute. If you can't make it fast, you can either not use the doctest syntax on it (so that it's ignored), or if you want to use the doctest syntax to show the results, you can add a comment `# doctest: +SKIP` at the end of the lines of code too long to execute
-- Each line of code that produces a result needs to have that result written below. You can ignore an output if you don't want to show it in your code example by adding a comment ` # doctest: +IGNORE_RESULT` at the end of the line of code producing it.
+Here is the full list of characters that can be used in this context: `= - ` : ' " ~ ^ _ * + # < >`

docs/TRANSLATING.md

@@ -1,57 +0,0 @@
-### Translating the Transformers documentation into your language
-
-As part of our mission to democratize machine learning, we'd love to make the Transformers library available in many more languages! Follow the steps below if you want to help translate the documentation into your language 🙏.
-
-**🗞️ Open an issue**
-
-To get started, navigate to the [Issues](https://github.com/huggingface/transformers/issues) page of this repo and check if anyone else has opened an issue for your language. If not, open a new issue by selecting the "Translation template" from the "New issue" button.
-
-Once an issue exists, post a comment to indicate which chapters you'd like to work on, and we'll add your name to the list.
-
-
-**🍴 Fork the repository**
-
-First, you'll need to [fork the Transformers repo](https://docs.github.com/en/get-started/quickstart/fork-a-repo). You can do this by clicking on the **Fork** button on the top-right corner of this repo's page.
-
-Once you've forked the repo, you'll want to get the files on your local machine for editing. You can do that by cloning the fork with Git as follows:
-
-```bash
-git clone https://github.com/YOUR-USERNAME/transformers.git
-```
-
-**📋 Copy-paste the English version with a new language code**
-
-The documentation files are in one leading directory:
-
-- [`docs/source`](https://github.com/huggingface/transformers/tree/main/docs/source): All the documentation materials are organized here by language.
-
-You'll only need to copy the files in the [`docs/source/en`](https://github.com/huggingface/transformers/tree/main/docs/source/en) directory, so first navigate to your fork of the repo and run the following:
-
-```bash
-cd ~/path/to/transformers/docs
-cp -r source/en source/LANG-ID
-```
-
-Here, `LANG-ID` should be one of the ISO 639-1 or ISO 639-2 language codes -- see [here](https://www.loc.gov/standards/iso639-2/php/code_list.php) for a handy table.
-
-**✍️ Start translating**
-
-The fun part comes - translating the text!
-
-The first thing we recommend is translating the part of the `_toctree.yml` file that corresponds to your doc chapter. This file is used to render the table of contents on the website. 
-
-> 🙋 If the `_toctree.yml` file doesn't yet exist for your language, you can create one by copy-pasting from the English version and deleting the sections unrelated to your chapter. Just make sure it exists in the `docs/source/LANG-ID/` directory!
-
-The fields you should add are `local` (with the name of the file containing the translation; e.g. `autoclass_tutorial`), and `title` (with the title of the doc in your language; e.g. `Load pretrained instances with an AutoClass`) -- as a reference, here is the `_toctree.yml` for [English](https://github.com/huggingface/transformers/blob/main/docs/source/en/_toctree.yml):
-
-```yaml
-- sections:
-  - local: pipeline_tutorial # Do not change this! Use the same name for your .md file
-    title: Pipelines for inference # Translate this!
-    ...
-  title: Tutorials # Translate this!
-```
-
-Once you have translated the `_toctree.yml` file, you can start translating the [MDX](https://mdxjs.com/) files associated with your docs chapter.
-
-> 🙋 If you'd like others to help you with the translation, you should [open an issue](https://github.com/huggingface/transformers/issues) and tag @sgugger.

docs/source/_config.py

@@ -1,14 +0,0 @@
-# docstyle-ignore
-INSTALL_CONTENT = """
-# Transformers installation
-! pip install transformers datasets
-# To install from source instead of the last release, comment the command above and uncomment the following one.
-# ! pip install git+https://github.com/huggingface/transformers.git
-"""
-
-notebook_first_cells = [{"type": "code", "content": INSTALL_CONTENT}]
-black_avoid_patterns = {
-    "{processor_class}": "FakeProcessorClass",
-    "{model_class}": "FakeModelClass",
-    "{object_class}": "FakeObjectClass",    
-}

docs/source/_static/css/code-snippets.css

@@ -0,0 +1,16 @@
+
+.highlight .c1, .highlight .sd{
+    color: #999
+}
+
+.highlight .nn, .highlight .k, .highlight .s1, .highlight .nb, .highlight .bp, .highlight .kc {
+    color: #FB8D68;
+}
+
+.highlight .kn, .highlight .nv, .highlight .s2, .highlight .ow {
+    color: #6670FF;
+}
+
+.highlight .gp {
+    color: #FB8D68;
+}

docs/source/_static/css/huggingface.css

@@ -0,0 +1,350 @@
+/* Our DOM objects */
+
+/* Colab dropdown */
+
+table.center-aligned-table td {
+    text-align: center;
+}
+
+table.center-aligned-table th {
+    text-align: center;
+    vertical-align: middle;
+}
+
+.colab-dropdown {
+    position: relative;
+    display: inline-block;
+}
+  
+.colab-dropdown-content {
+    display: none;
+    position: absolute;
+    background-color: #f9f9f9;
+    min-width: 117px;
+    box-shadow: 0px 8px 16px 0px rgba(0,0,0,0.2);
+    z-index: 1;
+}
+  
+.colab-dropdown-content button {
+    color: #6670FF;
+    background-color: #f9f9f9;
+    font-size: 12px;
+    border: none;
+    min-width: 117px;
+    padding: 5px 5px;
+    text-decoration: none;
+    display: block;
+}
+  
+.colab-dropdown-content button:hover {background-color: #eee;}
+  
+.colab-dropdown:hover .colab-dropdown-content {display: block;}
+
+/* Version control */
+
+.version-button {
+    background-color: #6670FF;
+    color: white;
+    border: none;
+    padding: 5px;
+    font-size: 15px;
+    cursor: pointer;
+}
+
+.version-button:hover, .version-button:focus {
+    background-color: #A6B0FF;
+}
+ 
+.version-dropdown {
+    display: none;
+    background-color: #6670FF;
+    min-width: 160px;
+    overflow: auto;
+    font-size: 15px;
+}
+  
+.version-dropdown a {
+    color: white;
+    padding: 3px 4px;
+    text-decoration: none;
+    display: block;
+}
+  
+.version-dropdown a:hover {
+    background-color: #A6B0FF;
+}
+  
+.version-show {
+    display: block;
+}
+
+/* Framework selector */
+
+.framework-selector {
+    display: flex;
+    flex-direction: row;
+    justify-content: flex-end;
+    margin-right: 30px;
+}
+
+.framework-selector > button {
+    background-color: white;
+    color: #6670FF;
+    border: 1px solid #6670FF;
+    padding: 5px;
+}
+
+.framework-selector > button.selected{
+    background-color: #6670FF;
+    color: white;
+    border: 1px solid #6670FF;
+    padding: 5px;
+}
+
+/* Copy button */
+
+a.copybtn {
+    margin: 3px;
+}
+
+/* The literal code blocks */
+.rst-content tt.literal, .rst-content tt.literal, .rst-content code.literal {
+    color: #6670FF;
+}
+
+/* To keep the logo centered */
+.wy-side-scroll {
+    width: auto;
+    font-size: 20px;
+}
+
+/* The div that holds the Hugging Face logo */
+.HuggingFaceDiv {
+    width: 100%
+}
+
+/* The research field on top of the toc tree */
+.wy-side-nav-search{
+    padding-top: 0;
+    background-color: #6670FF;
+}
+
+/* The toc tree */
+.wy-nav-side{
+    background-color: #6670FF;
+}
+
+/* The section headers in the toc tree */
+.wy-menu-vertical p.caption{
+    background-color: #4d59ff;
+    line-height: 40px;
+}
+
+/* The selected items in the toc tree */
+.wy-menu-vertical li.current{
+    background-color: #A6B0FF;
+}
+
+/* When a list item that does belong to the selected block from the toc tree is hovered */
+.wy-menu-vertical li.current a:hover{
+    background-color: #B6C0FF;
+}
+
+/* When a list item that does NOT belong to the selected block from the toc tree is hovered. */
+.wy-menu-vertical li a:hover{
+    background-color: #A7AFFB;
+}
+
+/* The text items on the toc tree */
+.wy-menu-vertical a {
+    color: #FFFFDD;
+    font-family: Calibre-Light, sans-serif;
+}
+.wy-menu-vertical header, .wy-menu-vertical p.caption{
+    color: white;
+    font-family: Calibre-Light, sans-serif;
+}
+
+/* The color inside the selected toc tree block */
+.wy-menu-vertical li.toctree-l2 a, .wy-menu-vertical li.toctree-l3 a, .wy-menu-vertical li.toctree-l4 a {
+    color: black;
+}
+
+/* Inside the depth-2 selected toc tree block */
+.wy-menu-vertical li.toctree-l2.current>a {
+    background-color: #B6C0FF
+}
+.wy-menu-vertical li.toctree-l2.current li.toctree-l3>a {
+    background-color: #C6D0FF
+}
+
+/* Inside the depth-3 selected toc tree block */
+.wy-menu-vertical li.toctree-l3.current li.toctree-l4>a{
+    background-color: #D6E0FF
+}
+
+/* Inside code snippets */
+.rst-content dl:not(.docutils) dt{
+    font-size: 15px;
+}
+
+/* Links */
+a {
+    color: #6670FF;
+}
+
+/* Content bars */
+.rst-content dl:not(.docutils) dt {
+    background-color: rgba(251, 141, 104, 0.1);
+    border-right: solid 2px #FB8D68;
+    border-left: solid 2px #FB8D68;
+    color: #FB8D68;
+    font-family: Calibre-Light, sans-serif;
+    border-top: none;
+    font-style: normal !important;
+}
+
+/* Expand button */
+.wy-menu-vertical li.toctree-l2 span.toctree-expand,
+.wy-menu-vertical li.on a span.toctree-expand, .wy-menu-vertical li.current>a span.toctree-expand,
+.wy-menu-vertical li.toctree-l3 span.toctree-expand{
+    color: black;
+}
+
+/* Max window size */
+.wy-nav-content{
+    max-width: 1200px;
+}
+
+/* Mobile header */
+.wy-nav-top{
+    background-color: #6670FF;
+}
+
+
+/* Source spans */
+.rst-content .viewcode-link, .rst-content .viewcode-back{
+    color: #6670FF;
+    font-size: 110%;
+    letter-spacing: 2px;
+    text-transform: uppercase;
+}
+
+/* It would be better for table to be visible without horizontal scrolling */
+.wy-table-responsive table td, .wy-table-responsive table th{
+    white-space: normal;
+}
+
+.footer {
+    margin-top: 20px;
+}
+
+.footer__Social {
+    display: flex;
+    flex-direction: row;
+}
+
+.footer__CustomImage {
+    margin: 2px 5px 0 0;
+}
+
+/* class and method names in doc */
+.rst-content dl:not(.docutils) tt.descname, .rst-content dl:not(.docutils) tt.descclassname, .rst-content dl:not(.docutils) tt.descname, .rst-content dl:not(.docutils) code.descname, .rst-content dl:not(.docutils) tt.descclassname, .rst-content dl:not(.docutils) code.descclassname{
+    font-family: Calibre, sans-serif;
+    font-size: 20px !important;
+}
+
+/* class name in doc*/
+.rst-content dl:not(.docutils) tt.descname, .rst-content dl:not(.docutils) tt.descname, .rst-content dl:not(.docutils) code.descname{
+    margin-right: 10px;
+    font-family: Calibre-Medium, sans-serif;
+}
+
+/* Method and class parameters */
+.sig-param{
+    line-height: 23px;
+}
+
+/* Class introduction "class" string at beginning */
+.rst-content dl:not(.docutils) .property{
+    font-size: 18px;
+    color: black;
+}
+
+
+/* FONTS */
+body{
+    font-family: Calibre, sans-serif;
+    font-size: 16px;
+}
+
+h1 {
+    font-family: Calibre-Thin, sans-serif;
+    font-size: 70px;
+}
+
+h2, .rst-content .toctree-wrapper p.caption, h3, h4, h5, h6, legend{
+    font-family: Calibre-Medium, sans-serif;
+}
+
+@font-face {
+    font-family: Calibre-Medium;
+    src: url(./Calibre-Medium.otf);
+    font-weight:400;
+}
+
+@font-face {
+    font-family: Calibre;
+    src: url(./Calibre-Regular.otf);
+    font-weight:400;
+}
+
+@font-face {
+    font-family: Calibre-Light;
+    src: url(./Calibre-Light.ttf);
+    font-weight:400;
+}
+
+@font-face {
+    font-family: Calibre-Thin;
+    src: url(./Calibre-Thin.otf);
+    font-weight:400;
+}
+
+
+/**
+ * Nav Links to other parts of huggingface.co
+ */
+ div.menu {
+    position: absolute;
+    top: 0;
+    right: 0;
+    padding-top: 20px;
+    padding-right: 20px;
+    z-index: 1000;
+}
+div.menu a {
+    font-size: 14px;
+    letter-spacing: 0.3px;
+    text-transform: uppercase;
+    color: white;
+    -webkit-font-smoothing: antialiased;
+    background: linear-gradient(0deg, #6671ffb8, #9a66ffb8 50%);
+    padding: 10px 16px 6px 16px;
+    border-radius: 3px;
+    margin-left: 12px;
+    position: relative;
+}
+div.menu a:active {
+    top: 1px;
+}
+@media (min-width: 768px) and (max-width: 1750px) {
+    .wy-breadcrumbs {
+        margin-top: 32px;
+    }
+}
+@media (max-width: 768px) {
+    div.menu {
+        display: none;
+    }
+}

docs/source/benchmarks.rst

@@ -0,0 +1,363 @@
+.. 
+    Copyright 2020 The HuggingFace Team. All rights reserved.
+
+    Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+    the License. You may obtain a copy of the License at
+
+        http://www.apache.org/licenses/LICENSE-2.0
+
+    Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+    an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+    specific language governing permissions and limitations under the License.
+
+Benchmarks
+=======================================================================================================================
+
+Let's take a look at how 🤗 Transformer models can be benchmarked, best practices, and already available benchmarks.
+
+A notebook explaining in more detail how to benchmark 🤗 Transformer models can be found :prefix_link:`here
+<notebooks/05-benchmark.ipynb>`.
+
+How to benchmark 🤗 Transformer models
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The classes :class:`~transformers.PyTorchBenchmark` and :class:`~transformers.TensorFlowBenchmark` allow to flexibly
+benchmark 🤗 Transformer models. The benchmark classes allow us to measure the `peak memory usage` and `required time`
+for both `inference` and `training`.
+
+.. note::
+
+  Hereby, `inference` is defined by a single forward pass, and `training` is defined by a single forward pass and
+  backward pass.
+
+The benchmark classes :class:`~transformers.PyTorchBenchmark` and :class:`~transformers.TensorFlowBenchmark` expect an
+object of type :class:`~transformers.PyTorchBenchmarkArguments` and
+:class:`~transformers.TensorFlowBenchmarkArguments`, respectively, for instantiation.
+:class:`~transformers.PyTorchBenchmarkArguments` and :class:`~transformers.TensorFlowBenchmarkArguments` are data
+classes and contain all relevant configurations for their corresponding benchmark class. In the following example, it
+is shown how a BERT model of type `bert-base-cased` can be benchmarked.
+
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> from transformers import PyTorchBenchmark, PyTorchBenchmarkArguments
+
+    >>> args = PyTorchBenchmarkArguments(models=["bert-base-uncased"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512])
+    >>> benchmark = PyTorchBenchmark(args)
+
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TensorFlowBenchmark, TensorFlowBenchmarkArguments
+
+    >>> args = TensorFlowBenchmarkArguments(models=["bert-base-uncased"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512])
+    >>> benchmark = TensorFlowBenchmark(args)
+
+
+Here, three arguments are given to the benchmark argument data classes, namely ``models``, ``batch_sizes``, and
+``sequence_lengths``. The argument ``models`` is required and expects a :obj:`list` of model identifiers from the
+`model hub <https://huggingface.co/models>`__ The :obj:`list` arguments ``batch_sizes`` and ``sequence_lengths`` define
+the size of the ``input_ids`` on which the model is benchmarked. There are many more parameters that can be configured
+via the benchmark argument data classes. For more detail on these one can either directly consult the files
+``src/transformers/benchmark/benchmark_args_utils.py``, ``src/transformers/benchmark/benchmark_args.py`` (for PyTorch)
+and ``src/transformers/benchmark/benchmark_args_tf.py`` (for Tensorflow). Alternatively, running the following shell
+commands from root will print out a descriptive list of all configurable parameters for PyTorch and Tensorflow
+respectively.
+
+.. code-block:: bash
+
+    ## PYTORCH CODE
+    python examples/pytorch/benchmarking/run_benchmark.py --help
+
+    ## TENSORFLOW CODE
+    python examples/tensorflow/benchmarking/run_benchmark_tf.py --help
+
+
+An instantiated benchmark object can then simply be run by calling ``benchmark.run()``.
+
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> results = benchmark.run()
+    >>> print(results)
+    ====================       INFERENCE - SPEED - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length     Time in s                  
+    --------------------------------------------------------------------------------
+    bert-base-uncased          8               8             0.006     
+    bert-base-uncased          8               32            0.006     
+    bert-base-uncased          8              128            0.018     
+    bert-base-uncased          8              512            0.088     
+    --------------------------------------------------------------------------------
+
+    ====================      INFERENCE - MEMORY - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length    Memory in MB 
+    --------------------------------------------------------------------------------
+    bert-base-uncased          8               8             1227
+    bert-base-uncased          8               32            1281
+    bert-base-uncased          8              128            1307
+    bert-base-uncased          8              512            1539
+    --------------------------------------------------------------------------------
+
+    ====================        ENVIRONMENT INFORMATION         ====================
+
+    - transformers_version: 2.11.0
+    - framework: PyTorch
+    - use_torchscript: False
+    - framework_version: 1.4.0
+    - python_version: 3.6.10
+    - system: Linux
+    - cpu: x86_64
+    - architecture: 64bit
+    - date: 2020-06-29
+    - time: 08:58:43.371351
+    - fp16: False
+    - use_multiprocessing: True
+    - only_pretrain_model: False
+    - cpu_ram_mb: 32088
+    - use_gpu: True
+    - num_gpus: 1
+    - gpu: TITAN RTX
+    - gpu_ram_mb: 24217
+    - gpu_power_watts: 280.0
+    - gpu_performance_state: 2
+    - use_tpu: False
+
+    >>> ## TENSORFLOW CODE
+    >>> results = benchmark.run()
+    >>> print(results)
+    ====================       INFERENCE - SPEED - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length     Time in s                  
+    --------------------------------------------------------------------------------
+    bert-base-uncased          8               8             0.005
+    bert-base-uncased          8               32            0.008
+    bert-base-uncased          8              128            0.022
+    bert-base-uncased          8              512            0.105
+    --------------------------------------------------------------------------------
+
+    ====================      INFERENCE - MEMORY - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length    Memory in MB 
+    --------------------------------------------------------------------------------
+    bert-base-uncased          8               8             1330
+    bert-base-uncased          8               32            1330
+    bert-base-uncased          8              128            1330
+    bert-base-uncased          8              512            1770
+    --------------------------------------------------------------------------------
+
+    ====================        ENVIRONMENT INFORMATION         ====================
+
+    - transformers_version: 2.11.0
+    - framework: Tensorflow
+    - use_xla: False
+    - framework_version: 2.2.0
+    - python_version: 3.6.10
+    - system: Linux
+    - cpu: x86_64
+    - architecture: 64bit
+    - date: 2020-06-29
+    - time: 09:26:35.617317
+    - fp16: False
+    - use_multiprocessing: True
+    - only_pretrain_model: False
+    - cpu_ram_mb: 32088
+    - use_gpu: True
+    - num_gpus: 1
+    - gpu: TITAN RTX
+    - gpu_ram_mb: 24217
+    - gpu_power_watts: 280.0
+    - gpu_performance_state: 2
+    - use_tpu: False
+
+By default, the `time` and the `required memory` for `inference` are benchmarked. In the example output above the first
+two sections show the result corresponding to `inference time` and `inference memory`. In addition, all relevant
+information about the computing environment, `e.g.` the GPU type, the system, the library versions, etc... are printed
+out in the third section under `ENVIRONMENT INFORMATION`. This information can optionally be saved in a `.csv` file
+when adding the argument :obj:`save_to_csv=True` to :class:`~transformers.PyTorchBenchmarkArguments` and
+:class:`~transformers.TensorFlowBenchmarkArguments` respectively. In this case, every section is saved in a separate
+`.csv` file. The path to each `.csv` file can optionally be defined via the argument data classes.
+
+Instead of benchmarking pre-trained models via their model identifier, `e.g.` `bert-base-uncased`, the user can
+alternatively benchmark an arbitrary configuration of any available model class. In this case, a :obj:`list` of
+configurations must be inserted with the benchmark args as follows.
+
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> from transformers import PyTorchBenchmark, PyTorchBenchmarkArguments, BertConfig
+
+    >>> args = PyTorchBenchmarkArguments(models=["bert-base", "bert-384-hid", "bert-6-lay"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512])
+    >>> config_base = BertConfig()
+    >>> config_384_hid = BertConfig(hidden_size=384)
+    >>> config_6_lay = BertConfig(num_hidden_layers=6)
+
+    >>> benchmark = PyTorchBenchmark(args, configs=[config_base, config_384_hid, config_6_lay])
+    >>> benchmark.run()
+    ====================       INFERENCE - SPEED - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length       Time in s                  
+    --------------------------------------------------------------------------------
+    bert-base                  8              128            0.006
+    bert-base                  8              512            0.006
+    bert-base                  8              128            0.018     
+    bert-base                  8              512            0.088     
+    bert-384-hid              8               8             0.006     
+    bert-384-hid              8               32            0.006     
+    bert-384-hid              8              128            0.011     
+    bert-384-hid              8              512            0.054     
+    bert-6-lay                 8               8             0.003     
+    bert-6-lay                 8               32            0.004     
+    bert-6-lay                 8              128            0.009     
+    bert-6-lay                 8              512            0.044
+    --------------------------------------------------------------------------------
+
+    ====================      INFERENCE - MEMORY - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length      Memory in MB 
+    --------------------------------------------------------------------------------
+    bert-base                  8               8             1277
+    bert-base                  8               32            1281
+    bert-base                  8              128            1307     
+    bert-base                  8              512            1539     
+    bert-384-hid              8               8             1005     
+    bert-384-hid              8               32            1027     
+    bert-384-hid              8              128            1035     
+    bert-384-hid              8              512            1255     
+    bert-6-lay                 8               8             1097     
+    bert-6-lay                 8               32            1101     
+    bert-6-lay                 8              128            1127     
+    bert-6-lay                 8              512            1359
+    --------------------------------------------------------------------------------
+
+    ====================        ENVIRONMENT INFORMATION         ====================
+
+    - transformers_version: 2.11.0
+    - framework: PyTorch
+    - use_torchscript: False
+    - framework_version: 1.4.0
+    - python_version: 3.6.10
+    - system: Linux
+    - cpu: x86_64
+    - architecture: 64bit
+    - date: 2020-06-29
+    - time: 09:35:25.143267
+    - fp16: False
+    - use_multiprocessing: True
+    - only_pretrain_model: False
+    - cpu_ram_mb: 32088
+    - use_gpu: True
+    - num_gpus: 1
+    - gpu: TITAN RTX
+    - gpu_ram_mb: 24217
+    - gpu_power_watts: 280.0
+    - gpu_performance_state: 2
+    - use_tpu: False
+
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TensorFlowBenchmark, TensorFlowBenchmarkArguments, BertConfig
+
+    >>> args = TensorFlowBenchmarkArguments(models=["bert-base", "bert-384-hid", "bert-6-lay"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512])
+    >>> config_base = BertConfig()
+    >>> config_384_hid = BertConfig(hidden_size=384)
+    >>> config_6_lay = BertConfig(num_hidden_layers=6)
+
+    >>> benchmark = TensorFlowBenchmark(args, configs=[config_base, config_384_hid, config_6_lay])
+    >>> benchmark.run()
+    ====================       INFERENCE - SPEED - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length       Time in s                  
+    --------------------------------------------------------------------------------
+    bert-base                  8               8             0.005
+    bert-base                  8               32            0.008
+    bert-base                  8              128            0.022
+    bert-base                  8              512            0.106
+    bert-384-hid              8               8             0.005
+    bert-384-hid              8               32            0.007
+    bert-384-hid              8              128            0.018
+    bert-384-hid              8              512            0.064
+    bert-6-lay                 8               8             0.002
+    bert-6-lay                 8               32            0.003
+    bert-6-lay                 8              128            0.0011
+    bert-6-lay                 8              512            0.074
+    --------------------------------------------------------------------------------
+
+    ====================      INFERENCE - MEMORY - RESULT       ====================
+    --------------------------------------------------------------------------------
+    Model Name             Batch Size     Seq Length      Memory in MB 
+    --------------------------------------------------------------------------------
+    bert-base                  8               8             1330
+    bert-base                  8               32            1330
+    bert-base                  8              128            1330
+    bert-base                  8              512            1770
+    bert-384-hid              8               8             1330
+    bert-384-hid              8               32            1330
+    bert-384-hid              8              128            1330
+    bert-384-hid              8              512            1540
+    bert-6-lay                 8               8             1330
+    bert-6-lay                 8               32            1330
+    bert-6-lay                 8              128            1330
+    bert-6-lay                 8              512            1540
+    --------------------------------------------------------------------------------
+
+    ====================        ENVIRONMENT INFORMATION         ====================
+
+    - transformers_version: 2.11.0
+    - framework: Tensorflow
+    - use_xla: False
+    - framework_version: 2.2.0
+    - python_version: 3.6.10
+    - system: Linux
+    - cpu: x86_64
+    - architecture: 64bit
+    - date: 2020-06-29
+    - time: 09:38:15.487125
+    - fp16: False
+    - use_multiprocessing: True
+    - only_pretrain_model: False
+    - cpu_ram_mb: 32088
+    - use_gpu: True
+    - num_gpus: 1
+    - gpu: TITAN RTX
+    - gpu_ram_mb: 24217
+    - gpu_power_watts: 280.0
+    - gpu_performance_state: 2
+    - use_tpu: False
+
+
+Again, `inference time` and `required memory` for `inference` are measured, but this time for customized configurations
+of the :obj:`BertModel` class. This feature can especially be helpful when deciding for which configuration the model
+should be trained.
+
+
+Benchmark best practices
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+This section lists a couple of best practices one should be aware of when benchmarking a model.
+
+- Currently, only single device benchmarking is supported. When benchmarking on GPU, it is recommended that the user
+  specifies on which device the code should be run by setting the ``CUDA_VISIBLE_DEVICES`` environment variable in the
+  shell, `e.g.` ``export CUDA_VISIBLE_DEVICES=0`` before running the code.
+- The option :obj:`no_multi_processing` should only be set to :obj:`True` for testing and debugging. To ensure accurate
+  memory measurement it is recommended to run each memory benchmark in a separate process by making sure
+  :obj:`no_multi_processing` is set to :obj:`True`.
+- One should always state the environment information when sharing the results of a model benchmark. Results can vary
+  heavily between different GPU devices, library versions, etc., so that benchmark results on their own are not very
+  useful for the community.
+
+
+Sharing your benchmark
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Previously all available core models (10 at the time) have been benchmarked for `inference time`, across many different
+settings: using PyTorch, with and without TorchScript, using TensorFlow, with and without XLA. All of those tests were
+done across CPUs (except for TensorFlow XLA) and GPUs.
+
+The approach is detailed in the `following blogpost
+<https://medium.com/huggingface/benchmarking-transformers-pytorch-and-tensorflow-e2917fb891c2>`__ and the results are
+available `here
+<https://docs.google.com/spreadsheets/d/1sryqufw2D0XlUH4sq3e9Wnxu5EAQkaohzrJbd5HdQ_w/edit?usp=sharing>`__.
+
+With the new `benchmark` tools, it is easier than ever to share your benchmark results with the community
+
+- :prefix_link:`PyTorch Benchmarking Results<examples/pytorch/benchmarking/README.md>`.
+- :prefix_link:`TensorFlow Benchmarking Results<examples/tensorflow/benchmarking/README.md>`.

docs/source/bertology.rst

@@ -0,0 +1,38 @@
+.. 
+    Copyright 2020 The HuggingFace Team. All rights reserved.
+
+    Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+    the License. You may obtain a copy of the License at
+
+        http://www.apache.org/licenses/LICENSE-2.0
+
+    Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+    an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+    specific language governing permissions and limitations under the License.
+
+BERTology
+-----------------------------------------------------------------------------------------------------------------------
+
+There is a growing field of study concerned with investigating the inner working of large-scale transformers like BERT
+(that some call "BERTology"). Some good examples of this field are:
+
+
+* BERT Rediscovers the Classical NLP Pipeline by Ian Tenney, Dipanjan Das, Ellie Pavlick:
+  https://arxiv.org/abs/1905.05950
+* Are Sixteen Heads Really Better than One? by Paul Michel, Omer Levy, Graham Neubig: https://arxiv.org/abs/1905.10650
+* What Does BERT Look At? An Analysis of BERT's Attention by Kevin Clark, Urvashi Khandelwal, Omer Levy, Christopher D.
+  Manning: https://arxiv.org/abs/1906.04341
+
+In order to help this new field develop, we have included a few additional features in the BERT/GPT/GPT-2 models to
+help people access the inner representations, mainly adapted from the great work of Paul Michel
+(https://arxiv.org/abs/1905.10650):
+
+
+* accessing all the hidden-states of BERT/GPT/GPT-2,
+* accessing all the attention weights for each head of BERT/GPT/GPT-2,
+* retrieving heads output values and gradients to be able to compute head importance score and prune head as explained
+  in https://arxiv.org/abs/1905.10650.
+
+To help you understand and use these features, we have added a specific example script: :prefix_link:`bertology.py
+<examples/research_projects/bertology/run_bertology.py>` while extract information and prune a model pre-trained on
+GLUE.

docs/source/community.md

@@ -1,4 +1,4 @@
-# Community
+# Community
 
 This page regroups resources around 🤗 Transformers developed by the community.
 
@@ -6,13 +6,12 @@ This page regroups resources around 🤗 Transformers developed by the community
 
 | Resource     |      Description      |      Author      |
 |:----------|:-------------|------:|
-| [Hugging Face Transformers Glossary Flashcards](https://www.darigovresearch.com/huggingface-transformers-glossary-flashcards) | A set of flashcards based on the [Transformers Docs Glossary](glossary) that has been put into a form which can be easily learnt/revised using [Anki ](https://apps.ankiweb.net/) an open source, cross platform app specifically designed for long term knowledge retention. See this [Introductory video on how to use the flashcards](https://www.youtube.com/watch?v=Dji_h7PILrw). | [Darigov Research](https://www.darigovresearch.com/) |
+| [Hugging Face Transformers Glossary Flashcards](https://www.darigovresearch.com/huggingface-transformers-glossary-flashcards) | A set of flashcards based on the [Transformers Docs Glossary](https://huggingface.co/transformers/master/glossary.html) that has been put into a form which can be easily learnt/revised using [Anki ](https://apps.ankiweb.net/) an open source, cross platform app specifically designed for long term knowledge retention. See this [Introductory video on how to use the flashcards](https://www.youtube.com/watch?v=Dji_h7PILrw). | [Darigov Research](https://www.darigovresearch.com/) |
 
 ## Community notebooks:
 
 | Notebook     |      Description      |      Author      |      |
 |:----------|:-------------|:-------------|------:|
-| [Fine-tune a pre-trained Transformer to generate lyrics](https://github.com/AlekseyKorshuk/huggingartists) | How to generate lyrics in the style of your favorite artist by fine-tuning a GPT-2 model |  [Aleksey Korshuk](https://github.com/AlekseyKorshuk) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/AlekseyKorshuk/huggingartists/blob/master/huggingartists-demo.ipynb) |
 | [Train T5 in Tensorflow 2 ](https://github.com/snapthat/TF-T5-text-to-text) | How to train T5 for any task using Tensorflow 2. This notebook demonstrates a Question & Answer task implemented in Tensorflow 2 using SQUAD | [Muhammad Harris](https://github.com/HarrisDePerceptron) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/snapthat/TF-T5-text-to-text/blob/master/snapthatT5/notebooks/TF-T5-Datasets%20Training.ipynb) |
 | [Train T5 on TPU](https://github.com/patil-suraj/exploring-T5/blob/master/T5_on_TPU.ipynb)  | How to train T5 on SQUAD with Transformers and Nlp | [Suraj Patil](https://github.com/patil-suraj) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/exploring-T5/blob/master/T5_on_TPU.ipynb#scrollTo=QLGiFCDqvuil) |
 | [Fine-tune T5 for Classification and Multiple Choice](https://github.com/patil-suraj/exploring-T5/blob/master/t5_fine_tuning.ipynb)  | How to fine-tune T5 for classification and multiple choice tasks using a text-to-text format with PyTorch Lightning |  [Suraj Patil](https://github.com/patil-suraj) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/exploring-T5/blob/master/t5_fine_tuning.ipynb) |
@@ -36,7 +35,7 @@ This page regroups resources around 🤗 Transformers developed by the community
 |[fine-tune a non-English GPT-2 Model with Trainer class](https://github.com/philschmid/fine-tune-GPT-2/blob/master/Fine_tune_a_non_English_GPT_2_Model_with_Huggingface.ipynb) | How to fine-tune a non-English GPT-2 Model with Trainer class | [Philipp Schmid](https://www.philschmid.de) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/philschmid/fine-tune-GPT-2/blob/master/Fine_tune_a_non_English_GPT_2_Model_with_Huggingface.ipynb)|
 |[Fine-tune a DistilBERT Model for Multi Label Classification task](https://github.com/DhavalTaunk08/Transformers_scripts/blob/master/Transformers_multilabel_distilbert.ipynb) | How to fine-tune a DistilBERT Model for Multi Label Classification task | [Dhaval Taunk](https://github.com/DhavalTaunk08) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/DhavalTaunk08/Transformers_scripts/blob/master/Transformers_multilabel_distilbert.ipynb)|
 |[Fine-tune ALBERT for sentence-pair classification](https://github.com/NadirEM/nlp-notebooks/blob/master/Fine_tune_ALBERT_sentence_pair_classification.ipynb) | How to fine-tune an ALBERT model or another BERT-based model for the sentence-pair classification task | [Nadir El Manouzi](https://github.com/NadirEM) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NadirEM/nlp-notebooks/blob/master/Fine_tune_ALBERT_sentence_pair_classification.ipynb)|
-|[Fine-tune Roberta for sentiment analysis](https://github.com/DhavalTaunk08/NLP_scripts/blob/master/sentiment_analysis_using_roberta.ipynb) | How to fine-tune a Roberta model for sentiment analysis | [Dhaval Taunk](https://github.com/DhavalTaunk08) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/DhavalTaunk08/NLP_scripts/blob/master/sentiment_analysis_using_roberta.ipynb)|
+|[Fine-tune Roberta for sentiment analysis](https://github.com/DhavalTaunk08/NLP_scripts/blob/master/sentiment_analysis_using_roberta.ipynb) | How to fine-tune an Roberta model for sentiment analysis | [Dhaval Taunk](https://github.com/DhavalTaunk08) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/DhavalTaunk08/NLP_scripts/blob/master/sentiment_analysis_using_roberta.ipynb)|
 |[Evaluating Question Generation Models](https://github.com/flexudy-pipe/qugeev) | How accurate are the answers to questions generated by your seq2seq transformer model? | [Pascal Zoleko](https://github.com/zolekode) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1bpsSqCQU-iw_5nNoRm_crPq6FRuJthq_?usp=sharing)|
 |[Classify text with DistilBERT and Tensorflow](https://github.com/peterbayerle/huggingface_notebook/blob/main/distilbert_tf.ipynb) | How to fine-tune DistilBERT for text classification in TensorFlow | [Peter Bayerle](https://github.com/peterbayerle) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/peterbayerle/huggingface_notebook/blob/main/distilbert_tf.ipynb)|
 |[Leverage BERT for Encoder-Decoder Summarization on CNN/Dailymail](https://github.com/patrickvonplaten/notebooks/blob/master/BERT2BERT_for_CNN_Dailymail.ipynb) | How to warm-start a *EncoderDecoderModel* with a *bert-base-uncased* checkpoint for summarization on CNN/Dailymail | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/BERT2BERT_for_CNN_Dailymail.ipynb)|
@@ -62,4 +61,3 @@ This page regroups resources around 🤗 Transformers developed by the community
 | [Speech Emotion Classification with Wav2Vec2](https://github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb) | How to leverage a pretrained Wav2Vec2 model for Emotion Classification on the MEGA dataset | [Mehrdad Farahani](https://github.com/m3hrdadfi) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb) |
 | [Detect objects in an image with DETR](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/DETR/DETR_minimal_example_(with_DetrFeatureExtractor).ipynb) | How to use a trained *DetrForObjectDetection* model to detect objects in an image and visualize attention | [Niels Rogge](https://github.com/NielsRogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/DETR_minimal_example_(with_DetrFeatureExtractor).ipynb) |
 | [Fine-tune DETR on a custom object detection dataset](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForObjectDetection_on_custom_dataset_(balloon).ipynb) | How to fine-tune *DetrForObjectDetection* on a custom object detection dataset | [Niels Rogge](https://github.com/NielsRogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForObjectDetection_on_custom_dataset_(balloon).ipynb) |
-| [Finetune T5 for Named Entity Recognition](https://github.com/ToluClassics/Notebooks/blob/main/T5_Ner_Finetuning.ipynb) | How to fine-tune *T5* on a Named Entity Recognition Task | [Ogundepo Odunayo](https://github.com/ToluClassics) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1obr78FY_cBmWY5ODViCmzdY6O1KB65Vc?usp=sharing) |

docs/source/conf.py

@@ -0,0 +1,218 @@
+# -*- coding: utf-8 -*-
+#
+# Configuration file for the Sphinx documentation builder.
+#
+# This file does only contain a selection of the most common options. For a
+# full list see the documentation:
+# http://www.sphinx-doc.org/en/master/config
+
+# -- Path setup --------------------------------------------------------------
+
+# If extensions (or modules to document with autodoc) are in another directory,
+# add these directories to sys.path here. If the directory is relative to the
+# documentation root, use os.path.abspath to make it absolute, like shown here.
+#
+import os
+import sys
+
+sys.path.insert(0, os.path.abspath("../../src"))
+
+
+# -- Project information -----------------------------------------------------
+
+project = "transformers"
+copyright = "2020, The Hugging Face Team, Licenced under the Apache License, Version 2.0"
+author = "huggingface"
+
+# The short X.Y version
+version = ""
+# The full version, including alpha/beta/rc tags
+release = u'4.7.0'
+
+
+
+# Prefix link to point to master, comment this during version release and uncomment below line
+extlinks = {"prefix_link": ("https://github.com/huggingface/transformers/blob/master/%s", "")}
+# Prefix link to always point to corresponding version, uncomment this during version release
+# extlinks = {'prefix_link': ('https://github.com/huggingface/transformers/blob/v'+ release + '/%s', '')}
+
+# -- General configuration ---------------------------------------------------
+
+# If your documentation needs a minimal Sphinx version, state it here.
+#
+# needs_sphinx = '1.0'
+
+# Add any Sphinx extension module names here, as strings. They can be
+# extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
+# ones.
+extensions = [
+    "sphinx.ext.autodoc",
+    "sphinx.ext.extlinks",
+    "sphinx.ext.coverage",
+    "sphinx.ext.napoleon",
+    "recommonmark",
+    "sphinx.ext.viewcode",
+    "sphinx_markdown_tables",
+    "sphinxext.opengraph",
+    "sphinx_copybutton",
+]
+
+# Add any paths that contain templates here, relative to this directory.
+templates_path = ["_templates"]
+
+# The suffix(es) of source filenames.
+# You can specify multiple suffix as a list of string:
+#
+source_suffix = [".rst", ".md"]
+# source_suffix = '.rst'
+
+# The master toctree document.
+master_doc = "index"
+
+# The language for content autogenerated by Sphinx. Refer to documentation
+# for a list of supported languages.
+#
+# This is also used if you do content translation via gettext catalogs.
+# Usually you set "language" from the command line for these cases.
+language = None
+
+# List of patterns, relative to source directory, that match files and
+# directories to ignore when looking for source files.
+# This pattern also affects html_static_path and html_extra_path.
+exclude_patterns = ["_build", "Thumbs.db", ".DS_Store"]
+
+# The name of the Pygments (syntax highlighting) style to use.
+pygments_style = None
+
+# Remove the prompt when copying examples
+copybutton_prompt_text = r">>> |\.\.\. "
+copybutton_prompt_is_regexp = True
+
+# -- Options for HTML output -------------------------------------------------
+
+# The theme to use for HTML and HTML Help pages.  See the documentation for
+# a list of builtin themes.
+#
+html_theme = "sphinx_rtd_theme"
+
+# Theme options are theme-specific and customize the look and feel of a theme
+# further.  For a list of options available for each theme, see the
+# documentation.
+#
+html_theme_options = {"analytics_id": "UA-83738774-2", "navigation_with_keys": True}
+
+#  Configuration for OpenGraph and Twitter Card Tags.
+# These are responsible for creating nice shareable social images https://ahrefs.com/blog/open-graph-meta-tags/
+# https://ogp.me/#type_website
+ogp_image = "https://huggingface.co/front/thumbnails/transformers.png"
+ogp_description = "State-of-the-art Natural Language Processing for PyTorch and TensorFlow 2.0. Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, question answering, summarization, translation, text generation, etc in 100+ languages. Its aim is to make cutting-edge NLP easier to use for everyone"
+ogp_description_length = 160
+
+ogp_custom_meta_tags = [
+    f'<meta name="twitter:image" content="{ogp_image}">',
+    f'<meta name="twitter:description" content="{ogp_description}">',
+]
+
+# Add any paths that contain custom static files (such as style sheets) here,
+# relative to this directory. They are copied after the builtin static files,
+# so a file named "default.css" will overwrite the builtin "default.css".
+html_static_path = ["_static"]
+
+# Custom sidebar templates, must be a dictionary that maps document names
+# to template names.
+#
+# The default sidebars (for documents that don't match any pattern) are
+# defined by theme itself.  Builtin themes are using these templates by
+# default: ``['localtoc.html', 'relations.html', 'sourcelink.html',
+# 'searchbox.html']``.
+#
+# html_sidebars = {}
+
+# This must be the name of an image file (path relative to the configuration
+# directory) that is the favicon of the docs. Modern browsers use this as
+# the icon for tabs, windows and bookmarks. It should be a Windows-style
+# icon file (.ico).
+html_favicon = "favicon.ico"
+
+
+# -- Options for HTMLHelp output ---------------------------------------------
+
+# Output file base name for HTML help builder.
+htmlhelp_basename = "transformersdoc"
+
+
+# -- Options for LaTeX output ------------------------------------------------
+
+latex_elements = {
+    # The paper size ('letterpaper' or 'a4paper').
+    #
+    # 'papersize': 'letterpaper',
+    # The font size ('10pt', '11pt' or '12pt').
+    #
+    # 'pointsize': '10pt',
+    # Additional stuff for the LaTeX preamble.
+    #
+    # 'preamble': '',
+    # Latex figure (float) alignment
+    #
+    # 'figure_align': 'htbp',
+}
+
+# Grouping the document tree into LaTeX files. List of tuples
+# (source start file, target name, title,
+#  author, documentclass [howto, manual, or own class]).
+latex_documents = [
+    (master_doc, "transformers.tex", "transformers Documentation", "huggingface", "manual"),
+]
+
+
+# -- Options for manual page output ------------------------------------------
+
+# One entry per manual page. List of tuples
+# (source start file, name, description, authors, manual section).
+man_pages = [(master_doc, "transformers", "transformers Documentation", [author], 1)]
+
+
+# -- Options for Texinfo output ----------------------------------------------
+
+# Grouping the document tree into Texinfo files. List of tuples
+# (source start file, target name, title, author,
+#  dir menu entry, description, category)
+texinfo_documents = [
+    (
+        master_doc,
+        "transformers",
+        "transformers Documentation",
+        author,
+        "transformers",
+        "One line description of project.",
+        "Miscellaneous",
+    ),
+]
+
+
+# -- Options for Epub output -------------------------------------------------
+
+# Bibliographic Dublin Core info.
+epub_title = project
+
+# The unique identifier of the text. This can be a ISBN number
+# or the project homepage.
+#
+# epub_identifier = ''
+
+# A unique identification for the text.
+#
+# epub_uid = ''
+
+# A list of files that should not be packed into the epub file.
+epub_exclude_files = ["search.html"]
+
+
+def setup(app):
+    app.add_css_file("css/huggingface.css")
+    app.add_css_file("css/code-snippets.css")
+    app.add_js_file("js/custom.js")
+
+
+# -- Extension configuration -------------------------------------------------

docs/source/contributing.md

@@ -0,0 +1 @@
+../../CONTRIBUTING.md

docs/source/converting_tensorflow_models.rst

@@ -0,0 +1,181 @@
+.. 
+    Copyright 2020 The HuggingFace Team. All rights reserved.
+
+    Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+    the License. You may obtain a copy of the License at
+
+        http://www.apache.org/licenses/LICENSE-2.0
+
+    Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+    an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+    specific language governing permissions and limitations under the License.
+
+Converting Tensorflow Checkpoints
+=======================================================================================================================
+
+A command-line interface is provided to convert original Bert/GPT/GPT-2/Transformer-XL/XLNet/XLM checkpoints in models
+than be loaded using the ``from_pretrained`` methods of the library.
+
+.. note::
+    Since 2.3.0 the conversion script is now part of the transformers CLI (**transformers-cli**) available in any
+    transformers >= 2.3.0 installation.
+
+    The documentation below reflects the **transformers-cli convert** command format.
+
+BERT
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+You can convert any TensorFlow checkpoint for BERT (in particular `the pre-trained models released by Google
+<https://github.com/google-research/bert#pre-trained-models>`_\ ) in a PyTorch save file by using the
+:prefix_link:`convert_bert_original_tf_checkpoint_to_pytorch.py
+<src/transformers/models/bert/convert_bert_original_tf_checkpoint_to_pytorch.py>` script.
+
+This CLI takes as input a TensorFlow checkpoint (three files starting with ``bert_model.ckpt``\ ) and the associated
+configuration file (\ ``bert_config.json``\ ), and creates a PyTorch model for this configuration, loads the weights
+from the TensorFlow checkpoint in the PyTorch model and saves the resulting model in a standard PyTorch save file that
+can be imported using ``from_pretrained()`` (see example in :doc:`quicktour` , :prefix_link:`run_glue.py
+<examples/pytorch/text-classification/run_glue.py>` \ ).
+
+You only need to run this conversion script **once** to get a PyTorch model. You can then disregard the TensorFlow
+checkpoint (the three files starting with ``bert_model.ckpt``\ ) but be sure to keep the configuration file (\
+``bert_config.json``\ ) and the vocabulary file (\ ``vocab.txt``\ ) as these are needed for the PyTorch model too.
+
+To run this specific conversion script you will need to have TensorFlow and PyTorch installed (\ ``pip install
+tensorflow``\ ). The rest of the repository only requires PyTorch.
+
+Here is an example of the conversion process for a pre-trained ``BERT-Base Uncased`` model:
+
+.. code-block:: shell
+
+    export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12
+
+    transformers-cli convert --model_type bert \
+      --tf_checkpoint $BERT_BASE_DIR/bert_model.ckpt \
+      --config $BERT_BASE_DIR/bert_config.json \
+      --pytorch_dump_output $BERT_BASE_DIR/pytorch_model.bin
+
+You can download Google's pre-trained models for the conversion `here
+<https://github.com/google-research/bert#pre-trained-models>`__.
+
+ALBERT
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Convert TensorFlow model checkpoints of ALBERT to PyTorch using the
+:prefix_link:`convert_albert_original_tf_checkpoint_to_pytorch.py
+<src/transformers/models/albert/convert_albert_original_tf_checkpoint_to_pytorch.py>` script.
+
+The CLI takes as input a TensorFlow checkpoint (three files starting with ``model.ckpt-best``\ ) and the accompanying
+configuration file (\ ``albert_config.json``\ ), then creates and saves a PyTorch model. To run this conversion you
+will need to have TensorFlow and PyTorch installed.
+
+Here is an example of the conversion process for the pre-trained ``ALBERT Base`` model:
+
+.. code-block:: shell
+
+    export ALBERT_BASE_DIR=/path/to/albert/albert_base
+
+    transformers-cli convert --model_type albert \
+      --tf_checkpoint $ALBERT_BASE_DIR/model.ckpt-best \
+      --config $ALBERT_BASE_DIR/albert_config.json \
+      --pytorch_dump_output $ALBERT_BASE_DIR/pytorch_model.bin
+
+You can download Google's pre-trained models for the conversion `here
+<https://github.com/google-research/albert#pre-trained-models>`__.
+
+OpenAI GPT
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Here is an example of the conversion process for a pre-trained OpenAI GPT model, assuming that your NumPy checkpoint
+save as the same format than OpenAI pretrained model (see `here <https://github.com/openai/finetune-transformer-lm>`__\
+)
+
+.. code-block:: shell
+
+    export OPENAI_GPT_CHECKPOINT_FOLDER_PATH=/path/to/openai/pretrained/numpy/weights
+
+    transformers-cli convert --model_type gpt \
+      --tf_checkpoint $OPENAI_GPT_CHECKPOINT_FOLDER_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--config OPENAI_GPT_CONFIG] \
+      [--finetuning_task_name OPENAI_GPT_FINETUNED_TASK] \
+
+
+OpenAI GPT-2
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Here is an example of the conversion process for a pre-trained OpenAI GPT-2 model (see `here
+<https://github.com/openai/gpt-2>`__\ )
+
+.. code-block:: shell
+
+    export OPENAI_GPT2_CHECKPOINT_PATH=/path/to/gpt2/pretrained/weights
+
+    transformers-cli convert --model_type gpt2 \
+      --tf_checkpoint $OPENAI_GPT2_CHECKPOINT_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--config OPENAI_GPT2_CONFIG] \
+      [--finetuning_task_name OPENAI_GPT2_FINETUNED_TASK]
+
+Transformer-XL
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Here is an example of the conversion process for a pre-trained Transformer-XL model (see `here
+<https://github.com/kimiyoung/transformer-xl/tree/master/tf#obtain-and-evaluate-pretrained-sota-models>`__\ )
+
+.. code-block:: shell
+
+    export TRANSFO_XL_CHECKPOINT_FOLDER_PATH=/path/to/transfo/xl/checkpoint
+
+    transformers-cli convert --model_type transfo_xl \
+      --tf_checkpoint $TRANSFO_XL_CHECKPOINT_FOLDER_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--config TRANSFO_XL_CONFIG] \
+      [--finetuning_task_name TRANSFO_XL_FINETUNED_TASK]
+
+
+XLNet
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Here is an example of the conversion process for a pre-trained XLNet model:
+
+.. code-block:: shell
+
+    export TRANSFO_XL_CHECKPOINT_PATH=/path/to/xlnet/checkpoint
+    export TRANSFO_XL_CONFIG_PATH=/path/to/xlnet/config
+
+    transformers-cli convert --model_type xlnet \
+      --tf_checkpoint $TRANSFO_XL_CHECKPOINT_PATH \
+      --config $TRANSFO_XL_CONFIG_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--finetuning_task_name XLNET_FINETUNED_TASK] \
+
+
+XLM
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Here is an example of the conversion process for a pre-trained XLM model:
+
+.. code-block:: shell
+
+    export XLM_CHECKPOINT_PATH=/path/to/xlm/checkpoint
+
+    transformers-cli convert --model_type xlm \
+      --tf_checkpoint $XLM_CHECKPOINT_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT
+     [--config XML_CONFIG] \
+     [--finetuning_task_name XML_FINETUNED_TASK]
+
+
+T5
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Here is an example of the conversion process for a pre-trained T5 model:
+
+.. code-block:: shell
+
+    export T5=/path/to/t5/uncased_L-12_H-768_A-12
+
+    transformers-cli convert --model_type t5 \
+      --tf_checkpoint $T5/t5_model.ckpt \
+      --config $T5/t5_config.json \
+      --pytorch_dump_output $T5/pytorch_model.bin

docs/source/custom_datasets.rst

@@ -0,0 +1,729 @@
+.. 
+    Copyright 2020 The HuggingFace Team. All rights reserved.
+
+    Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+    the License. You may obtain a copy of the License at
+
+        http://www.apache.org/licenses/LICENSE-2.0
+
+    Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+    an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+    specific language governing permissions and limitations under the License.
+
+Fine-tuning with custom datasets
+=======================================================================================================================
+
+.. note::
+
+    The datasets used in this tutorial are available and can be more easily accessed using the `🤗 Datasets library
+    <https://github.com/huggingface/datasets>`_. We do not use this library to access the datasets here since this
+    tutorial meant to illustrate how to work with your own data. A brief of introduction can be found at the end of the
+    tutorial in the section ":ref:`datasetslib`".
+
+This tutorial will take you through several examples of using 🤗 Transformers models with your own datasets. The guide
+shows one of many valid workflows for using these models and is meant to be illustrative rather than definitive. We
+show examples of reading in several data formats, preprocessing the data for several types of tasks, and then preparing
+the data into PyTorch/TensorFlow ``Dataset`` objects which can easily be used either with
+:class:`~transformers.Trainer`/:class:`~transformers.TFTrainer` or with native PyTorch/TensorFlow.
+
+We include several examples, each of which demonstrates a different type of common downstream task:
+
+  - :ref:`seq_imdb`
+  - :ref:`tok_ner`
+  - :ref:`qa_squad`
+  - :ref:`resources`
+
+.. _seq_imdb:
+
+Sequence Classification with IMDb Reviews
+-----------------------------------------------------------------------------------------------------------------------
+
+.. note::
+
+    This dataset can be explored in the Hugging Face model hub (`IMDb <https://huggingface.co/datasets/imdb>`_), and
+    can be alternatively downloaded with the 🤗 Datasets library with ``load_dataset("imdb")``.
+
+In this example, we'll show how to download, tokenize, and train a model on the IMDb reviews dataset. This task takes
+the text of a review and requires the model to predict whether the sentiment of the review is positive or negative.
+Let's start by downloading the dataset from the `Large Movie Review Dataset
+<http://ai.stanford.edu/~amaas/data/sentiment/>`_ webpage.
+
+.. code-block:: bash
+
+    wget http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz
+    tar -xf aclImdb_v1.tar.gz
+
+This data is organized into ``pos`` and ``neg`` folders with one text file per example. Let's write a function that can
+read this in.
+
+.. code-block:: python
+
+    from pathlib import Path
+
+    def read_imdb_split(split_dir):
+        split_dir = Path(split_dir)
+        texts = []
+        labels = []
+        for label_dir in ["pos", "neg"]:
+            for text_file in (split_dir/label_dir).iterdir():
+                texts.append(text_file.read_text())
+                labels.append(0 if label_dir is "neg" else 1)
+
+        return texts, labels
+
+    train_texts, train_labels = read_imdb_split('aclImdb/train')
+    test_texts, test_labels = read_imdb_split('aclImdb/test')
+
+We now have a train and test dataset, but let's also also create a validation set which we can use for for evaluation
+and tuning without tainting our test set results. Sklearn has a convenient utility for creating such splits:
+
+.. code-block:: python
+
+    from sklearn.model_selection import train_test_split
+    train_texts, val_texts, train_labels, val_labels = train_test_split(train_texts, train_labels, test_size=.2)
+
+Alright, we've read in our dataset. Now let's tackle tokenization. We'll eventually train a classifier using
+pre-trained DistilBert, so let's use the DistilBert tokenizer.
+
+.. code-block:: python
+
+    from transformers import DistilBertTokenizerFast
+    tokenizer = DistilBertTokenizerFast.from_pretrained('distilbert-base-uncased')
+
+Now we can simply pass our texts to the tokenizer. We'll pass ``truncation=True`` and ``padding=True``, which will
+ensure that all of our sequences are padded to the same length and are truncated to be no longer model's maximum input
+length. This will allow us to feed batches of sequences into the model at the same time.
+
+.. code-block:: python
+
+    train_encodings = tokenizer(train_texts, truncation=True, padding=True)
+    val_encodings = tokenizer(val_texts, truncation=True, padding=True)
+    test_encodings = tokenizer(test_texts, truncation=True, padding=True)
+
+Now, let's turn our labels and encodings into a Dataset object. In PyTorch, this is done by subclassing a
+``torch.utils.data.Dataset`` object and implementing ``__len__`` and ``__getitem__``. In TensorFlow, we pass our input
+encodings and labels to the ``from_tensor_slices`` constructor method. We put the data in this format so that the data
+can be easily batched such that each key in the batch encoding corresponds to a named parameter of the
+:meth:`~transformers.DistilBertForSequenceClassification.forward` method of the model we will train.
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    import torch
+
+    class IMDbDataset(torch.utils.data.Dataset):
+        def __init__(self, encodings, labels):
+            self.encodings = encodings
+            self.labels = labels
+
+        def __getitem__(self, idx):
+            item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
+            item['labels'] = torch.tensor(self.labels[idx])
+            return item
+
+        def __len__(self):
+            return len(self.labels)
+
+    train_dataset = IMDbDataset(train_encodings, train_labels)
+    val_dataset = IMDbDataset(val_encodings, val_labels)
+    test_dataset = IMDbDataset(test_encodings, test_labels)
+    ## TENSORFLOW CODE
+    import tensorflow as tf
+
+    train_dataset = tf.data.Dataset.from_tensor_slices((
+        dict(train_encodings),
+        train_labels
+    ))
+    val_dataset = tf.data.Dataset.from_tensor_slices((
+        dict(val_encodings),
+        val_labels
+    ))
+    test_dataset = tf.data.Dataset.from_tensor_slices((
+        dict(test_encodings),
+        test_labels
+    ))
+
+Now that our datasets our ready, we can fine-tune a model either with the 🤗
+:class:`~transformers.Trainer`/:class:`~transformers.TFTrainer` or with native PyTorch/TensorFlow. See :doc:`training
+<training>`.
+
+.. _ft_trainer:
+
+Fine-tuning with Trainer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The steps above prepared the datasets in the way that the trainer is expected. Now all we need to do is create a model
+to fine-tune, define the :class:`~transformers.TrainingArguments`/:class:`~transformers.TFTrainingArguments` and
+instantiate a :class:`~transformers.Trainer`/:class:`~transformers.TFTrainer`.
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    from transformers import DistilBertForSequenceClassification, Trainer, TrainingArguments
+
+    training_args = TrainingArguments(
+        output_dir='./results',          # output directory
+        num_train_epochs=3,              # total number of training epochs
+        per_device_train_batch_size=16,  # batch size per device during training
+        per_device_eval_batch_size=64,   # batch size for evaluation
+        warmup_steps=500,                # number of warmup steps for learning rate scheduler
+        weight_decay=0.01,               # strength of weight decay
+        logging_dir='./logs',            # directory for storing logs
+        logging_steps=10,
+    )
+
+    model = DistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")
+
+    trainer = Trainer(
+        model=model,                         # the instantiated 🤗 Transformers model to be trained
+        args=training_args,                  # training arguments, defined above
+        train_dataset=train_dataset,         # training dataset
+        eval_dataset=val_dataset             # evaluation dataset
+    )
+
+    trainer.train()
+    ## TENSORFLOW CODE
+    from transformers import TFDistilBertForSequenceClassification, TFTrainer, TFTrainingArguments
+
+    training_args = TFTrainingArguments(
+        output_dir='./results',          # output directory
+        num_train_epochs=3,              # total number of training epochs
+        per_device_train_batch_size=16,  # batch size per device during training
+        per_device_eval_batch_size=64,   # batch size for evaluation
+        warmup_steps=500,                # number of warmup steps for learning rate scheduler
+        weight_decay=0.01,               # strength of weight decay
+        logging_dir='./logs',            # directory for storing logs
+        logging_steps=10,
+    )
+
+    with training_args.strategy.scope():
+        model = TFDistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")
+
+    trainer = TFTrainer(
+        model=model,                         # the instantiated 🤗 Transformers model to be trained
+        args=training_args,                  # training arguments, defined above
+        train_dataset=train_dataset,         # training dataset
+        eval_dataset=val_dataset             # evaluation dataset
+    )
+
+    trainer.train()
+
+.. _ft_native:
+
+Fine-tuning with native PyTorch/TensorFlow
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+We can also train use native PyTorch or TensorFlow:
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    from torch.utils.data import DataLoader
+    from transformers import DistilBertForSequenceClassification, AdamW
+
+    device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
+
+    model = DistilBertForSequenceClassification.from_pretrained('distilbert-base-uncased')
+    model.to(device)
+    model.train()
+
+    train_loader = DataLoader(train_dataset, batch_size=16, shuffle=True)
+
+    optim = AdamW(model.parameters(), lr=5e-5)
+
+    for epoch in range(3):
+        for batch in train_loader:
+            optim.zero_grad()
+            input_ids = batch['input_ids'].to(device)
+            attention_mask = batch['attention_mask'].to(device)
+            labels = batch['labels'].to(device)
+            outputs = model(input_ids, attention_mask=attention_mask, labels=labels)
+            loss = outputs[0]
+            loss.backward()
+            optim.step()
+
+    model.eval()
+    ## TENSORFLOW CODE
+    from transformers import TFDistilBertForSequenceClassification
+
+    model = TFDistilBertForSequenceClassification.from_pretrained('distilbert-base-uncased')
+
+    optimizer = tf.keras.optimizers.Adam(learning_rate=5e-5)
+    model.compile(optimizer=optimizer, loss=model.compute_loss) # can also use any keras loss fn
+    model.fit(train_dataset.shuffle(1000).batch(16), epochs=3, batch_size=16)
+
+.. _tok_ner:
+
+Token Classification with W-NUT Emerging Entities
+-----------------------------------------------------------------------------------------------------------------------
+
+.. note::
+
+    This dataset can be explored in the Hugging Face model hub (`WNUT-17 <https://huggingface.co/datasets/wnut_17>`_),
+    and can be alternatively downloaded with the 🤗 Datasets library with ``load_dataset("wnut_17")``.
+
+Next we will look at token classification. Rather than classifying an entire sequence, this task classifies token by
+token. We'll demonstrate how to do this with `Named Entity Recognition
+<http://nlpprogress.com/english/named_entity_recognition.html>`_, which involves identifying tokens which correspond to
+a predefined set of "entities". Specifically, we'll use the `W-NUT Emerging and Rare entities
+<http://noisy-text.github.io/2017/emerging-rare-entities.html>`_ corpus. The data is given as a collection of
+pre-tokenized documents where each token is assigned a tag.
+
+Let's start by downloading the data.
+
+.. code-block:: bash
+
+    wget http://noisy-text.github.io/2017/files/wnut17train.conll
+
+In this case, we'll just download the train set, which is a single text file. Each line of the file contains either (1)
+a word and tag separated by a tab, or (2) a blank line indicating the end of a document. Let's write a function to read
+this in. We'll take in the file path and return ``token_docs`` which is a list of lists of token strings, and
+``token_tags`` which is a list of lists of tag strings.
+
+.. code-block:: python
+
+    from pathlib import Path
+    import re
+
+    def read_wnut(file_path):
+        file_path = Path(file_path)
+
+        raw_text = file_path.read_text().strip()
+        raw_docs = re.split(r'\n\t?\n', raw_text)
+        token_docs = []
+        tag_docs = []
+        for doc in raw_docs:
+            tokens = []
+            tags = []
+            for line in doc.split('\n'):
+                token, tag = line.split('\t')
+                tokens.append(token)
+                tags.append(tag)
+            token_docs.append(tokens)
+            tag_docs.append(tags)
+
+        return token_docs, tag_docs
+
+    texts, tags = read_wnut('wnut17train.conll')
+
+Just to see what this data looks like, let's take a look at a segment of the first document.
+
+.. code-block:: python
+
+    >>> print(texts[0][10:17], tags[0][10:17], sep='\n')
+    ['for', 'two', 'weeks', '.', 'Empire', 'State', 'Building']
+    ['O', 'O', 'O', 'O', 'B-location', 'I-location', 'I-location']
+
+``location`` is an entity type, ``B-`` indicates the beginning of an entity, and ``I-`` indicates consecutive positions
+of the same entity ("Empire State Building" is considered one entity). ``O`` indicates the token does not correspond to
+any entity.
+
+Now that we've read the data in, let's create a train/validation split:
+
+.. code-block:: python
+
+    from sklearn.model_selection import train_test_split
+    train_texts, val_texts, train_tags, val_tags = train_test_split(texts, tags, test_size=.2)
+
+Next, let's create encodings for our tokens and tags. For the tags, we can start by just create a simple mapping which
+we'll use in a moment:
+
+.. code-block:: python
+
+    unique_tags = set(tag for doc in tags for tag in doc)
+    tag2id = {tag: id for id, tag in enumerate(unique_tags)}
+    id2tag = {id: tag for tag, id in tag2id.items()}
+
+To encode the tokens, we'll use a pre-trained DistilBert tokenizer. We can tell the tokenizer that we're dealing with
+ready-split tokens rather than full sentence strings by passing ``is_split_into_words=True``. We'll also pass
+``padding=True`` and ``truncation=True`` to pad the sequences to be the same length. Lastly, we can tell the model to
+return information about the tokens which are split by the wordpiece tokenization process, which we will need in a
+moment.
+
+.. code-block:: python
+
+    from transformers import DistilBertTokenizerFast
+    tokenizer = DistilBertTokenizerFast.from_pretrained('distilbert-base-cased')
+    train_encodings = tokenizer(train_texts, is_split_into_words=True, return_offsets_mapping=True, padding=True, truncation=True)
+    val_encodings = tokenizer(val_texts, is_split_into_words=True, return_offsets_mapping=True, padding=True, truncation=True)
+
+Great, so now our tokens are nicely encoded in the format that they need to be in to feed them into our DistilBert
+model below.
+
+Now we arrive at a common obstacle with using pre-trained models for token-level classification: many of the tokens in
+the W-NUT corpus are not in DistilBert's vocabulary. Bert and many models like it use a method called WordPiece
+Tokenization, meaning that single words are split into multiple tokens such that each token is likely to be in the
+vocabulary. For example, DistilBert's tokenizer would split the Twitter handle ``@huggingface`` into the tokens ``['@',
+'hugging', '##face']``. This is a problem for us because we have exactly one tag per token. If the tokenizer splits a
+token into multiple sub-tokens, then we will end up with a mismatch between our tokens and our labels.
+
+One way to handle this is to only train on the tag labels for the first subtoken of a split token. We can do this in 🤗
+Transformers by setting the labels we wish to ignore to ``-100``. In the example above, if the label for
+``@HuggingFace`` is ``3`` (indexing ``B-corporation``), we would set the labels of ``['@', 'hugging', '##face']`` to
+``[3, -100, -100]``.
+
+Let's write a function to do this. This is where we will use the ``offset_mapping`` from the tokenizer as mentioned
+above. For each sub-token returned by the tokenizer, the offset mapping gives us a tuple indicating the sub-token's
+start position and end position relative to the original token it was split from. That means that if the first position
+in the tuple is anything other than ``0``, we will set its corresponding label to ``-100``. While we're at it, we can
+also set labels to ``-100`` if the second position of the offset mapping is ``0``, since this means it must be a
+special token like ``[PAD]`` or ``[CLS]``.
+
+.. note::
+
+    Due to a recently fixed bug, -1 must be used instead of -100 when using TensorFlow in 🤗 Transformers <= 3.02.
+
+.. code-block:: python
+
+    import numpy as np
+
+    def encode_tags(tags, encodings):
+        labels = [[tag2id[tag] for tag in doc] for doc in tags]
+        encoded_labels = []
+        for doc_labels, doc_offset in zip(labels, encodings.offset_mapping):
+            # create an empty array of -100
+            doc_enc_labels = np.ones(len(doc_offset),dtype=int) * -100
+            arr_offset = np.array(doc_offset)
+
+            # set labels whose first offset position is 0 and the second is not 0
+            doc_enc_labels[(arr_offset[:,0] == 0) & (arr_offset[:,1] != 0)] = doc_labels
+            encoded_labels.append(doc_enc_labels.tolist())
+
+        return encoded_labels
+
+    train_labels = encode_tags(train_tags, train_encodings)
+    val_labels = encode_tags(val_tags, val_encodings)
+
+The hard part is now done. Just as in the sequence classification example above, we can create a dataset object:
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    import torch
+
+    class WNUTDataset(torch.utils.data.Dataset):
+        def __init__(self, encodings, labels):
+            self.encodings = encodings
+            self.labels = labels
+
+        def __getitem__(self, idx):
+            item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
+            item['labels'] = torch.tensor(self.labels[idx])
+            return item
+
+        def __len__(self):
+            return len(self.labels)
+
+    train_encodings.pop("offset_mapping") # we don't want to pass this to the model
+    val_encodings.pop("offset_mapping")
+    train_dataset = WNUTDataset(train_encodings, train_labels)
+    val_dataset = WNUTDataset(val_encodings, val_labels)
+    ## TENSORFLOW CODE
+    import tensorflow as tf
+
+    train_encodings.pop("offset_mapping") # we don't want to pass this to the model
+    val_encodings.pop("offset_mapping")
+
+    train_dataset = tf.data.Dataset.from_tensor_slices((
+        dict(train_encodings),
+        train_labels
+    ))
+    val_dataset = tf.data.Dataset.from_tensor_slices((
+        dict(val_encodings),
+        val_labels
+    ))
+
+Now load in a token classification model and specify the number of labels:
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    from transformers import DistilBertForTokenClassification
+    model = DistilBertForTokenClassification.from_pretrained('distilbert-base-cased', num_labels=len(unique_tags))
+    ## TENSORFLOW CODE
+    from transformers import TFDistilBertForTokenClassification
+    model = TFDistilBertForTokenClassification.from_pretrained('distilbert-base-cased', num_labels=len(unique_tags))
+
+The data and model are both ready to go. You can train the model either with
+:class:`~transformers.Trainer`/:class:`~transformers.TFTrainer` or with native PyTorch/TensorFlow, exactly as in the
+sequence classification example above.
+
+  - :ref:`ft_trainer`
+  - :ref:`ft_native`
+
+.. _qa_squad:
+
+Question Answering with SQuAD 2.0
+-----------------------------------------------------------------------------------------------------------------------
+
+.. note::
+
+    This dataset can be explored in the Hugging Face model hub (`SQuAD V2
+    <https://huggingface.co/datasets/squad_v2>`_), and can be alternatively downloaded with the 🤗 Datasets library with
+    ``load_dataset("squad_v2")``.
+
+Question answering comes in many forms. In this example, we'll look at the particular type of extractive QA that
+involves answering a question about a passage by highlighting the segment of the passage that answers the question.
+This involves fine-tuning a model which predicts a start position and an end position in the passage. We will use the
+`Stanford Question Answering Dataset (SQuAD) 2.0 <https://rajpurkar.github.io/SQuAD-explorer/>`_.
+
+We will start by downloading the data:
+
+.. code-block:: bash
+
+    mkdir squad
+    wget https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v2.0.json -O squad/train-v2.0.json
+    wget https://rajpurkar.github.io/SQuAD-explorer/dataset/dev-v2.0.json -O squad/dev-v2.0.json
+
+Each split is in a structured json file with a number of questions and answers for each passage (or context). We'll
+take this apart into parallel lists of contexts, questions, and answers (note that the contexts here are repeated since
+there are multiple questions per context):
+
+.. code-block:: python
+
+    import json
+    from pathlib import Path
+
+    def read_squad(path):
+        path = Path(path)
+        with open(path, 'rb') as f:
+            squad_dict = json.load(f)
+
+        contexts = []
+        questions = []
+        answers = []
+        for group in squad_dict['data']:
+            for passage in group['paragraphs']:
+                context = passage['context']
+                for qa in passage['qas']:
+                    question = qa['question']
+                    for answer in qa['answers']:
+                        contexts.append(context)
+                        questions.append(question)
+                        answers.append(answer)
+
+        return contexts, questions, answers
+
+    train_contexts, train_questions, train_answers = read_squad('squad/train-v2.0.json')
+    val_contexts, val_questions, val_answers = read_squad('squad/dev-v2.0.json')
+
+The contexts and questions are just strings. The answers are dicts containing the subsequence of the passage with the
+correct answer as well as an integer indicating the character at which the answer begins. In order to train a model on
+this data we need (1) the tokenized context/question pairs, and (2) integers indicating at which *token* positions the
+answer begins and ends.
+
+First, let's get the *character* position at which the answer ends in the passage (we are given the starting position).
+Sometimes SQuAD answers are off by one or two characters, so we will also adjust for that.
+
+.. code-block:: python
+
+    def add_end_idx(answers, contexts):
+        for answer, context in zip(answers, contexts):
+            gold_text = answer['text']
+            start_idx = answer['answer_start']
+            end_idx = start_idx + len(gold_text)
+
+            # sometimes squad answers are off by a character or two – fix this
+            if context[start_idx:end_idx] == gold_text:
+                answer['answer_end'] = end_idx
+            elif context[start_idx-1:end_idx-1] == gold_text:
+                answer['answer_start'] = start_idx - 1
+                answer['answer_end'] = end_idx - 1     # When the gold label is off by one character
+            elif context[start_idx-2:end_idx-2] == gold_text:
+                answer['answer_start'] = start_idx - 2
+                answer['answer_end'] = end_idx - 2     # When the gold label is off by two characters
+
+    add_end_idx(train_answers, train_contexts)
+    add_end_idx(val_answers, val_contexts)
+
+Now ``train_answers`` and ``val_answers`` include the character end positions and the corrected start positions. Next,
+let's tokenize our context/question pairs. 🤗 Tokenizers can accept parallel lists of sequences and encode them together
+as sequence pairs.
+
+.. code-block:: python
+
+    from transformers import DistilBertTokenizerFast
+    tokenizer = DistilBertTokenizerFast.from_pretrained('distilbert-base-uncased')
+
+    train_encodings = tokenizer(train_contexts, train_questions, truncation=True, padding=True)
+    val_encodings = tokenizer(val_contexts, val_questions, truncation=True, padding=True)
+
+Next we need to convert our character start/end positions to token start/end positions. When using 🤗 Fast Tokenizers,
+we can use the built in :func:`~transformers.BatchEncoding.char_to_token` method.
+
+.. code-block:: python
+
+    def add_token_positions(encodings, answers):
+        start_positions = []
+        end_positions = []
+        for i in range(len(answers)):
+            start_positions.append(encodings.char_to_token(i, answers[i]['answer_start']))
+            end_positions.append(encodings.char_to_token(i, answers[i]['answer_end'] - 1))
+
+            # if start position is None, the answer passage has been truncated
+            if start_positions[-1] is None:
+                start_positions[-1] = tokenizer.model_max_length
+            if end_positions[-1] is None:
+                end_positions[-1] = tokenizer.model_max_length
+
+        encodings.update({'start_positions': start_positions, 'end_positions': end_positions})
+
+    add_token_positions(train_encodings, train_answers)
+    add_token_positions(val_encodings, val_answers)
+
+Our data is ready. Let's just put it in a PyTorch/TensorFlow dataset so that we can easily use it for training. In
+PyTorch, we define a custom ``Dataset`` class. In TensorFlow, we pass a tuple of ``(inputs_dict, labels_dict)`` to the
+``from_tensor_slices`` method.
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    import torch
+
+    class SquadDataset(torch.utils.data.Dataset):
+        def __init__(self, encodings):
+            self.encodings = encodings
+
+        def __getitem__(self, idx):
+            return {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
+
+        def __len__(self):
+            return len(self.encodings.input_ids)
+
+    train_dataset = SquadDataset(train_encodings)
+    val_dataset = SquadDataset(val_encodings)
+    ## TENSORFLOW CODE
+    import tensorflow as tf
+
+    train_dataset = tf.data.Dataset.from_tensor_slices((
+        {key: train_encodings[key] for key in ['input_ids', 'attention_mask']},
+        {key: train_encodings[key] for key in ['start_positions', 'end_positions']}
+    ))
+    val_dataset = tf.data.Dataset.from_tensor_slices((
+        {key: val_encodings[key] for key in ['input_ids', 'attention_mask']},
+        {key: val_encodings[key] for key in ['start_positions', 'end_positions']}
+    ))
+
+Now we can use a DistilBert model with a QA head for training:
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    from transformers import DistilBertForQuestionAnswering
+    model = DistilBertForQuestionAnswering.from_pretrained("distilbert-base-uncased")
+    ## TENSORFLOW CODE
+    from transformers import TFDistilBertForQuestionAnswering
+    model = TFDistilBertForQuestionAnswering.from_pretrained("distilbert-base-uncased")
+
+
+The data and model are both ready to go. You can train the model with
+:class:`~transformers.Trainer`/:class:`~transformers.TFTrainer` exactly as in the sequence classification example
+above. If using native PyTorch, replace ``labels`` with ``start_positions`` and ``end_positions`` in the training
+example. If using Keras's ``fit``, we need to make a minor modification to handle this example since it involves
+multiple model outputs.
+
+  - :ref:`ft_trainer`
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    from torch.utils.data import DataLoader
+    from transformers import AdamW
+
+    device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
+
+    model.to(device)
+    model.train()
+
+    train_loader = DataLoader(train_dataset, batch_size=16, shuffle=True)
+
+    optim = AdamW(model.parameters(), lr=5e-5)
+
+    for epoch in range(3):
+        for batch in train_loader:
+            optim.zero_grad()
+            input_ids = batch['input_ids'].to(device)
+            attention_mask = batch['attention_mask'].to(device)
+            start_positions = batch['start_positions'].to(device)
+            end_positions = batch['end_positions'].to(device)
+            outputs = model(input_ids, attention_mask=attention_mask, start_positions=start_positions, end_positions=end_positions)
+            loss = outputs[0]
+            loss.backward()
+            optim.step()
+
+    model.eval()
+    ## TENSORFLOW CODE
+    # Keras will expect a tuple when dealing with labels
+    train_dataset = train_dataset.map(lambda x, y: (x, (y['start_positions'], y['end_positions'])))
+
+    # Keras will assign a separate loss for each output and add them together. So we'll just use the standard CE loss
+    # instead of using the built-in model.compute_loss, which expects a dict of outputs and averages the two terms.
+    # Note that this means the loss will be 2x of when using TFTrainer since we're adding instead of averaging them.
+    loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
+    model.distilbert.return_dict = False # if using 🤗 Transformers >3.02, make sure outputs are tuples
+
+    optimizer = tf.keras.optimizers.Adam(learning_rate=5e-5)
+    model.compile(optimizer=optimizer, loss=loss) # can also use any keras loss fn
+    model.fit(train_dataset.shuffle(1000).batch(16), epochs=3, batch_size=16)
+
+.. _resources:
+
+Additional Resources
+-----------------------------------------------------------------------------------------------------------------------
+
+  - `How to train a new language model from scratch using Transformers and Tokenizers
+    <https://huggingface.co/blog/how-to-train>`_. Blog post showing the steps to load in Esperanto data and train a
+    masked language model from scratch.
+  - :doc:`Preprocessing <preprocessing>`. Docs page on data preprocessing.
+  - :doc:`Training <training>`. Docs page on training and fine-tuning.
+
+.. _datasetslib:
+
+Using the 🤗 Datasets & Metrics library
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+This tutorial demonstrates how to read in datasets from various raw text formats and prepare them for training with 🤗
+Transformers so that you can do the same thing with your own custom datasets. However, we recommend users use the `🤗
+Datasets library <https://github.com/huggingface/datasets>`_ for working with the 150+ datasets included in the `hub
+<https://huggingface.co/datasets>`_, including the three datasets used in this tutorial. As a very brief overview, we
+will show how to use the Datasets library to download and prepare the IMDb dataset from the first example,
+:ref:`seq_imdb`.
+
+Start by downloading the dataset:
+
+.. code-block:: python
+
+    from datasets import load_dataset
+    train = load_dataset("imdb", split="train")
+
+Each dataset has multiple columns corresponding to different features. Let's see what our columns are.
+
+.. code-block:: python
+
+    >>> print(train.column_names)
+    ['label', 'text']
+
+Great. Now let's tokenize the text. We can do this using the ``map`` method. We'll also rename the ``label`` column to
+``labels`` to match the model's input arguments.
+
+.. code-block:: python
+
+    train = train.map(lambda batch: tokenizer(batch["text"], truncation=True, padding=True), batched=True)
+    train.rename_column_("label", "labels")
+
+Lastly, we can use the ``set_format`` method to determine which columns and in what data format we want to access
+dataset elements.
+
+.. code-block:: python
+
+    ## PYTORCH CODE
+    >>> train.set_format("torch", columns=["input_ids", "attention_mask", "labels"])
+    >>> {key: val.shape for key, val in train[0].items()})
+    {'labels': torch.Size([]), 'input_ids': torch.Size([512]), 'attention_mask': torch.Size([512])}
+    ## TENSORFLOW CODE
+    >>> train.set_format("tensorflow", columns=["input_ids", "attention_mask", "labels"])
+    >>> {key: val.shape for key, val in train[0].items()})
+    {'labels': TensorShape([]), 'input_ids': TensorShape([512]), 'attention_mask': TensorShape([512])}
+
+We now have a fully-prepared dataset. Check out `the 🤗 Datasets docs
+<https://huggingface.co/docs/datasets/processing.html>`_ for a more thorough introduction.

docs/source/de/_config.py

@@ -1,14 +0,0 @@
-# docstyle-ignore
-INSTALL_CONTENT = """
-# Transformers installation
-! pip install transformers datasets
-# To install from source instead of the last release, comment the command above and uncomment the following one.
-# ! pip install git+https://github.com/huggingface/transformers.git
-"""
-
-notebook_first_cells = [{"type": "code", "content": INSTALL_CONTENT}]
-black_avoid_patterns = {
-    "{processor_class}": "FakeProcessorClass",
-    "{model_class}": "FakeModelClass",
-    "{object_class}": "FakeObjectClass",
-}

docs/source/de/_toctree.yml

@@ -1,22 +0,0 @@
-- sections:
-  - local: index
-    title: 🤗 Transformers
-  - local: quicktour
-    title: Schnellstart
-  - local: installation
-    title: Installation
-  title: Erste Schritte
-- sections:
-  - local: pipeline_tutorial
-    title: Pipelines für Inferenzen
-  - local: autoclass_tutorial
-    title: Laden von vortrainierten Instanzen mit einer AutoClass
-  - local: preprocessing
-    title: Vorverarbeiten
-  - local: training
-    title: Optimierung eines vortrainierten Modells
-  - local: accelerate
-    title: Verteiltes Training mit 🤗 Accelerate
-  - local: model_sharing
-    title: Ein Modell teilen
-  title: Tutorials

docs/source/de/accelerate.mdx

@@ -1,132 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Verteiltes Training mit 🤗 Accelerate
-
-Da die Modelle immer größer werden, hat sich die Parallelität als Strategie zum Trainieren größerer Modelle auf begrenzter Hardware und zur Beschleunigung der Trainingsgeschwindigkeit um mehrere Größenordnungen erwiesen. Bei Hugging Face haben wir die Bibliothek [🤗 Accelerate](https://huggingface.co/docs/accelerate) entwickelt, um Nutzern zu helfen, ein 🤗 Transformers-Modell auf jeder Art von verteiltem Setup zu trainieren, egal ob es sich um mehrere GPUs auf einer Maschine oder mehrere GPUs auf mehreren Maschinen handelt. In diesem Tutorial lernen Sie, wie Sie Ihre native PyTorch-Trainingsschleife anpassen, um das Training in einer verteilten Umgebung zu ermöglichen.
-
-## Einrichtung
-
-Beginnen Sie mit der Installation von 🤗 Accelerate:
-
-```bash
-pip install accelerate
-```
-
-Dann importieren und erstellen Sie ein [`~accelerate.Accelerator`]-Objekt. Der [`~accelerate.Accelerator`] wird automatisch Ihre Art der verteilten Einrichtung erkennen und alle notwendigen Komponenten für das Training initialisieren. Sie müssen Ihr Modell nicht explizit auf einem Gerät platzieren.
-
-```py
->>> from accelerate import Accelerator
-
->>> accelerator = Accelerator()
-```
-
-## Vorbereiten auf die Beschleunigung
-
-Der nächste Schritt ist die Übergabe aller relevanten Trainingsobjekte an die Methode [`~accelerate.Accelerator.prepare`]. Dazu gehören Ihre Trainings- und Evaluierungs-DataLoader, ein Modell und ein Optimierer:
-
-```py
->>> train_dataloader, eval_dataloader, model, optimizer = accelerator.prepare(
-...     train_dataloader, eval_dataloader, model, optimizer
-... )
-```
-
-## Rückwärts
-
-Die letzte Ergänzung besteht darin, das typische `loss.backward()` in der Trainingsschleife durch die 🤗 Accelerate-Methode [`~accelerate.Accelerator.backward`] zu ersetzen:
-
-```py
->>> for epoch in range(num_epochs):
-...     for batch in train_dataloader:
-...         outputs = model(**batch)
-...         loss = outputs.loss
-...         accelerator.backward(loss)
-
-...         optimizer.step()
-...         lr_scheduler.step()
-...         optimizer.zero_grad()
-...         progress_bar.update(1)
-```
-
-Wie Sie im folgenden Code sehen können, müssen Sie nur vier zusätzliche Codezeilen zu Ihrer Trainingsschleife hinzufügen, um verteiltes Training zu ermöglichen!
-
-```diff
-+ from accelerate import Accelerator
-  from transformers import AdamW, AutoModelForSequenceClassification, get_scheduler
-
-+ accelerator = Accelerator()
-
-  model = AutoModelForSequenceClassification.from_pretrained(checkpoint, num_labels=2)
-  optimizer = AdamW(model.parameters(), lr=3e-5)
-
-- device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
-- model.to(device)
-
-+ train_dataloader, eval_dataloader, model, optimizer = accelerator.prepare(
-+     train_dataloader, eval_dataloader, model, optimizer
-+ )
-
-  num_epochs = 3
-  num_training_steps = num_epochs * len(train_dataloader)
-  lr_scheduler = get_scheduler(
-      "linear",
-      optimizer=optimizer,
-      num_warmup_steps=0,
-      num_training_steps=num_training_steps
-  )
-
-  progress_bar = tqdm(range(num_training_steps))
-
-  model.train()
-  for epoch in range(num_epochs):
-      for batch in train_dataloader:
--         batch = {k: v.to(device) for k, v in batch.items()}
-          outputs = model(**batch)
-          loss = outputs.loss
--         loss.backward()
-+         accelerator.backward(loss)
-
-          optimizer.step()
-          lr_scheduler.step()
-          optimizer.zero_grad()
-          progress_bar.update(1)
-```
-
-## Trainieren
-
-Sobald Sie die entsprechenden Codezeilen hinzugefügt haben, starten Sie Ihr Training in einem Skript oder einem Notebook wie Colaboratory.
-
-### Trainieren mit einem Skript
-
-Wenn Sie Ihr Training mit einem Skript durchführen, führen Sie den folgenden Befehl aus, um eine Konfigurationsdatei zu erstellen und zu speichern:
-
-```bash
-accelerate config
-```
-
-Dann starten Sie Ihr Training mit:
-
-```bash
-accelerate launch train.py
-```
-
-### Trainieren mit einem Notebook
-
-🤗 Accelerate kann auch in einem Notebook laufen, wenn Sie planen, die TPUs von Colaboratory zu verwenden. Verpacken Sie den gesamten Code, der für das Training verantwortlich ist, in eine Funktion und übergeben Sie diese an [`~accelerate.notebook_launcher`]:
-
-```py
->>> from accelerate import notebook_launcher
-
->>> notebook_launcher(training_function)
-```
-
-Weitere Informationen über 🤗 Accelerate und seine umfangreichen Funktionen finden Sie in der [Dokumentation](https://huggingface.co/docs/accelerate).

docs/source/de/autoclass_tutorial.mdx

@@ -1,127 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Vortrainierte Instanzen mit einer AutoClass laden
-
-Bei so vielen verschiedenen Transformator-Architekturen kann es eine Herausforderung sein, eine für Ihren Checkpoint zu erstellen. Als Teil der 🤗 Transformers Kernphilosophie, die Bibliothek leicht, einfach und flexibel nutzbar zu machen, leitet eine `AutoClass` automatisch die richtige Architektur aus einem gegebenen Checkpoint ab und lädt sie. Mit der Methode `from_pretrained()` kann man schnell ein vortrainiertes Modell für eine beliebige Architektur laden, so dass man keine Zeit und Ressourcen aufwenden muss, um ein Modell von Grund auf zu trainieren. Die Erstellung dieser Art von Checkpoint-agnostischem Code bedeutet, dass Ihr Code, wenn er für einen Checkpoint funktioniert, auch mit einem anderen Checkpoint funktionieren wird - solange er für eine ähnliche Aufgabe trainiert wurde - selbst wenn die Architektur unterschiedlich ist.
-
-<Tip>
-
-Denken Sie daran, dass sich die Architektur auf das Skelett des Modells bezieht und die Checkpoints die Gewichte für eine bestimmte Architektur sind. Zum Beispiel ist [BERT](https://huggingface.co/bert-base-uncased) eine Architektur, während `bert-base-uncased` ein Checkpoint ist. Modell ist ein allgemeiner Begriff, der entweder Architektur oder Prüfpunkt bedeuten kann.
-
-</Tip>
-
-In dieser Anleitung lernen Sie, wie man:
-
-* Einen vortrainierten Tokenizer lädt.
-* Einen vortrainierten Merkmalsextraktor lädt.
-* Einen vortrainierten Prozessor lädt.
-* Ein vortrainiertes Modell lädt.
-
-## AutoTokenizer
-
-Nahezu jede NLP-Aufgabe beginnt mit einem Tokenizer. Ein Tokenizer wandelt Ihre Eingabe in ein Format um, das vom Modell verarbeitet werden kann.
-
-Laden Sie einen Tokenizer mit [`AutoTokenizer.from_pretrained`]:
-
-```py
->>> from transformers import AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
-```
-
-Dann tokenisieren Sie Ihre Eingabe wie unten gezeigt:
-
-```py
->>> sequence = "In a hole in the ground there lived a hobbit."
->>> print(tokenizer(sequence))
-{'input_ids': [101, 1999, 1037, 4920, 1999, 1996, 2598, 2045, 2973, 1037, 7570, 10322, 4183, 1012, 102], 
- 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
- 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
-```
-
-## AutoFeatureExtractor
-
-Für Audio- und Bildverarbeitungsaufgaben verarbeitet ein Merkmalsextraktor das Audiosignal oder Bild in das richtige Eingabeformat.
-
-Laden Sie einen Merkmalsextraktor mit [`AutoFeatureExtractor.from_pretrained`]:
-
-```py
->>> from transformers import AutoFeatureExtractor
-
->>> feature_extractor = AutoFeatureExtractor.from_pretrained(
-...     "ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition"
-... )
-```
-
-## AutoProcessor
-
-Multimodale Aufgaben erfordern einen Prozessor, der zwei Arten von Vorverarbeitungswerkzeugen kombiniert. Das Modell [LayoutLMV2](model_doc/layoutlmv2) beispielsweise benötigt einen Feature-Extraktor für Bilder und einen Tokenizer für Text; ein Prozessor kombiniert beide.
-
-Laden Sie einen Prozessor mit [`AutoProcessor.from_pretrained`]:
-
-```py
->>> from transformers import AutoProcessor
-
->>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased")
-```
-
-## AutoModel
-
-<frameworkcontent>
-<pt>
-Mit den `AutoModelFor`-Klassen können Sie schließlich ein vortrainiertes Modell für eine bestimmte Aufgabe laden (siehe [hier](model_doc/auto) für eine vollständige Liste der verfügbaren Aufgaben). Laden Sie zum Beispiel ein Modell für die Sequenzklassifikation mit [`AutoModelForSequenceClassification.from_pretrained`]:
-
-```py
->>> from transformers import AutoModelForSequenceClassification
-
->>> model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-Sie können denselben Prüfpunkt problemlos wiederverwenden, um eine Architektur für eine andere Aufgabe zu laden:
-
-```py
->>> from transformers import AutoModelForTokenClassification
-
->>> model = AutoModelForTokenClassification.from_pretrained("distilbert-base-uncased")
-```
-
-<Tip warning={true}>
-
-Für PyTorch-Modelle verwendet die Methode `from_pretrained()` `torch.load()`, die intern `pickle` verwendet und als unsicher bekannt ist. Generell sollte man niemals ein Modell laden, das aus einer nicht vertrauenswürdigen Quelle stammen könnte, oder das manipuliert worden sein könnte. Dieses Sicherheitsrisiko wird für öffentliche Modelle, die auf dem Hugging Face Hub gehostet werden, teilweise gemildert, da diese bei jeder Übertragung [auf Malware](https://huggingface.co/docs/hub/security-malware) gescannt werden. Siehe die [Hub-Dokumentation](https://huggingface.co/docs/hub/security) für Best Practices wie [signierte Commit-Verifizierung](https://huggingface.co/docs/hub/security-gpg#signing-commits-with-gpg) mit GPG.
-
-TensorFlow- und Flax-Checkpoints sind nicht betroffen und können in PyTorch-Architekturen mit den Kwargs `from_tf` und `from_flax` für die Methode `from_pretrained` geladen werden, um dieses Problem zu umgehen.
-
-</Tip>
-
-Im Allgemeinen empfehlen wir die Verwendung der Klasse "AutoTokenizer" und der Klasse "AutoModelFor", um trainierte Instanzen von Modellen zu laden. Dadurch wird sichergestellt, dass Sie jedes Mal die richtige Architektur laden. Im nächsten [Tutorial] (Vorverarbeitung) erfahren Sie, wie Sie Ihren neu geladenen Tokenizer, Feature Extractor und Prozessor verwenden, um einen Datensatz für die Feinabstimmung vorzuverarbeiten.
-</pt>
-<tf>
-Mit den Klassen `TFAutoModelFor` schließlich können Sie ein vortrainiertes Modell für eine bestimmte Aufgabe laden (siehe [hier](model_doc/auto) für eine vollständige Liste der verfügbaren Aufgaben). Laden Sie zum Beispiel ein Modell für die Sequenzklassifikation mit [`TFAutoModelForSequenceClassification.from_pretrained`]:
-
-```py
->>> from transformers import TFAutoModelForSequenceClassification
-
->>> model = TFAutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-Sie können denselben Prüfpunkt problemlos wiederverwenden, um eine Architektur für eine andere Aufgabe zu laden:
-
-```py
->>> from transformers import TFAutoModelForTokenClassification
-
->>> model = TFAutoModelForTokenClassification.from_pretrained("distilbert-base-uncased")
-```
-
-Im Allgemeinen empfehlen wir, die Klasse "AutoTokenizer" und die Klasse "TFAutoModelFor" zu verwenden, um vortrainierte Instanzen von Modellen zu laden. Dadurch wird sichergestellt, dass Sie jedes Mal die richtige Architektur laden. Im nächsten [Tutorial] (Vorverarbeitung) erfahren Sie, wie Sie Ihren neu geladenen Tokenizer, Feature Extractor und Prozessor verwenden, um einen Datensatz für die Feinabstimmung vorzuverarbeiten.
-</tf>
-</frameworkcontent>

docs/source/de/index.mdx

@@ -1,322 +0,0 @@
-<!--Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# 🤗 Transformers
-
-Maschinelles Lernen auf dem neuesten Stand der Technik für PyTorch, TensorFlow und JAX.
-
-🤗 Transformers bietet APIs zum einfachen Herunterladen und Trainieren von vortrainierten Modellen auf dem neuesten Stand der Technik. Die Verwendung von vortrainierten Modellen kann Rechenkosten sparen und den CO2-Fußabdruck reduzieren und Zeit sparen, die für das Training eines Modells von Grund auf benötigt wird. Die Modelle können für verschiedene Modalitäten verwendet werden, wie z. B.:
-
-* 📝 Text: Textklassifizierung, Informationsextrahierung, Beantwortung von Fragen, Zusammenfassung, Übersetzung und Texterstellung in über 100 Sprachen.
-* 🖼️ Bilder: Bildklassifizierung, Objekterkennung und Segmentierung.
-* 🗣️ Audio: Spracherkennung und Audioklassifizierung.
-* 🐙 Multimodal: Beantwortung von Tabellenfragen, optische Zeichenerkennung, Informationsextraktion aus gescannten Dokumenten, Videoklassifizierung und Beantwortung visueller Fragen.
-
-Unsere Bibliothek unterstützt die nahtlose Integration von drei der beliebtesten Deep-Learning-Bibliotheken: [PyTorch](https://pytorch.org/), [TensorFlow](https://www.tensorflow.org/) und [JAX](https://jax.readthedocs.io/en/latest/). Trainieren Sie Ihr Modell in drei Codezeilen in einem Framework und laden Sie es zur Inferenz mit einem anderen.
-
-Jede 🤗 Transformers-Architektur ist in einem eigenständigen Python-Modul definiert, so dass sie leicht für Forschung und Experimente angepasst werden kann.
-
-## Wenn Sie auf der Suche nach individueller Unterstützung durch das Hugging Face-Team sind
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-## Inhalt
-
-Die Dokumentation ist in fünf Teile gegliedert:
-
-- **GET STARTED** enthält eine kurze Tour und Installationsanweisungen, um mit 🤗 Transformers loszulegen.
-- **TUTORIALS** sind ein hervorragender Ausgangspunkt, wenn Sie neu in unserer Bibliothek sind. Dieser Abschnitt hilft Ihnen, die grundlegenden Fähigkeiten zu erlangen, die Sie benötigen, um mit 🤗 Transformers zu arbeiten.
-- **HOW-TO GUIDES** zeigen Ihnen, wie Sie ein bestimmtes Ziel erreichen können, z. B. die Feinabstimmung eines vortrainierten Modells für die Sprachmodellierung oder die Erstellung eines benutzerdefinierten Modellkopfs.
-- **KONZEPTUELLE ANLEITUNGEN** bietet weitere Diskussionen und Erklärungen zu den zugrunde liegenden Konzepten und Ideen hinter Modellen, Aufgaben und der Designphilosophie von 🤗 Transformers. 
-- **API** beschreibt jede Klasse und Funktion, gruppiert in:
-
-  - **MAIN CLASSES** für die Hauptklassen, die die wichtigsten APIs der Bibliothek darstellen.
-  - MODELLE** für die Klassen und Funktionen, die zu jedem in der Bibliothek implementierten Modell gehören.
-  - **INTERNAL HELPERS** für die Klassen und Funktionen, die wir intern verwenden.
-
-Die Bibliothek enthält derzeit JAX-, PyTorch- und TensorFlow-Implementierungen, vortrainierte Modellgewichte, Nutzungsskripte und Konvertierungsprogramme für die folgenden Modelle.
-
-### Unterstütze Modelle
-
-<!--This list is updated automatically from the README with _make fix-copies_. Do not update manually! -->
-
-1. **[ALBERT](model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[BART](model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[Blenderbot](model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLOOM](model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[ByT5](model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[CLIP](model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CodeGen](model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[ConvBERT](model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[CPM](model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CTRL](model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[DeiT](model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DETR](model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DistilBERT](model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
-1. **[DiT](model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[DPR](model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[ELECTRA](model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EncoderDecoder](model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[FlauBERT](model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[Funnel Transformer](model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[GLPN](model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT-2](model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GroupViT](model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[Hubert](model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[ImageGPT](model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[LayoutLM](model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[Longformer](model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MaskFormer](model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[mBART](model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[Megatron-BERT](model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[mLUKE](model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MobileBERT](model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileViT](model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MT5](model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MVP](model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[Nezha](model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nyströmformer](model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OPT](master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[Perceiver IO](model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[PhoBERT](model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[PLBart](model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[ProphetNet](model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[QDQBert](model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoFormer](model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[SegFormer](model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[SEW](model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechToTextTransformer](model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[Swin Transformer](model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[T5](model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[TAPAS](model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Trajectory Transformer](model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[UL2](model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UniSpeech](model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[VAN](model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViTMAE](model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[Wav2Vec2](model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[XGLM](model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLNet](model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-
-
-### Unterstützte Frameworks
-
-Die folgende Tabelle zeigt die derzeitige Unterstützung in der Bibliothek für jedes dieser Modelle, unabhängig davon, ob sie einen Python
-Tokenizer haben (als "langsam" bezeichnet), ein "schneller" Tokenizer, der von der 🤗 Tokenizers Bibliothek unterstützt wird, ob sie Unterstützung in Jax (via
-Flax), PyTorch, und/oder TensorFlow haben.
-
-<!--This table is updated automatically from the auto modules with _make fix-copies_. Do not update manually!-->
-
-|            Model            | Tokenizer slow | Tokenizer fast | PyTorch support | TensorFlow support | Flax Support |
-|:---------------------------:|:--------------:|:--------------:|:---------------:|:------------------:|:------------:|
-|           ALBERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|            BART             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|            BEiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
-|            BERT             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|       Bert Generation       |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           BigBird           |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
-|       BigBird-Pegasus       |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         Blenderbot          |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|       BlenderbotSmall       |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|            BLOOM            |       ❌       |       ✅       |       ✅        |         ❌         |      ❌      |
-|          CamemBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|           CANINE            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            CLIP             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|           CodeGen           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|          ConvBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|          ConvNeXT           |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|             CvT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|        Data2VecAudio        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|        Data2VecText         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|       Data2VecVision        |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|           DeBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|         DeBERTa-v2          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|    Decision Transformer     |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            DeiT             |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|            DETR             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         DistilBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|             DPR             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|             DPT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
-| FairSeq Machine-Translation |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|          FlauBERT           |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|            FLAVA            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            FNet             |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|     Funnel Transformer      |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|            GLPN             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           GPT Neo           |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
-|          GPT NeoX           |       ❌       |       ✅       |       ✅        |         ❌         |      ❌      |
-|            GPT-J            |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
-|          GroupViT           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           Hubert            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|           I-BERT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|          ImageGPT           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|         LayoutLMv2          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|         LayoutLMv3          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|             LED             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|            LeViT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         Longformer          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|           LongT5            |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
-|            LUKE             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           LXMERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|           M-CTC-T           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           M2M100            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           Marian            |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
-|         MaskFormer          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            mBART            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|        Megatron-BERT        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         MobileBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|          MobileViT          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            MPNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|             MT5             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|             MVP             |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|            Nezha            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|        Nyströmformer        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         OpenAI GPT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|             OPT             |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
-|           OWL-ViT           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           Pegasus           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|          Perceiver          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           PLBart            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         PoolFormer          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         ProphetNet          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|           QDQBert           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|             RAG             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|            REALM            |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|          Reformer           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|           RegNet            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|           RemBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|           ResNet            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|          RetriBERT          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|          RoFormer           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|          SegFormer          |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|             SEW             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            SEW-D            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|   Speech Encoder decoder    |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
-|         Speech2Text         |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|        Speech2Text2         |       ✅       |       ❌       |       ❌        |         ❌         |      ❌      |
-|          Splinter           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|         SqueezeBERT         |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-|      Swin Transformer       |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|     Swin Transformer V2     |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|            TAPAS            |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|   Trajectory Transformer    |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|            TrOCR            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|          UniSpeech          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|        UniSpeechSat         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|             VAN             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|          VideoMAE           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            ViLT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|   Vision Encoder decoder    |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
-|    VisionTextDualEncoder    |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
-|         VisualBERT          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|             ViT             |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
-|           ViTMAE            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
-|          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
-|     Wav2Vec2-Conformer      |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            WavLM            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            XGLM             |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
-|             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|       XLM-ProphetNet        |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-|         XLM-RoBERTa         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-|       XLM-RoBERTa-XL        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            XLNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-|            YOLOS            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|            YOSO             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-
-<!-- End table-->

docs/source/de/installation.mdx

@@ -1,246 +0,0 @@
-<!---
-Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
--->
-
-# Installation
-
-Installieren Sie 🤗 Transformers für die Deep-Learning-Bibliothek, mit der Sie arbeiten, richten Sie Ihren Cache ein und konfigurieren Sie 🤗 Transformers optional für den Offline-Betrieb.
-
-🤗 Transformers wurde unter Python 3.6+, PyTorch 1.1.0+, TensorFlow 2.0+, und Flax getestet. Folgen Sie den Installationsanweisungen unten für die von Ihnen verwendete Deep-Learning-Bibliothek:
-
-* [PyTorch](https://pytorch.org/get-started/locally/) installation instructions.
-* [TensorFlow 2.0](https://www.tensorflow.org/install/pip) installation instructions.
-* [Flax](https://flax.readthedocs.io/en/latest/) installation instructions.
-
-## Installation mit pip
-
-Sie sollten 🤗 Transformers in einer [virtuellen Umgebung](https://docs.python.org/3/library/venv.html) installieren. Wenn Sie mit virtuellen Python-Umgebungen nicht vertraut sind, werfen Sie einen Blick auf diese [Anleitung](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/). Eine virtuelle Umgebung macht es einfacher, verschiedene Projekte zu verwalten und Kompatibilitätsprobleme zwischen Abhängigkeiten zu vermeiden.
-
-Beginnen wir mit der Erstellung einer virtuellen Umgebung in Ihrem Projektverzeichnis:
-
-
-```bash
-python -m venv .env
-```
-
-Aktivieren wir die virtuelle Umgebung. Unter Linux und MacOs:
-
-```bash
-source .env/bin/activate
-```
-Aktivieren wir die virtuelle Umgebung unter Windows
-
-```bash
-.env/Scripts/activate
-```
-
-Jetzt können wir die 🤗 Transformers mit dem folgenden Befehl installieren:
-
-```bash
-pip install transformers
-```
-
-Bei reiner CPU-Unterstützung können wir 🤗 Transformers und eine Deep-Learning-Bibliothek bequem in einer Zeile installieren. Installieren wir zum Beispiel 🤗 Transformers und PyTorch mit:
-
-```bash
-pip install transformers[torch]
-```
-
-🤗 Transformers und TensorFlow 2.0:
-
-```bash
-pip install transformers[tf-cpu]
-```
-
-🤗 Transformers und Flax:
-
-```bash
-pip install transformers[flax]
-```
-
-Überprüfen wir abschließend, ob 🤗 Transformers ordnungsgemäß installiert wurde, indem wir den folgenden Befehl ausführen. Es wird ein vortrainiertes Modell heruntergeladen:
-
-```bash
-python -c "from transformers import pipeline; print(pipeline('sentiment-analysis')('we love you'))"
-```
-
-Dann wird die Kategorie und die Wahrscheinlichkeit ausgegeben:
-
-```bash
-[{'label': 'POSITIVE', 'score': 0.9998704791069031}]
-```
-
-## Installation aus dem Code
-
-Installieren wir 🤗 Transformers aus dem Quellcode mit dem folgenden Befehl:
-
-```bash
-pip install git+https://github.com/huggingface/transformers
-```
-
-Dieser Befehl installiert die aktuelle `main` Version und nicht die neueste `stable` Version. Die `main`-Version ist nützlich, um mit den neuesten Entwicklungen Schritt zu halten. Zum Beispiel, wenn ein Fehler seit der letzten offiziellen Version behoben wurde, aber eine neue Version noch nicht veröffentlicht wurde. Das bedeutet jedoch, dass die "Hauptversion" nicht immer stabil ist. Wir bemühen uns, die Hauptversion einsatzbereit zu halten, und die meisten Probleme werden normalerweise innerhalb weniger Stunden oder eines Tages behoben. Wenn Sie auf ein Problem stoßen, öffnen Sie bitte ein [Issue] (https://github.com/huggingface/transformers/issues), damit wir es noch schneller beheben können!
-
-Überprüfen wir, ob 🤗 Transformers richtig installiert wurde, indem Sie den folgenden Befehl ausführen:
-
-
-```bash
-python -c "from transformers import pipeline; print(pipeline('sentiment-analysis')('I love you'))"
-```
-
-## Editierbare Installation
-
-Sie benötigen eine bearbeitbare Installation, wenn Sie:
-
-* die "Haupt"-Version des Quellcodes verwenden möchten.
-* Zu 🤗 Transformers beitragen und Änderungen am Code testen wollen.
-
-Klonen Sie das Repository und installieren 🤗 Transformers mit den folgenden Befehlen:
-
-```bash
-git clone https://github.com/huggingface/transformers.git
-cd transformers
-pip install -e .
-```
-
-Diese Befehle verknüpfen den Ordner, in den Sie das Repository geklont haben, mit den Pfaden Ihrer Python-Bibliotheken. Python wird nun in dem Ordner suchen, in den Sie geklont haben, zusätzlich zu den normalen Bibliothekspfaden. Wenn zum Beispiel Ihre Python-Pakete normalerweise in `~/anaconda3/envs/main/lib/python3.7/site-packages/` installiert sind, wird Python auch den Ordner durchsuchen, in den Sie geklont haben: `~/transformers/`.
-
-
-<Tip warning={true}>
-
-Sie müssen den Ordner `transformers` behalten, wenn Sie die Bibliothek weiter verwenden wollen.
-
-</Tip>
-
-Jetzt können Sie Ihren Klon mit dem folgenden Befehl ganz einfach auf die neueste Version von 🤗 Transformers aktualisieren:
-
-
-```bash
-cd ~/transformers/
-git pull
-```
-
-Ihre Python-Umgebung wird beim nächsten Ausführen die `main`-Version von 🤗 Transformers finden.
-
-## Installation mit conda
-
-Installation von dem conda Kanal `huggingface`:
-
-```bash
-conda install -c huggingface transformers
-```
-
-## Cache Einrichtung
-
-Vorgefertigte Modelle werden heruntergeladen und lokal zwischengespeichert unter: `~/.cache/huggingface/hub`. Dies ist das Standardverzeichnis, das durch die Shell-Umgebungsvariable "TRANSFORMERS_CACHE" vorgegeben ist. Unter Windows wird das Standardverzeichnis durch `C:\Benutzer\Benutzername\.cache\huggingface\hub` angegeben. Sie können die unten aufgeführten Shell-Umgebungsvariablen - in der Reihenfolge ihrer Priorität - ändern, um ein anderes Cache-Verzeichnis anzugeben:
-
-1. Shell-Umgebungsvariable (Standard): `HUGGINGFACE_HUB_CACHE` oder `TRANSFORMERS_CACHE`.
-2. Shell-Umgebungsvariable: `HF_HOME`.
-3. Shell-Umgebungsvariable: `XDG_CACHE_HOME` + `/huggingface`.
-
-
-<Tip>
-
-Transformers verwendet die Shell-Umgebungsvariablen `PYTORCH_TRANSFORMERS_CACHE` oder `PYTORCH_PRETRAINED_BERT_CACHE`, wenn Sie von einer früheren Iteration dieser Bibliothek kommen und diese Umgebungsvariablen gesetzt haben, sofern Sie nicht die Shell-Umgebungsvariable `TRANSFORMERS_CACHE` angeben.
-  
-</Tip>
-
-## Offline Modus
-
-Transformers ist in der Lage, in einer Firewall- oder Offline-Umgebung zu laufen, indem es nur lokale Dateien verwendet. Setzen Sie die Umgebungsvariable `TRANSFORMERS_OFFLINE=1`, um dieses Verhalten zu aktivieren.
-
-<Tip>
-
-Fügen sie [🤗 Datasets](https://huggingface.co/docs/datasets/) zu Ihrem Offline-Trainingsworkflow hinzufügen, indem Sie die Umgebungsvariable `HF_DATASETS_OFFLINE=1` setzen.
-
-</Tip>
-
-So würden Sie beispielsweise ein Programm in einem normalen Netzwerk mit einer Firewall für externe Instanzen mit dem folgenden Befehl ausführen:
-
-```bash
-python examples/pytorch/translation/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
-```
-
-Führen Sie das gleiche Programm in einer Offline-Instanz mit aus:
-
-```bash
-HF_DATASETS_OFFLINE=1 TRANSFORMERS_OFFLINE=1 \
-python examples/pytorch/translation/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
-```
-
-Das Skript sollte nun laufen, ohne sich aufzuhängen oder eine Zeitüberschreitung abzuwarten, da es weiß, dass es nur nach lokalen Dateien suchen soll.
-
-
-### Abrufen von Modellen und Tokenizern zur Offline-Verwendung
-
-Eine andere Möglichkeit, 🤗 Transformers offline zu verwenden, besteht darin, die Dateien im Voraus herunterzuladen und dann auf ihren lokalen Pfad zu verweisen, wenn Sie sie offline verwenden müssen. Es gibt drei Möglichkeiten, dies zu tun:
-
-* Laden Sie eine Datei über die Benutzeroberfläche des [Model Hub](https://huggingface.co/models) herunter, indem Sie auf das ↓-Symbol klicken.
-
-    ![download-icon](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/download-icon.png)
-
-* Verwenden Sie den [PreTrainedModel.from_pretrained] und [PreTrainedModel.save_pretrained] Workflow:
-
-    1. Laden Sie Ihre Dateien im Voraus mit [`PreTrainedModel.from_pretrained`] herunter:
-
-    ```py
-    >>> from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-
-    >>> tokenizer = AutoTokenizer.from_pretrained("bigscience/T0_3B")
-    >>> model = AutoModelForSeq2SeqLM.from_pretrained("bigscience/T0_3B")
-    ```
-
-    2. Speichern Sie Ihre Dateien in einem bestimmten Verzeichnis mit [`PreTrainedModel.save_pretrained`]:
-
-    ```py
-    >>> tokenizer.save_pretrained("./your/path/bigscience_t0")
-    >>> model.save_pretrained("./your/path/bigscience_t0")
-    ```
-
-    3. Wenn Sie nun offline sind, laden Sie Ihre Dateien mit [`PreTrainedModel.from_pretrained`] aus dem bestimmten Verzeichnis:
-
-    ```py
-    >>> tokenizer = AutoTokenizer.from_pretrained("./your/path/bigscience_t0")
-    >>> model = AutoModel.from_pretrained("./your/path/bigscience_t0")
-    ```
-
-* Programmatisches Herunterladen von Dateien mit der [huggingface_hub](https://github.com/huggingface/huggingface_hub/tree/main/src/huggingface_hub) Bibliothek:
-
-    1. Installieren Sie die "huggingface_hub"-Bibliothek in Ihrer virtuellen Umgebung:
-
-    ```bash
-    python -m pip install huggingface_hub
-    ```
-
-    2. Verwenden Sie die Funktion [`hf_hub_download`](https://huggingface.co/docs/hub/adding-a-library#download-files-from-the-hub), um eine Datei in einen bestimmten Pfad herunterzuladen. Der folgende Befehl lädt zum Beispiel die Datei "config.json" aus dem Modell [T0](https://huggingface.co/bigscience/T0_3B) in den gewünschten Pfad herunter:
-
-    ```py
-    >>> from huggingface_hub import hf_hub_download
-
-    >>> hf_hub_download(repo_id="bigscience/T0_3B", filename="config.json", cache_dir="./your/path/bigscience_t0")
-    ```
-
-Sobald Ihre Datei heruntergeladen und lokal zwischengespeichert ist, geben Sie den lokalen Pfad an, um sie zu laden und zu verwenden:
-
-```py
->>> from transformers import AutoConfig
-
->>> config = AutoConfig.from_pretrained("./your/path/bigscience_t0/config.json")
-```
-
-<Tip>
-
-Weitere Informationen zum Herunterladen von Dateien, die auf dem Hub gespeichert sind, finden Sie im Abschnitt [Wie man Dateien vom Hub herunterlädt] (https://huggingface.co/docs/hub/how-to-downstream).
-  
-</Tip>

docs/source/de/model_sharing.mdx

@@ -1,228 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Ein Modell teilen
-
-Die letzten beiden Tutorials haben gezeigt, wie man ein Modell mit PyTorch, Keras und 🤗 Accelerate für verteilte Setups feinabstimmen kann. Der nächste Schritt besteht darin, Ihr Modell mit der Community zu teilen! Bei Hugging Face glauben wir an den offenen Austausch von Wissen und Ressourcen, um künstliche Intelligenz für alle zu demokratisieren. Wir ermutigen Sie, Ihr Modell mit der Community zu teilen, um anderen zu helfen, Zeit und Ressourcen zu sparen.
-
-In diesem Tutorial lernen Sie zwei Methoden kennen, wie Sie ein trainiertes oder verfeinertes Modell auf dem [Model Hub](https://huggingface.co/models) teilen können:
-
-- Programmgesteuertes Übertragen Ihrer Dateien auf den Hub.
-- Ziehen Sie Ihre Dateien per Drag-and-Drop über die Weboberfläche in den Hub.
-
-<iframe width="560" height="315" src="https://www.youtube.com/embed/XvSGPZFEjDY" title="YouTube video player"
-frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope;
-picture-in-picture" allowfullscreen></iframe>
-
-<Tip>
-
-Um ein Modell mit der Öffentlichkeit zu teilen, benötigen Sie ein Konto auf [huggingface.co](https://huggingface.co/join). Sie können auch einer bestehenden Organisation beitreten oder eine neue Organisation gründen.
-
-</Tip>
-
-## Repository-Funktionen
-
-Jedes Repository im Model Hub verhält sich wie ein typisches GitHub-Repository. Unsere Repositorys bieten Versionierung, Commit-Historie und die Möglichkeit, Unterschiede zu visualisieren.
-
-Die integrierte Versionierung des Model Hub basiert auf Git und [git-lfs](https://git-lfs.github.com/). Mit anderen Worten: Sie können ein Modell als ein Repository behandeln, was eine bessere Zugriffskontrolle und Skalierbarkeit ermöglicht. Die Versionskontrolle ermöglicht *Revisionen*, eine Methode zum Anheften einer bestimmten Version eines Modells mit einem Commit-Hash, Tag oder Branch.
-
-Folglich können Sie eine bestimmte Modellversion mit dem Parameter "Revision" laden:
-
-```py
->>> model = AutoModel.from_pretrained(
-...     "julien-c/EsperBERTo-small", revision="v2.0.1"  # tag name, or branch name, or commit hash
-... )
-```
-
-Dateien lassen sich auch in einem Repository leicht bearbeiten, und Sie können die Commit-Historie sowie die Unterschiede einsehen:
-
-![vis_diff](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/vis_diff.png)
-
-## Einrichtung
-
-Bevor Sie ein Modell für den Hub freigeben, benötigen Sie Ihre Hugging Face-Anmeldedaten. Wenn Sie Zugang zu einem Terminal haben, führen Sie den folgenden Befehl in der virtuellen Umgebung aus, in der 🤗 Transformers installiert ist. Dadurch werden Ihre Zugangsdaten in Ihrem Hugging Face-Cache-Ordner (standardmäßig `~/.cache/`) gespeichert:
-
-```bash
-huggingface-cli login
-```
-
-Wenn Sie ein Notebook wie Jupyter oder Colaboratory verwenden, stellen Sie sicher, dass Sie die [`huggingface_hub`](https://huggingface.co/docs/hub/adding-a-library) Bibliothek installiert haben. Diese Bibliothek ermöglicht Ihnen die programmatische Interaktion mit dem Hub.
-
-```bash
-pip install huggingface_hub
-```
-
-Verwenden Sie dann `notebook_login`, um sich beim Hub anzumelden, und folgen Sie dem Link [hier](https://huggingface.co/settings/token), um ein Token für die Anmeldung zu generieren:
-
-```py
->>> from huggingface_hub import notebook_login
-
->>> notebook_login()
-```
-
-## Ein Modell für alle Frameworks konvertieren
-
-Um sicherzustellen, dass Ihr Modell von jemandem verwendet werden kann, der mit einem anderen Framework arbeitet, empfehlen wir Ihnen, Ihr Modell sowohl mit PyTorch- als auch mit TensorFlow-Checkpoints zu konvertieren und hochzuladen. Während Benutzer immer noch in der Lage sind, Ihr Modell von einem anderen Framework zu laden, wenn Sie diesen Schritt überspringen, wird es langsamer sein, weil 🤗 Transformers den Checkpoint on-the-fly konvertieren müssen.
-
-Die Konvertierung eines Checkpoints für ein anderes Framework ist einfach. Stellen Sie sicher, dass Sie PyTorch und TensorFlow installiert haben (siehe [hier](installation) für Installationsanweisungen), und finden Sie dann das spezifische Modell für Ihre Aufgabe in dem anderen Framework. 
-
-<frameworkcontent>
-<pt>
-Geben Sie `from_tf=True` an, um einen Prüfpunkt von TensorFlow nach PyTorch zu konvertieren:
-
-```py
->>> pt_model = DistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_tf=True)
->>> pt_model.save_pretrained("path/to/awesome-name-you-picked")
-```
-</pt>
-<tf>
-Geben Sie `from_pt=True` an, um einen Prüfpunkt von PyTorch nach TensorFlow zu konvertieren:
-
-```py
->>> tf_model = TFDistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_pt=True)
-```
-
-Dann können Sie Ihr neues TensorFlow-Modell mit seinem neuen Checkpoint speichern:
-
-```py
->>> tf_model.save_pretrained("path/to/awesome-name-you-picked")
-```
-</tf>
-<jax>
-Wenn ein Modell in Flax verfügbar ist, können Sie auch einen Kontrollpunkt von PyTorch nach Flax konvertieren:
-
-```py
->>> flax_model = FlaxDistilBertForSequenceClassification.from_pretrained(
-...     "path/to/awesome-name-you-picked", from_pt=True
-... )
-```
-</jax>
-</frameworkcontent>
-
-## Ein Modell während des Trainings hochladen
-
-<frameworkcontent>
-<pt>
-<Youtube id="Z1-XMy-GNLQ"/>
-
-Die Weitergabe eines Modells an den Hub ist so einfach wie das Hinzufügen eines zusätzlichen Parameters oder Rückrufs. Erinnern Sie sich an das [Feinabstimmungs-Tutorial](training), in der Klasse [`TrainingArguments`] geben Sie Hyperparameter und zusätzliche Trainingsoptionen an. Eine dieser Trainingsoptionen beinhaltet die Möglichkeit, ein Modell direkt an den Hub zu pushen. Setzen Sie `push_to_hub=True` in Ihrer [`TrainingArguments`]:
-
-```py
->>> training_args = TrainingArguments(output_dir="my-awesome-model", push_to_hub=True)
-```
-
-Übergeben Sie Ihre Trainingsargumente wie gewohnt an [`Trainer`]:
-
-```py
->>> trainer = Trainer(
-...     model=model,
-...     args=training_args,
-...     train_dataset=small_train_dataset,
-...     eval_dataset=small_eval_dataset,
-...     compute_metrics=compute_metrics,
-... )
-```
-
-Nach der Feinabstimmung Ihres Modells rufen Sie [`~transformers.Trainer.push_to_hub`] auf [`Trainer`] auf, um das trainierte Modell an den Hub zu übertragen. Transformers fügt sogar automatisch Trainings-Hyperparameter, Trainingsergebnisse und Framework-Versionen zu Ihrer Modellkarte hinzu!
-
-```py
->>> trainer.push_to_hub()
-```
-</pt>
-<tf>
-Geben Sie ein Modell mit [`PushToHubCallback`] an den Hub weiter. In der [`PushToHubCallback`] Funktion, fügen Sie hinzu:
-
-- Ein Ausgabeverzeichnis für Ihr Modell.
-- Einen Tokenizer.
-- Die `hub_model_id`, die Ihr Hub-Benutzername und Modellname ist.
-
-```py
->>> from transformers.keras.callbacks import PushToHubCallback
-
->>> push_to_hub_callback = PushToHubCallback(
-...     output_dir="./your_model_save_path", tokenizer=tokenizer, hub_model_id="your-username/my-awesome-model"
-... )
-```
-
-Fügen Sie den Callback zu [`fit`](https://keras.io/api/models/model_training_apis/) hinzu, und 🤗 Transformers wird das trainierte Modell an den Hub weiterleiten:
-
-```py
->>> model.fit(tf_train_dataset, validation_data=tf_validation_dataset, epochs=3, callbacks=push_to_hub_callback)
-```
-</tf>
-</frameworkcontent>
-
-## Verwenden Sie die Funktion `push_to_hub`.
-
-Sie können `push_to_hub` auch direkt für Ihr Modell aufrufen, um es in den Hub hochzuladen.
-
-Geben Sie den Namen Ihres Modells in "push_to_hub" an:
-
-```py
->>> pt_model.push_to_hub("my-awesome-model")
-```
-
-Dadurch wird ein Repository unter Ihrem Benutzernamen mit dem Modellnamen `my-awesome-model` erstellt. Benutzer können nun Ihr Modell mit der Funktion `from_pretrained` laden:
-
-```py
->>> from transformers import AutoModel
-
->>> model = AutoModel.from_pretrained("your_username/my-awesome-model")
-```
-
-Wenn Sie zu einer Organisation gehören und Ihr Modell stattdessen unter dem Namen der Organisation pushen wollen, fügen Sie diesen einfach zur `repo_id` hinzu:
-
-```py
->>> pt_model.push_to_hub("my-awesome-org/my-awesome-model")
-```
-
-Die Funktion "push_to_hub" kann auch verwendet werden, um andere Dateien zu einem Modell-Repository hinzuzufügen. Zum Beispiel kann man einen Tokenizer zu einem Modell-Repository hinzufügen:
-
-```py
->>> tokenizer.push_to_hub("my-awesome-model")
-```
-
-Oder vielleicht möchten Sie die TensorFlow-Version Ihres fein abgestimmten PyTorch-Modells hinzufügen:
-
-```py
->>> tf_model.push_to_hub("my-awesome-model")
-```
-
-Wenn Sie nun zu Ihrem Hugging Face-Profil navigieren, sollten Sie Ihr neu erstelltes Modell-Repository sehen. Wenn Sie auf die Registerkarte **Dateien** klicken, werden alle Dateien angezeigt, die Sie in das Repository hochgeladen haben.
-
-Weitere Einzelheiten zum Erstellen und Hochladen von Dateien in ein Repository finden Sie in der Hub-Dokumentation [hier](https://huggingface.co/docs/hub/how-to-upstream).
-
-## Hochladen mit der Weboberfläche
-
-Benutzer, die einen no-code Ansatz bevorzugen, können ein Modell über das Webinterface des Hubs hochladen. Besuchen Sie [huggingface.co/new](https://huggingface.co/new) um ein neues Repository zu erstellen:
-
-![new_model_repo](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/new_model_repo.png)
-
-Fügen Sie von hier aus einige Informationen über Ihr Modell hinzu:
-
-- Wählen Sie den **Besitzer** des Repositorys. Dies können Sie selbst oder eine der Organisationen sein, denen Sie angehören.
-- Wählen Sie einen Namen für Ihr Modell, der auch der Name des Repositorys sein wird.
-- Wählen Sie, ob Ihr Modell öffentlich oder privat ist.
-- Geben Sie die Lizenzverwendung für Ihr Modell an.
-
-Klicken Sie nun auf die Registerkarte **Dateien** und klicken Sie auf die Schaltfläche **Datei hinzufügen**, um eine neue Datei in Ihr Repository hochzuladen. Ziehen Sie dann eine Datei per Drag-and-Drop hoch und fügen Sie eine Übergabemeldung hinzu.
-
-![upload_file](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/upload_file.png)
-
-## Hinzufügen einer Modellkarte
-
-Um sicherzustellen, dass die Benutzer die Fähigkeiten, Grenzen, möglichen Verzerrungen und ethischen Aspekte Ihres Modells verstehen, fügen Sie bitte eine Modellkarte zu Ihrem Repository hinzu. Die Modellkarte wird in der Datei `README.md` definiert. Sie können eine Modellkarte hinzufügen, indem Sie:
-
-* Manuelles Erstellen und Hochladen einer "README.md"-Datei.
-* Klicken Sie auf die Schaltfläche **Modellkarte bearbeiten** in Ihrem Modell-Repository.
-
-Werfen Sie einen Blick auf die DistilBert [model card](https://huggingface.co/distilbert-base-uncased) als gutes Beispiel für die Art von Informationen, die eine Modellkarte enthalten sollte. Weitere Details über andere Optionen, die Sie in der Datei "README.md" einstellen können, wie z.B. den Kohlenstoff-Fußabdruck eines Modells oder Beispiele für Widgets, finden Sie in der Dokumentation [hier](https://huggingface.co/docs/hub/models-cards).

docs/source/de/pipeline_tutorial.mdx

@@ -1,171 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Pipelines für Inferenzen
-
-Die [`pipeline`] macht es einfach, jedes beliebige Modell aus dem [Hub](https://huggingface.co/models) für die Inferenz auf jede Sprache, Computer Vision, Sprache und multimodale Aufgaben zu verwenden. Selbst wenn Sie keine Erfahrung mit einer bestimmten Modalität haben oder nicht mit dem zugrundeliegenden Code hinter den Modellen vertraut sind, können Sie sie mit der [`pipeline`] für Inferenzen verwenden! In diesem Beispiel lernen Sie, wie:
-
-* Eine [`pipeline`] für Inferenz zu verwenden.
-* Einen bestimmten Tokenizer oder ein bestimmtes Modell zu verwenden.
-* Eine [`pipeline`] für Audio-, Vision- und multimodale Aufgaben zu verwenden.
-
-<Tip>
-
-Eine vollständige Liste der unterstützten Aufgaben und verfügbaren Parameter finden Sie in der [`pipeline`]-Dokumentation.
-
-</Tip>
-
-## Verwendung von Pipelines
-
-Obwohl jede Aufgabe eine zugehörige [`pipeline`] hat, ist es einfacher, die allgemeine [`pipeline`]-Abstraktion zu verwenden, die alle aufgabenspezifischen Pipelines enthält. Die [`pipeline`] lädt automatisch ein Standardmodell und eine Vorverarbeitungsklasse, die für Ihre Aufgabe inferenzfähig ist.
-
-1. Beginnen Sie mit der Erstellung einer [`pipeline`] und geben Sie eine Inferenzaufgabe an:
-
-```py
->>> from transformers import pipeline
-
->>> generator = pipeline(task="text-generation")
-```
-
-2. Übergeben Sie Ihren Eingabetext an die [`pipeline`]:
-
-```py
->>> generator(
-...     "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone"
-... )  # doctest: +SKIP
-[{'generated_text': 'Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone, Seven for the Iron-priests at the door to the east, and thirteen for the Lord Kings at the end of the mountain'}]
-```
-
-Wenn Sie mehr als eine Eingabe haben, übergeben Sie die Eingabe als Liste:
-
-```py
->>> generator(
-...     [
-...         "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone",
-...         "Nine for Mortal Men, doomed to die, One for the Dark Lord on his dark throne",
-...     ]
-... )  # doctest: +SKIP
-```
-
-Alle zusätzlichen Parameter für Ihre Aufgabe können auch in die [`pipeline`] aufgenommen werden. Die Aufgabe `Text-Generierung` hat eine [`~generation_utils.GenerationMixin.generate`]-Methode mit mehreren Parametern zur Steuerung der Ausgabe. Wenn Sie zum Beispiel mehr als eine Ausgabe erzeugen wollen, setzen Sie den Parameter `num_return_sequences`:
-
-```py
->>> generator(
-...     "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone",
-...     num_return_sequences=2,
-... )  # doctest: +SKIP
-```
-
-### Wählen Sie ein Modell und einen Tokenizer
-
-Die [`pipeline`] akzeptiert jedes Modell aus dem [Hub] (https://huggingface.co/models). Auf dem Hub gibt es Tags, mit denen Sie nach einem Modell filtern können, das Sie für Ihre Aufgabe verwenden möchten. Sobald Sie ein passendes Modell ausgewählt haben, laden Sie es mit der entsprechenden `AutoModelFor` und [`AutoTokenizer`] Klasse. Laden Sie zum Beispiel die Klasse [`AutoModelForCausalLM`] für eine kausale Sprachmodellierungsaufgabe:
-
-```py
->>> from transformers import AutoTokenizer, AutoModelForCausalLM
-
->>> tokenizer = AutoTokenizer.from_pretrained("distilgpt2")
->>> model = AutoModelForCausalLM.from_pretrained("distilgpt2")
-```
-
-Erstellen Sie eine [`pipeline`] für Ihre Aufgabe, und geben Sie das Modell und den Tokenizer an, die Sie geladen haben:
-
-```py
->>> from transformers import pipeline
-
->>> generator = pipeline(task="text-generation", model=model, tokenizer=tokenizer)
-```
-
-Übergeben Sie Ihren Eingabetext an die [`pipeline`] , um einen Text zu erzeugen:
-
-```py
->>> generator(
-...     "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone"
-... )  # doctest: +SKIP
-[{'generated_text': 'Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone, Seven for the Dragon-lords (for them to rule in a world ruled by their rulers, and all who live within the realm'}]
-```
-
-## Audio-Pipeline
-
-Die [`pipeline`] unterstützt auch Audioaufgaben wie Audioklassifizierung und automatische Spracherkennung.
-
-Lassen Sie uns zum Beispiel die Emotion in diesem Audioclip klassifizieren:
-
-```py
->>> from datasets import load_dataset
->>> import torch
-
->>> torch.manual_seed(42)  # doctest: +IGNORE_RESULT
->>> ds = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
->>> audio_file = ds[0]["audio"]["path"]
-```
-
-Finden Sie ein [Audioklassifikation](https://huggingface.co/models?pipeline_tag=audio-classification) Modell auf dem Model Hub für Emotionserkennung und laden Sie es in die [`pipeline`]:
-
-```py
->>> from transformers import pipeline
-
->>> audio_classifier = pipeline(
-...     task="audio-classification", model="ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition"
-... )
-```
-
-Übergeben Sie die Audiodatei an die [`pipeline`]:
-
-```py
->>> preds = audio_classifier(audio_file)
->>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
->>> preds
-[{'score': 0.1315, 'label': 'calm'}, {'score': 0.1307, 'label': 'neutral'}, {'score': 0.1274, 'label': 'sad'}, {'score': 0.1261, 'label': 'fearful'}, {'score': 0.1242, 'label': 'happy'}]
-```
-
-## Bildverarbeitungs-Pipeline
-
-Die Verwendung einer [`pipeline`] für Bildverarbeitungsaufgaben ist praktisch identisch.
-
-Geben Sie Ihre Aufgabe an und übergeben Sie Ihr Bild an den Klassifikator. Das Bild kann ein Link oder ein lokaler Pfad zu dem Bild sein. Zum Beispiel: Welche Katzenart ist unten abgebildet?
-
-![pipeline-cat-chonk](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg)
-
-```py
->>> from transformers import pipeline
-
->>> vision_classifier = pipeline(task="image-classification")
->>> preds = vision_classifier(
-...     images="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"
-... )
->>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
->>> preds
-[{'score': 0.4335, 'label': 'lynx, catamount'}, {'score': 0.0348, 'label': 'cougar, puma, catamount, mountain lion, painter, panther, Felis concolor'}, {'score': 0.0324, 'label': 'snow leopard, ounce, Panthera uncia'}, {'score': 0.0239, 'label': 'Egyptian cat'}, {'score': 0.0229, 'label': 'tiger cat'}]
-```
-
-## Multimodale Pipeline
-
-Die [`pipeline`] unterstützt mehr als eine Modalität. Eine Aufgabe zur Beantwortung visueller Fragen (VQA) kombiniert zum Beispiel Text und Bild. Verwenden Sie einen beliebigen Bildlink und eine Frage, die Sie zu dem Bild stellen möchten. Das Bild kann eine URL oder ein lokaler Pfad zu dem Bild sein.
-
-Wenn Sie zum Beispiel das gleiche Bild wie in der obigen Vision-Pipeline verwenden:
-
-```py
->>> image = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"
->>> question = "Where is the cat?"
-```
-
-Erstellen Sie eine Pipeline für "vqa" und übergeben Sie ihr das Bild und die Frage:
-
-```py
->>> from transformers import pipeline
-
->>> vqa = pipeline(task="vqa")
->>> preds = vqa(image=image, question=question)
->>> preds = [{"score": round(pred["score"], 4), "answer": pred["answer"]} for pred in preds]
->>> preds
-[{'score': 0.9112, 'answer': 'snow'}, {'score': 0.8796, 'answer': 'in snow'}, {'score': 0.6717, 'answer': 'outside'}, {'score': 0.0291, 'answer': 'on ground'}, {'score': 0.027, 'answer': 'ground'}]
-```

docs/source/de/preprocessing.mdx

@@ -1,502 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Vorverarbeiten
-
-[[open-in-colab]]
-
-Bevor Sie Ihre Daten in einem Modell verwenden können, müssen die Daten in ein für das Modell akzeptables Format gebracht werden. Ein Modell versteht keine Rohtexte, Bilder oder Audiodaten. Diese Eingaben müssen in Zahlen umgewandelt und zu Tensoren zusammengesetzt werden. In dieser Anleitung werden Sie:
-
-* Textdaten mit einem Tokenizer vorverarbeiten.
-* Bild- oder Audiodaten mit einem Feature Extractor vorverarbeiten.
-* Daten für eine multimodale Aufgabe mit einem Prozessor vorverarbeiten.
-
-## NLP
-
-<Youtube id="Yffk5aydLzg"/>
-
-Das wichtigste Werkzeug zur Verarbeitung von Textdaten ist ein [Tokenizer](main_classes/tokenizer). Ein Tokenizer zerlegt Text zunächst nach einer Reihe von Regeln in *Token*. Die Token werden in Zahlen umgewandelt, die zum Aufbau von Tensoren als Eingabe für ein Modell verwendet werden. Alle zusätzlichen Eingaben, die ein Modell benötigt, werden ebenfalls vom Tokenizer hinzugefügt.
-
-<Tip>
-
-Wenn Sie ein vortrainiertes Modell verwenden möchten, ist es wichtig, den zugehörigen vortrainierten Tokenizer zu verwenden. Dadurch wird sichergestellt, dass der Text auf die gleiche Weise aufgeteilt wird wie das Pretraining-Korpus und die gleichen entsprechenden Token-zu-Index (in der Regel als *vocab* bezeichnet) während des Pretrainings verwendet werden.
-
-</Tip>
-
-Laden Sie einen vortrainierten Tokenizer mit der Klasse [AutoTokenizer], um schnell loszulegen. Damit wird das *vocab* heruntergeladen, das verwendet wird, wenn ein Modell vortrainiert wird.
-
-### Tokenize
-
-Laden Sie einen vortrainierten Tokenizer mit [`AutoTokenizer.from_pretrained`]:
-
-```py
->>> from transformers import AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
-```
-
-Dann übergeben Sie Ihren Satz an den Tokenizer:
-
-```py
->>> encoded_input = tokenizer("Do not meddle in the affairs of wizards, for they are subtle and quick to anger.")
->>> print(encoded_input)
-{'input_ids': [101, 2079, 2025, 19960, 10362, 1999, 1996, 3821, 1997, 16657, 1010, 2005, 2027, 2024, 11259, 1998, 4248, 2000, 4963, 1012, 102], 
- 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
- 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
-```
-
-Der Tokenizer gibt ein Wörterbuch mit drei wichtigen Elementen zurück:
-
-* [input_ids](glossary#input-ids) sind die Indizes, die den einzelnen Token im Satz entsprechen.
-* [attention_mask](glossary#attention-mask) gibt an, ob ein Token beachtet werden soll oder nicht.
-* [token_type_ids](glossary#token-type-ids) gibt an, zu welcher Sequenz ein Token gehört, wenn es mehr als eine Sequenz gibt.
-
-Sie können die `input_ids` dekodieren, um die ursprüngliche Eingabe zurückzugeben:
-
-```py
->>> tokenizer.decode(encoded_input["input_ids"])
-'[CLS] Do not meddle in the affairs of wizards, for they are subtle and quick to anger. [SEP]'
-```
-
-Wie Sie sehen können, hat der Tokenisierer zwei spezielle Token - `CLS` und `SEP` (Klassifikator und Separator) - zum Satz hinzugefügt. Nicht alle Modelle benötigen
-spezielle Token, aber wenn dies der Fall ist, fügt der Tokenisierer sie automatisch für Sie hinzu.
-
-Wenn Sie mehrere Sätze verarbeiten wollen, übergeben Sie die Sätze als Liste an den Tokenizer:
-
-```py
->>> batch_sentences = [
-...     "But what about second breakfast?",
-...     "Don't think he knows about second breakfast, Pip.",
-...     "What about elevensies?",
-... ]
->>> encoded_inputs = tokenizer(batch_sentences)
->>> print(encoded_inputs)
-{'input_ids': [[101, 1252, 1184, 1164, 1248, 6462, 136, 102], 
-               [101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102], 
-               [101, 1327, 1164, 5450, 23434, 136, 102]], 
- 'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0], 
-                    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
-                    [0, 0, 0, 0, 0, 0, 0]], 
- 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1], 
-                    [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 
-                    [1, 1, 1, 1, 1, 1, 1]]}
-```
-
-### Pad
-
-Dies bringt uns zu einem wichtigen Thema. Wenn Sie einen Haufen von Sätzen verarbeiten, sind diese nicht immer gleich lang. Das ist ein Problem, weil Tensoren, die Eingabe für das Modell, eine einheitliche Form haben müssen. Padding ist eine Strategie, die sicherstellt, dass Tensoren rechteckig sind, indem ein spezielles *Padding-Token* zu Sätzen mit weniger Token hinzugefügt wird.
-
-Setzen Sie den Parameter "padding" auf "true", um die kürzeren Sequenzen im Stapel so aufzufüllen, dass sie der längsten Sequenz entsprechen:
-
-```py
->>> batch_sentences = [
-...     "But what about second breakfast?",
-...     "Don't think he knows about second breakfast, Pip.",
-...     "What about elevensies?",
-... ]
->>> encoded_input = tokenizer(batch_sentences, padding=True)
->>> print(encoded_input)
-{'input_ids': [[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0], 
-               [101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102], 
-               [101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]], 
- 'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
-                    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
-                    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], 
- 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0], 
-                    [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 
-                    [1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]]}
-```
-
-Beachten Sie, dass der Tokenizer den ersten und den dritten Satz mit einer "0" aufgefüllt hat, weil sie kürzer sind!
-
-### Kürzung
-
-Auf der anderen Seite des Spektrums kann es vorkommen, dass eine Sequenz zu lang für ein Modell ist. In diesem Fall müssen Sie die Sequenz auf eine kürzere Länge kürzen.
-
-Setzen Sie den Parameter "truncation" auf "true", um eine Sequenz auf die vom Modell akzeptierte Höchstlänge zu kürzen:
-
-```py
->>> batch_sentences = [
-...     "But what about second breakfast?",
-...     "Don't think he knows about second breakfast, Pip.",
-...     "What about elevensies?",
-... ]
->>> encoded_input = tokenizer(batch_sentences, padding=True, truncation=True)
->>> print(encoded_input)
-{'input_ids': [[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0], 
-               [101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102], 
-               [101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]], 
- 'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
-                    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
-                    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], 
- 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0], 
-                    [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 
-                    [1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]]}
-```
-
-### Tensoren erstellen
-
-Schließlich möchten Sie, dass der Tokenizer die tatsächlichen Tensoren zurückgibt, die dem Modell zugeführt werden.
-
-Setzen Sie den Parameter `return_tensors` entweder auf `pt` für PyTorch, oder `tf` für TensorFlow:
-
-<frameworkcontent>
-<pt>
-
-```py
->>> batch_sentences = [
-...     "But what about second breakfast?",
-...     "Don't think he knows about second breakfast, Pip.",
-...     "What about elevensies?",
-... ]
->>> encoded_input = tokenizer(batch_sentences, padding=True, truncation=True, return_tensors="pt")
->>> print(encoded_input)
-{'input_ids': tensor([[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0],
-                      [101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
-                      [101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]]), 
- 'token_type_ids': tensor([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-                           [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-                           [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]), 
- 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
-                           [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
-                           [1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]])}
-```
-</pt>
-<tf>
-```py
->>> batch_sentences = [
-...     "But what about second breakfast?",
-...     "Don't think he knows about second breakfast, Pip.",
-...     "What about elevensies?",
-... ]
->>> encoded_input = tokenizer(batch_sentences, padding=True, truncation=True, return_tensors="tf")
->>> print(encoded_input)
-{'input_ids': <tf.Tensor: shape=(2, 9), dtype=int32, numpy=
-array([[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0],
-       [101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
-       [101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]],
-      dtype=int32)>, 
- 'token_type_ids': <tf.Tensor: shape=(2, 9), dtype=int32, numpy=
-array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=int32)>, 
- 'attention_mask': <tf.Tensor: shape=(2, 9), dtype=int32, numpy=
-array([[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
-       [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
-       [1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=int32)>}
-```
-</tf>
-</frameworkcontent>
-
-## Audio
-
-Audioeingaben werden anders vorverarbeitet als Texteingaben, aber das Endziel bleibt dasselbe: numerische Sequenzen zu erstellen, die das Modell verstehen kann. Ein [feature extractor](main_classes/feature_extractor) dient dem ausdrücklichen Zweck, Merkmale aus Rohbild- oder Audiodaten zu extrahieren und in Tensoren zu konvertieren. Bevor Sie beginnen, installieren Sie 🤗 Datasets, um einen Audio-Datensatz zu laden, mit dem Sie experimentieren können:
-
-```bash
-pip install datasets
-```
-
-Laden Sie den [MInDS-14](https://huggingface.co/datasets/PolyAI/minds14) Datensatz (weitere Informationen zum Laden eines Datensatzes finden Sie im 🤗 [Datasets tutorial](https://huggingface.co/docs/datasets/load_hub.html)):
-
-```py
->>> from datasets import load_dataset, Audio
-
->>> dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")
-```
-
-Greifen Sie auf das erste Element der `audio`-Spalte zu, um einen Blick auf die Eingabe zu werfen. Durch den Aufruf der Spalte "audio" wird die Audiodatei automatisch geladen und neu gesampelt:
-
-```py
->>> dataset[0]["audio"]
-{'array': array([ 0.        ,  0.00024414, -0.00024414, ..., -0.00024414,
-         0.        ,  0.        ], dtype=float32),
- 'path': '/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-US~JOINT_ACCOUNT/602ba55abb1e6d0fbce92065.wav',
- 'sampling_rate': 8000}
-```
-
-Dies gibt drei Elemente zurück:
-
-* "array" ist das Sprachsignal, das als 1D-Array geladen - und möglicherweise neu gesampelt - wurde.
-* Pfad" zeigt auf den Speicherort der Audiodatei.
-* `sampling_rate` bezieht sich darauf, wie viele Datenpunkte im Sprachsignal pro Sekunde gemessen werden.
-
-### Resample
-
-Für dieses Tutorial werden Sie das Modell [Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base) verwenden. Wie Sie aus der Modellkarte ersehen können, ist das Wav2Vec2-Modell auf 16kHz abgetastetes Sprachaudio vortrainiert. Es ist wichtig, dass die Abtastrate Ihrer Audiodaten mit der Abtastrate des Datensatzes übereinstimmt, der für das Pre-Training des Modells verwendet wurde. Wenn die Abtastrate Ihrer Daten nicht dieselbe ist, müssen Sie Ihre Audiodaten neu abtasten. 
-
-Der Datensatz [MInDS-14](https://huggingface.co/datasets/PolyAI/minds14) hat zum Beispiel eine Abtastrate von 8000 kHz. Um das Wav2Vec2-Modell mit diesem Datensatz verwenden zu können, müssen Sie die Abtastrate auf 16 kHz erhöhen:
-
-```py
->>> dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")
->>> dataset[0]["audio"]
-{'array': array([ 0.        ,  0.00024414, -0.00024414, ..., -0.00024414,
-         0.        ,  0.        ], dtype=float32),
- 'path': '/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-US~JOINT_ACCOUNT/602ba55abb1e6d0fbce92065.wav',
- 'sampling_rate': 8000}
-```
-
-1. Verwenden Sie die Methode [~datasets.Dataset.cast_column] von 🤗 Datasets, um die Abtastrate auf 16kHz zu erhöhen:
-
-```py
->>> dataset = dataset.cast_column("audio", Audio(sampling_rate=16_000))
-```
-
-2. Laden Sie die Audiodatei:
-
-```py
->>> dataset[0]["audio"]
-{'array': array([ 2.3443763e-05,  2.1729663e-04,  2.2145823e-04, ...,
-         3.8356509e-05, -7.3497440e-06, -2.1754686e-05], dtype=float32),
- 'path': '/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-US~JOINT_ACCOUNT/602ba55abb1e6d0fbce92065.wav',
- 'sampling_rate': 16000}
-```
-
-Wie Sie sehen können, ist die Abtastrate jetzt 16kHz!
-
-### Merkmalsextraktor
-
-Der nächste Schritt ist das Laden eines Merkmalsextraktors, um die Eingabe zu normalisieren und aufzufüllen. Beim Auffüllen von Textdaten wird für kürzere Sequenzen ein `0` hinzugefügt. Die gleiche Idee gilt für Audiodaten, und der Audio-Feature-Extraktor fügt eine `0` - interpretiert als Stille - zu `array` hinzu.
-
-Laden Sie den Merkmalsextraktor mit [`AutoFeatureExtractor.from_pretrained`]:
-
-```py
->>> from transformers import AutoFeatureExtractor
-
->>> feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base")
-```
-
-Übergeben Sie das Audio-"Array" an den Feature-Extraktor. Wir empfehlen auch, das Argument `sampling_rate` im Feature Extractor hinzuzufügen, um eventuell auftretende stille Fehler besser zu beheben.
-
-```py
->>> audio_input = [dataset[0]["audio"]["array"]]
->>> feature_extractor(audio_input, sampling_rate=16000)
-{'input_values': [array([ 3.8106556e-04,  2.7506407e-03,  2.8015103e-03, ...,
-        5.6335266e-04,  4.6588284e-06, -1.7142107e-04], dtype=float32)]}
-```
-
-### Auffüllen und Kürzen
-
-Genau wie beim Tokenizer können Sie variable Sequenzen in einem Stapel durch Auffüllen oder Abschneiden behandeln. Werfen Sie einen Blick auf die Sequenzlänge dieser beiden Audiobeispiele:
-
-```py
->>> dataset[0]["audio"]["array"].shape
-(173398,)
-
->>> dataset[1]["audio"]["array"].shape
-(106496,)
-```
-
-Wie Sie sehen können, hat das erste Beispiel eine längere Sequenz als das zweite Beispiel. Lassen Sie uns eine Funktion erstellen, die den Datensatz vorverarbeitet. Geben Sie eine maximale Länge der Probe an, und der Feature-Extraktor wird die Sequenzen entweder auffüllen oder abschneiden, damit sie dieser Länge entsprechen:
-
-```py
->>> def preprocess_function(examples):
-...     audio_arrays = [x["array"] for x in examples["audio"]]
-...     inputs = feature_extractor(
-...         audio_arrays,
-...         sampling_rate=16000,
-...         padding=True,
-...         max_length=100000,
-...         truncation=True,
-...     )
-...     return inputs
-```
-
-Wenden Sie die Funktion auf die ersten paar Beispiele im Datensatz an:
-
-```py
->>> processed_dataset = preprocess_function(dataset[:5])
-```
-
-Schauen Sie sich nun noch einmal die verarbeiteten Beispiel-Längen an:
-
-```py
->>> processed_dataset["input_values"][0].shape
-(100000,)
-
->>> processed_dataset["input_values"][1].shape
-(100000,)
-```
-
-Die Länge der ersten beiden Beispiele entspricht nun der von Ihnen angegebenen Maximallänge.
-
-## Bildverarbeitung
-
-Ein Merkmalsextraktor wird auch verwendet, um Bilder für Bildverarbeitungsaufgaben zu verarbeiten. Auch hier besteht das Ziel darin, das Rohbild in eine Reihe von Tensoren als Eingabe zu konvertieren.
-
-Laden wir den [food101](https://huggingface.co/datasets/food101) Datensatz für dieses Tutorial. Verwenden Sie den Parameter 🤗 Datasets `split`, um nur eine kleine Stichprobe aus dem Trainingssplit zu laden, da der Datensatz recht groß ist:
-
-```py
->>> from datasets import load_dataset
-
->>> dataset = load_dataset("food101", split="train[:100]")
-```
-
-Als Nächstes sehen Sie sich das Bild mit dem Merkmal 🤗 Datensätze [Bild] (https://huggingface.co/docs/datasets/package_reference/main_classes.html?highlight=image#datasets.Image) an:
-
-```py
->>> dataset[0]["image"]
-```
-
-![vision-preprocess-tutorial.png](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/vision-preprocess-tutorial.png)
-
-### Merkmalsextraktor
-
-Laden Sie den Merkmalsextraktor mit [`AutoFeatureExtractor.from_pretrained`]:
-
-```py
->>> from transformers import AutoFeatureExtractor
-
->>> feature_extractor = AutoFeatureExtractor.from_pretrained("google/vit-base-patch16-224")
-```
-
-### Datenerweiterung
-
-Bei Bildverarbeitungsaufgaben ist es üblich, den Bildern als Teil der Vorverarbeitung eine Art von Datenerweiterung hinzuzufügen. Sie können Erweiterungen mit jeder beliebigen Bibliothek hinzufügen, aber in diesem Tutorial werden Sie das Modul [`transforms`](https://pytorch.org/vision/stable/transforms.html) von torchvision verwenden.
-
-1. Normalisieren Sie das Bild und verwenden Sie [`Compose`](https://pytorch.org/vision/master/generated/torchvision.transforms.Compose.html), um einige Transformationen - [`RandomResizedCrop`](https://pytorch.org/vision/main/generated/torchvision.transforms.RandomResizedCrop.html) und [`ColorJitter`](https://pytorch.org/vision/main/generated/torchvision.transforms.ColorJitter.html) - miteinander zu verknüpfen:
-
-```py
->>> from torchvision.transforms import Compose, Normalize, RandomResizedCrop, ColorJitter, ToTensor
-
->>> normalize = Normalize(mean=feature_extractor.image_mean, std=feature_extractor.image_std)
->>> _transforms = Compose(
-...     [RandomResizedCrop(feature_extractor.size), ColorJitter(brightness=0.5, hue=0.5), ToTensor(), normalize]
-... )
-```
-
-2. Das Modell akzeptiert [`pixel_values`](model_doc/visionencoderdecoder#transformers.VisionEncoderDecoderModel.forward.pixel_values) als Eingabe. Dieser Wert wird vom Merkmalsextraktor erzeugt. Erstellen Sie eine Funktion, die `pixel_values` aus den Transformationen erzeugt:
-
-```py
->>> def transforms(examples):
-...     examples["pixel_values"] = [_transforms(image.convert("RGB")) for image in examples["image"]]
-...     return examples
-```
-
-3. Dann verwenden Sie 🤗 Datasets [`set_transform`](https://huggingface.co/docs/datasets/process.html#format-transform), um die Transformationen im laufenden Betrieb anzuwenden:
-
-```py
->>> dataset.set_transform(transforms)
-```
-
-4. Wenn Sie nun auf das Bild zugreifen, werden Sie feststellen, dass der Feature Extractor die Modelleingabe "pixel_values" hinzugefügt hat:
-
-```py
->>> dataset[0]["image"]
-{'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=384x512 at 0x7F1A7B0630D0>,
- 'label': 6,
- 'pixel_values': tensor([[[ 0.0353,  0.0745,  0.1216,  ..., -0.9922, -0.9922, -0.9922],
-          [-0.0196,  0.0667,  0.1294,  ..., -0.9765, -0.9843, -0.9922],
-          [ 0.0196,  0.0824,  0.1137,  ..., -0.9765, -0.9686, -0.8667],
-          ...,
-          [ 0.0275,  0.0745,  0.0510,  ..., -0.1137, -0.1216, -0.0824],
-          [ 0.0667,  0.0824,  0.0667,  ..., -0.0588, -0.0745, -0.0980],
-          [ 0.0353,  0.0353,  0.0431,  ..., -0.0039, -0.0039, -0.0588]],
- 
-         [[ 0.2078,  0.2471,  0.2863,  ..., -0.9451, -0.9373, -0.9451],
-          [ 0.1608,  0.2471,  0.3098,  ..., -0.9373, -0.9451, -0.9373],
-          [ 0.2078,  0.2706,  0.3020,  ..., -0.9608, -0.9373, -0.8275],
-          ...,
-          [-0.0353,  0.0118, -0.0039,  ..., -0.2392, -0.2471, -0.2078],
-          [ 0.0196,  0.0353,  0.0196,  ..., -0.1843, -0.2000, -0.2235],
-          [-0.0118, -0.0039, -0.0039,  ..., -0.0980, -0.0980, -0.1529]],
- 
-         [[ 0.3961,  0.4431,  0.4980,  ..., -0.9216, -0.9137, -0.9216],
-          [ 0.3569,  0.4510,  0.5216,  ..., -0.9059, -0.9137, -0.9137],
-          [ 0.4118,  0.4745,  0.5216,  ..., -0.9137, -0.8902, -0.7804],
-          ...,
-          [-0.2314, -0.1922, -0.2078,  ..., -0.4196, -0.4275, -0.3882],
-          [-0.1843, -0.1686, -0.2000,  ..., -0.3647, -0.3804, -0.4039],
-          [-0.1922, -0.1922, -0.1922,  ..., -0.2941, -0.2863, -0.3412]]])}
-```
-
-Hier sehen Sie, wie das Bild nach der Vorverarbeitung aussieht. Wie von den angewandten Transformationen zu erwarten, wurde das Bild willkürlich beschnitten und seine Farbeigenschaften sind anders.
-
-```py
->>> import numpy as np
->>> import matplotlib.pyplot as plt
-
->>> img = dataset[0]["pixel_values"]
->>> plt.imshow(img.permute(1, 2, 0))
-```
-
-![preprocessed_image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/preprocessed_image.png)
-
-## Multimodal
-
-Für multimodale Aufgaben werden Sie eine Kombination aus allem, was Sie bisher gelernt haben, verwenden und Ihre Fähigkeiten auf eine Aufgabe der automatischen Spracherkennung (ASR) anwenden. Dies bedeutet, dass Sie einen:
-
-* Feature Extractor zur Vorverarbeitung der Audiodaten.
-* Tokenizer, um den Text zu verarbeiten.
-
-Kehren wir zum [LJ Speech](https://huggingface.co/datasets/lj_speech) Datensatz zurück:
-
-```py
->>> from datasets import load_dataset
-
->>> lj_speech = load_dataset("lj_speech", split="train")
-```
-
-Da Sie hauptsächlich an den Spalten "Audio" und "Text" interessiert sind, entfernen Sie die anderen Spalten:
-
-```py
->>> lj_speech = lj_speech.map(remove_columns=["file", "id", "normalized_text"])
-```
-
-Schauen Sie sich nun die Spalten "Audio" und "Text" an:
-
-```py
->>> lj_speech[0]["audio"]
-{'array': array([-7.3242188e-04, -7.6293945e-04, -6.4086914e-04, ...,
-         7.3242188e-04,  2.1362305e-04,  6.1035156e-05], dtype=float32),
- 'path': '/root/.cache/huggingface/datasets/downloads/extracted/917ece08c95cf0c4115e45294e3cd0dee724a1165b7fc11798369308a465bd26/LJSpeech-1.1/wavs/LJ001-0001.wav',
- 'sampling_rate': 22050}
-
->>> lj_speech[0]["text"]
-'Printing, in the only sense with which we are at present concerned, differs from most if not from all the arts and crafts represented in the Exhibition'
-```
-
-Erinnern Sie sich an den früheren Abschnitt über die Verarbeitung von Audiodaten: Sie sollten immer die Abtastrate Ihrer Audiodaten [resample](preprocessing#audio), damit sie mit der Abtastrate des Datensatzes übereinstimmt, der für das Vortraining eines Modells verwendet wird:
-
-```py
->>> lj_speech = lj_speech.cast_column("audio", Audio(sampling_rate=16_000))
-```
-
-### Prozessor
-
-Ein Processor kombiniert einen Feature-Extraktor und einen Tokenizer. Laden Sie einen Processor mit [`AutoProcessor.from_pretrained]:
-
-```py
->>> from transformers import AutoProcessor
-
->>> processor = AutoProcessor.from_pretrained("facebook/wav2vec2-base-960h")
-```
-
-1. Erstellen Sie eine Funktion, die die Audiodaten zu `input_values` verarbeitet und den Text zu `labels` tokenisiert. Dies sind Ihre Eingaben für das Modell:
-
-```py
->>> def prepare_dataset(example):
-...     audio = example["audio"]
-
-...     example.update(processor(audio=audio["array"], text=example["text"], sampling_rate=16000))
-
-...     return example
-```
-
-2. Wenden Sie die Funktion "prepare_dataset" auf ein Beispiel an:
-
-```py
->>> prepare_dataset(lj_speech[0])
-```
-
-Beachten Sie, dass der Processor `input_values` und `labels` hinzugefügt hat. Auch die Abtastrate wurde korrekt auf 16kHz heruntergerechnet.
-
-Toll, Sie sollten jetzt in der Lage sein, Daten für jede Modalität vorzuverarbeiten und sogar verschiedene Modalitäten zu kombinieren! Im nächsten Kurs lernen Sie, wie Sie ein Modell mit Ihren neu aufbereiteten Daten feinabstimmen können.

docs/source/de/quicktour.mdx

@@ -1,428 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Schnellstart
-
-[[open-in-colab]]
-
-Mit 🤗 Transformers können Sie sofort loslegen! Verwenden Sie die [`pipeline`] für schnelle Inferenz und laden Sie schnell ein vortrainiertes Modell und einen Tokenizer mit einer [AutoClass](./model_doc/auto), um Ihre Text-, Bild- oder Audioaufgabe zu lösen.
-
-<Tip>
-
-Alle in der Dokumentation vorgestellten Codebeispiele haben oben links einen Umschalter für PyTorch und TensorFlow. Wenn
-nicht, wird erwartet, dass der Code für beide Backends ohne Änderungen funktioniert.
-
-</Tip>
-
-## Pipeline
-
-[`pipeline`] ist der einfachste Weg, ein vortrainiertes Modell für eine bestimmte Aufgabe zu verwenden.
-
-<Youtube id="tiZFewofSLM"/>
-
-Die [`pipeline`] unterstützt viele gängige Aufgaben:
-
-**Text**:
-* Stimmungsanalyse: Klassifizierung der Polarität eines gegebenen Textes.
-* Textgenerierung (auf Englisch): Generierung von Text aus einer gegebenen Eingabe.
-* Name-Entity-Recognition (NER): Kennzeichnung jedes Worts mit der Entität, die es repräsentiert (Person, Datum, Ort usw.).
-* Beantwortung von Fragen: Extrahieren der Antwort aus dem Kontext, wenn ein gewisser Kontext und eine Frage gegeben sind.
-* Fill-mask: Ausfüllen von Lücken in einem Text mit maskierten Wörtern.
-* Zusammenfassung: Erstellung einer Zusammenfassung einer langen Text- oder Dokumentensequenz.
-* Übersetzung: Übersetzen eines Textes in eine andere Sprache.
-* Merkmalsextraktion: Erstellen einer Tensordarstellung des Textes.
-
-**Bild**:
-* Bildklassifizierung: Klassifizierung eines Bildes.
-* Bildsegmentierung: Klassifizierung jedes Pixels in einem Bild.
-* Objekterkennung: Erkennen von Objekten innerhalb eines Bildes.
-
-**Audio**:
-* Audioklassifizierung: Zuweisung eines Labels zu einem bestimmten Audiosegment.
-* Automatische Spracherkennung (ASR): Transkription von Audiodaten in Text.
-
-<Tip>
-
-Für mehr Details über die [`pipeline`] und assoziierte Aufgaben, schauen Sie in die Dokumentation [hier](./main_classes/pipelines).
-
-</Tip>
-
-### Verwendung der Pipeline
-
-Im folgenden Beispiel werden Sie die [`pipeline`] für die Stimmungsanalyse verwenden.
-
-Installieren Sie die folgenden Abhängigkeiten, falls Sie dies nicht bereits getan haben:
-
-<frameworkcontent>
-<pt>
-```bash
-pip install torch
-```
-</pt>
-<tf>
-```bash
-pip install tensorflow
-```
-</tf>
-</frameworkcontent>
-
-Importieren sie die [`pipeline`] und spezifizieren sie die Aufgabe, welche sie lösen möchten:
-
-```py
->>> from transformers import pipeline
-
->>> classifier = pipeline("sentiment-analysis")
-```
-
-Die Pipeline lädt ein standardmäßiges [vortrainiertes Modell] (https://huggingface.co/distilbert-base-uncased-finetuned-sst-2-english) und einen Tokenizer für die Stimmungs-Analyse herunter und speichert sie. Jetzt können Sie den "Klassifikator" auf Ihren Zieltext anwenden:
-
-```py
->>> classifier("We are very happy to show you the 🤗 Transformers library.")
-[{'label': 'POSITIVE', 'score': 0.9998}]
-```
-
-For more than one sentence, pass a list of sentences to the [`pipeline`] which returns a list of dictionaries:
-
-```py
->>> results = classifier(["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."])
->>> for result in results:
-...     print(f"label: {result['label']}, with score: {round(result['score'], 4)}")
-label: POSITIVE, with score: 0.9998
-label: NEGATIVE, with score: 0.5309
-```
-
-Die [`pipeline`] kann auch über einen ganzen Datensatz iterieren. Starten wir mit der Installation der [🤗 Datasets](https://huggingface.co/docs/datasets/) Bibliothek:
-
-```bash
-pip install datasets 
-```
-
-Erstellen wir eine [`pipeline`] mit der Aufgabe die wir lösen und dem Modell welches wir nutzen möchten.
-
-```py
->>> import torch
->>> from transformers import pipeline
-
->>> speech_recognizer = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")
-```
-
-Als nächstes laden wir den Datensatz (siehe 🤗 Datasets [Quick Start](https://huggingface.co/docs/datasets/quickstart.html) für mehr Details) welches wir nutzen möchten. Zum Beispiel laden wir den [MInDS-14](https://huggingface.co/datasets/PolyAI/minds14) Datensatz:
-
-```py
->>> from datasets import load_dataset, Audio
-
->>> dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")  # doctest: +IGNORE_RESULT
-```
-
-Wir müssen sicherstellen, dass die Abtastrate des Datensatzes der Abtastrate entspricht, mit der `facebook/wav2vec2-base-960h` trainiert wurde.
-
-```py
->>> dataset = dataset.cast_column("audio", Audio(sampling_rate=speech_recognizer.feature_extractor.sampling_rate))
-```
-
-Audiodateien werden automatisch geladen und neu abgetastet, wenn die Spalte "audio" aufgerufen wird.
-Extrahieren wir die rohen Wellenform-Arrays der ersten 4 Beispiele und übergeben wir sie als Liste an die Pipeline:
-
-```py
->>> result = speech_recognizer(dataset[:4]["audio"])
->>> print([d["text"] for d in result])
-['I WOULD LIKE TO SET UP A JOINT ACCOUNT WITH MY PARTNER HOW DO I PROCEED WITH DOING THAT', "FODING HOW I'D SET UP A JOIN TO HET WITH MY WIFE AND WHERE THE AP MIGHT BE", "I I'D LIKE TOY SET UP A JOINT ACCOUNT WITH MY PARTNER I'M NOT SEEING THE OPTION TO DO IT ON THE AP SO I CALLED IN TO GET SOME HELP CAN I JUST DO IT OVER THE PHONE WITH YOU AND GIVE YOU THE INFORMATION OR SHOULD I DO IT IN THE AP AND I'M MISSING SOMETHING UQUETTE HAD PREFERRED TO JUST DO IT OVER THE PHONE OF POSSIBLE THINGS", 'HOW DO I THURN A JOIN A COUNT']
-```
-
-Bei einem größeren Datensatz mit vielen Eingaben (wie bei Sprache oder Bildverarbeitung) sollten Sie einen Generator anstelle einer Liste übergeben, der alle Eingaben in den Speicher lädt. Weitere Informationen finden Sie in der [Pipeline-Dokumentation](./main_classes/pipelines).
-
-### Ein anderes Modell und einen anderen Tokenizer in der Pipeline verwenden
-
-Die [`pipeline`] kann jedes Modell aus dem [Model Hub] (https://huggingface.co/models) verwenden, wodurch es einfach ist, die [`pipeline`] für andere Anwendungsfälle anzupassen. Wenn Sie beispielsweise ein Modell wünschen, das französischen Text verarbeiten kann, verwenden Sie die Tags im Model Hub, um nach einem geeigneten Modell zu filtern. Das oberste gefilterte Ergebnis liefert ein mehrsprachiges [BERT-Modell](https://huggingface.co/nlptown/bert-base-multilingual-uncased-sentiment), das auf die Stimmungsanalyse abgestimmt ist. Großartig, verwenden wir dieses Modell!
-
-```py
->>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
-```
-
-<frameworkcontent>
-<pt>
-Use the [`AutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the pretrained model and it's associated tokenizer (more on an `AutoClass` below):
-
-```py
->>> from transformers import AutoTokenizer, AutoModelForSequenceClassification
-
->>> model = AutoModelForSequenceClassification.from_pretrained(model_name)
->>> tokenizer = AutoTokenizer.from_pretrained(model_name)
-```
-</pt>
-<tf>
-Use the [`TFAutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the pretrained model and it's associated tokenizer (more on an `TFAutoClass` below):
-
-```py
->>> from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
-
->>> model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
->>> tokenizer = AutoTokenizer.from_pretrained(model_name)
-```
-</tf>
-</frameworkcontent>
-
-Dann können Sie das Modell und den Tokenizer in der [`pipeline`] angeben und den `Klassifikator` auf Ihren Zieltext anwenden:
-
-```py
->>> classifier = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)
->>> classifier("Nous sommes très heureux de vous présenter la bibliothèque 🤗 Transformers.")
-[{'label': '5 stars', 'score': 0.7273}]
-```
-
-Wenn Sie kein Modell für Ihren Anwendungsfall finden können, müssen Sie ein vortrainiertes Modell auf Ihren Daten feinabstimmen. Schauen Sie sich unser [Feinabstimmungs-Tutorial](./training) an, um zu erfahren, wie das geht. Und schließlich, nachdem Sie Ihr trainiertes Modell verfeinert haben, sollten Sie es mit der Community im Model Hub teilen (siehe Tutorial [hier](./model_sharing)), um NLP für alle zu demokratisieren! 🤗
-
-## AutoClass
-
-<Youtube id="AhChOFRegn4"/>
-
-Unter der Haube arbeiten die Klassen [`AutoModelForSequenceClassification`] und [`AutoTokenizer`] zusammen, um die [`pipeline`] zu betreiben. Eine [`AutoClass`](./model_doc/auto) ist eine Abkürzung, die automatisch die Architektur eines trainierten Modells aus dessen Namen oder Pfad abruft. Sie müssen nur die passende `AutoClass` für Ihre Aufgabe und den zugehörigen Tokenizer mit [`AutoTokenizer`] auswählen. 
-
-Kehren wir zu unserem Beispiel zurück und sehen wir uns an, wie Sie die `AutoClass` verwenden können, um die Ergebnisse der [`pipeline`] zu replizieren.
-
-### AutoTokenizer
-
-Ein Tokenizer ist für die Vorverarbeitung von Text in ein für das Modell verständliches Format zuständig. Zunächst zerlegt der Tokenisierer den Text in Wörter, die *Token* genannt werden. Es gibt mehrere Regeln für den Tokenisierungsprozess, z. B. wie und auf welcher Ebene ein Wort aufgespalten wird (weitere Informationen über Tokenisierung [hier](./tokenizer_summary)). Das Wichtigste ist jedoch, dass Sie den Tokenizer mit demselben Modellnamen instanziieren müssen, um sicherzustellen, dass Sie dieselben Tokenisierungsregeln verwenden, mit denen ein Modell zuvor trainiert wurde.
-Laden sie einen Tokenizer mit [`AutoTokenizer`]:
-
-```py
->>> from transformers import AutoTokenizer
-
->>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
->>> tokenizer = AutoTokenizer.from_pretrained(model_name)
-```
-
-Anschließend wandelt der Tokenizer die Token in Zahlen um, um einen Tensor als Eingabe für das Modell zu konstruieren. Dieser wird als *Vokabular* des Modells bezeichnet.
-
-Übergeben Sie Ihren Text an den Tokenizer:
-
-```py
->>> encoding = tokenizer("We are very happy to show you the 🤗 Transformers library.")
->>> print(encoding)
-{'input_ids': [101, 11312, 10320, 12495, 19308, 10114, 11391, 10855, 10103, 100, 58263, 13299, 119, 102],
- 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
- 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
-```
-
-Der Tokenizer gibt ein Wörterbuch zurück, das Folgendes enthält:
-
-* [input_ids](./glossary#input-ids): numerische Repräsentationen Ihrer Token.
-* [atttention_mask](.glossary#attention-mask): gibt an, welche Token beachtet werden sollen.
-
-Genau wie die [`pipeline`] akzeptiert der Tokenizer eine Liste von Eingaben. Darüber hinaus kann der Tokenizer den Text auch auffüllen und kürzen, um einen Stapel mit einheitlicher Länge zurückzugeben:
-
-<frameworkcontent>
-<pt>
-```py
->>> pt_batch = tokenizer(
-...     ["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."],
-...     padding=True,
-...     truncation=True,
-...     max_length=512,
-...     return_tensors="pt",
-... )
-```
-</pt>
-<tf>
-```py
->>> tf_batch = tokenizer(
-...     ["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."],
-...     padding=True,
-...     truncation=True,
-...     max_length=512,
-...     return_tensors="tf",
-... )
-```
-</tf>
-</frameworkcontent>
-
-Lesen Sie das Tutorial [preprocessing](./preprocessing) für weitere Details zur Tokenisierung.
-
-### AutoModel
-
-<frameworkcontent>
-<pt>
-🤗 Transformers bietet eine einfache und einheitliche Möglichkeit, vortrainierte Instanzen zu laden. Das bedeutet, dass Sie ein [`AutoModel`] laden können, wie Sie einen [`AutoTokenizer`] laden würden. Der einzige Unterschied ist die Auswahl des richtigen [`AutoModel`] für die Aufgabe. Da Sie eine Text- oder Sequenzklassifizierung vornehmen, laden Sie [`AutoModelForSequenceClassification`]:
-
-```py
->>> from transformers import AutoModelForSequenceClassification
-
->>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
->>> pt_model = AutoModelForSequenceClassification.from_pretrained(model_name)
-```
-
-<Tip>
-
-In der [Aufgabenzusammenfassung](./task_summary) steht, welche [AutoModel]-Klasse für welche Aufgabe zu verwenden ist.
-
-</Tip>
-
-Jetzt können Sie Ihren vorverarbeiteten Stapel von Eingaben direkt an das Modell übergeben. Sie müssen nur das Wörterbuch entpacken, indem Sie `**` hinzufügen:
-
-```py
->>> pt_outputs = pt_model(**pt_batch)
-```
-
-Das Modell gibt die endgültigen Aktivierungen in dem Attribut "logits" aus. Wenden Sie die Softmax-Funktion auf die "logits" an, um die Wahrscheinlichkeiten zu erhalten:
-  
-```py
->>> from torch import nn
-
->>> pt_predictions = nn.functional.softmax(pt_outputs.logits, dim=-1)
->>> print(pt_predictions)
-tensor([[0.0021, 0.0018, 0.0115, 0.2121, 0.7725],
-        [0.2084, 0.1826, 0.1969, 0.1755, 0.2365]], grad_fn=<SoftmaxBackward0>)
-```
-</pt>
-<tf>
-🤗 Transformers bietet eine einfache und einheitliche Methode zum Laden von vortrainierten Instanzen. Das bedeutet, dass Sie ein [`TFAutoModel`] genauso laden können, wie Sie einen [`AutoTokenizer`] laden würden. Der einzige Unterschied ist die Auswahl des richtigen [`TFAutoModel`] für die Aufgabe. Da Sie Text - oder Sequenz - Klassifizierung machen, laden Sie [`TFAutoModelForSequenceClassification`]:
-
-```py
->>> from transformers import TFAutoModelForSequenceClassification
-
->>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
->>> tf_model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
-```
-
-<Tip>
-
-In der [Aufgabenzusammenfassung](./task_summary) steht, welche [AutoModel]-Klasse für welche Aufgabe zu verwenden ist.
-
-</Tip>
-
-Jetzt können Sie Ihren vorverarbeiteten Stapel von Eingaben direkt an das Modell übergeben, indem Sie die Wörterbuchschlüssel direkt an die Tensoren übergeben:
-  
-```py
->>> tf_outputs = tf_model(tf_batch)
-```
-
-Das Modell gibt die endgültigen Aktivierungen in dem Attribut "logits" aus. Wenden Sie die Softmax-Funktion auf die "logits" an, um die Wahrscheinlichkeiten zu erhalten:
-
-```py
->>> import tensorflow as tf
-
->>> tf_predictions = tf.nn.softmax(tf_outputs.logits, axis=-1)
->>> tf_predictions  # doctest: +IGNORE_RESULT
-```
-</tf>
-</frameworkcontent>
-
-<Tip>
-
-Alle 🤗 Transformers-Modelle (PyTorch oder TensorFlow) geben die Tensoren *vor* der endgültigen Aktivierungsfunktion
-Funktion (wie Softmax) aus, da die endgültige Aktivierungsfunktion oft mit dem Verlusten verschmolzen ist.
-
-</Tip>
-
-Modelle sind ein standardmäßiges [`torch.nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) oder ein [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model), sodass Sie sie in Ihrer üblichen Trainingsschleife verwenden können. Um jedoch die Dinge einfacher zu machen, bietet 🤗 Transformers eine [`Trainer`]-Klasse für PyTorch, die Funktionalität für verteiltes Training, gemischte Präzision und mehr bietet. Für TensorFlow können Sie die Methode `fit` aus [Keras](https://keras.io/) verwenden. Siehe das [training tutorial](./training) für weitere Details.
-
-<Tip>
-
-Transformers-Modellausgaben sind spezielle Datenklassen, so dass ihre Attribute in einer IDE automatisch vervollständigt werden.
-Die Modellausgänge verhalten sich auch wie ein Tupel oder ein Wörterbuch (z.B. können Sie mit einem Integer, einem Slice oder einem String indexieren), wobei die Attribute, die "None" sind, ignoriert werden.
-
-</Tip>
-
-### Modell speichern
-
-<frameworkcontent>
-<pt>
-Sobald Ihr Modell feinabgestimmt ist, können Sie es mit seinem Tokenizer speichern, indem Sie [`PreTrainedModel.save_pretrained`] verwenden:
-
-```py
->>> pt_save_directory = "./pt_save_pretrained"
->>> tokenizer.save_pretrained(pt_save_directory)  # doctest: +IGNORE_RESULT
->>> pt_model.save_pretrained(pt_save_directory)
-```
-
-Wenn Sie bereit sind, das Modell erneut zu verwenden, laden Sie es mit [`PreTrainedModel.from_pretrained`]:
-
-```py
->>> pt_model = AutoModelForSequenceClassification.from_pretrained("./pt_save_pretrained")
-```
-</pt>
-<tf>
-Sobald Ihr Modell feinabgestimmt ist, können Sie es mit seinem Tokenizer unter Verwendung von [`TFPreTrainedModel.save_pretrained`] speichern:
-
-```py
->>> tf_save_directory = "./tf_save_pretrained"
->>> tokenizer.save_pretrained(tf_save_directory)  # doctest: +IGNORE_RESULT
->>> tf_model.save_pretrained(tf_save_directory)
-```
-
-Wenn Sie bereit sind, das Modell wieder zu verwenden, laden Sie es mit [`TFPreTrainedModel.from_pretrained`]:
-
-```py
->>> tf_model = TFAutoModelForSequenceClassification.from_pretrained("./tf_save_pretrained")
-```
-</tf>
-</frameworkcontent>
-
-Ein besonders cooles 🤗 Transformers-Feature ist die Möglichkeit, ein Modell zu speichern und es entweder als PyTorch- oder TensorFlow-Modell wieder zu laden. Der Parameter "from_pt" oder "from_tf" kann das Modell von einem Framework in das andere konvertieren:
-
-<frameworkcontent>
-<pt>
-```py
->>> from transformers import AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained(tf_save_directory)
->>> pt_model = AutoModelForSequenceClassification.from_pretrained(tf_save_directory, from_tf=True)
-```
-</pt>
-<tf>
-```py
->>> from transformers import TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained(pt_save_directory)
->>> tf_model = TFAutoModelForSequenceClassification.from_pretrained(pt_save_directory, from_pt=True)
-```
-</tf>
-</frameworkcontent>
-
-## Custom model builds
-
-Sie können die Konfigurationsklasse des Modells ändern, um zu bestimmen, wie ein Modell aufgebaut ist. Die Konfiguration legt die Attribute eines Modells fest, z. B. die Anzahl der verborgenen Schichten oder der Aufmerksamkeitsköpfe. Wenn Sie ein Modell aus einer benutzerdefinierten Konfigurationsklasse initialisieren, beginnen Sie bei Null. Die Modellattribute werden zufällig initialisiert, und Sie müssen das Modell trainieren, bevor Sie es verwenden können, um aussagekräftige Ergebnisse zu erhalten.
-
-Beginnen Sie mit dem Import von [`AutoConfig`] und laden Sie dann das trainierte Modell, das Sie ändern möchten. Innerhalb von [`AutoConfig.from_pretrained`] können Sie das Attribut angeben, das Sie ändern möchten, z. B. die Anzahl der Aufmerksamkeitsköpfe:
-
-```py
->>> from transformers import AutoConfig
-
->>> my_config = AutoConfig.from_pretrained("distilbert-base-uncased", n_heads=12)
-```
-
-<frameworkcontent>
-<pt>
-Create a model from your custom configuration with [`AutoModel.from_config`]:
-
-```py
->>> from transformers import AutoModel
-
->>> my_model = AutoModel.from_config(my_config)
-```
-</pt>
-<tf>
-Create a model from your custom configuration with [`TFAutoModel.from_config`]:
-
-```py
->>> from transformers import TFAutoModel
-
->>> my_model = TFAutoModel.from_config(my_config)
-```
-</tf>
-</frameworkcontent>
-
-Weitere Informationen zur Erstellung von benutzerdefinierten Konfigurationen finden Sie in der Anleitung [Erstellen einer benutzerdefinierten Architektur](./create_a_model).
-
-## Wie geht es weiter?
-
-Nachdem Sie nun die 🤗 Transformers-Kurztour abgeschlossen haben, schauen Sie sich unsere Anleitungen an und erfahren Sie, wie Sie spezifischere Dinge tun können, wie das Schreiben eines benutzerdefinierten Modells, die Feinabstimmung eines Modells für eine Aufgabe und wie man ein Modell mit einem Skript trainiert. Wenn Sie mehr über die Kernkonzepte von 🤗 Transformers erfahren möchten, nehmen Sie sich eine Tasse Kaffee und werfen Sie einen Blick auf unsere konzeptionellen Leitfäden!

docs/source/de/training.mdx

@@ -1,427 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Optimierung eines vortrainierten Modells
-
-[[open-in-colab]]
-
-Die Verwendung eines vorab trainierten Modells hat erhebliche Vorteile. Es reduziert die Rechenkosten und den CO2-Fußabdruck und ermöglicht Ihnen die Verwendung von Modellen, die dem neuesten Stand der Technik entsprechen, ohne dass Sie ein Modell von Grund auf neu trainieren müssen. Transformers bietet Zugang zu Tausenden von vortrainierten Modellen für eine Vielzahl von Aufgaben. Wenn Sie ein vorab trainiertes Modell verwenden, trainieren Sie es auf einem für Ihre Aufgabe spezifischen Datensatz. Dies wird als Feinabstimmung bezeichnet und ist eine unglaublich leistungsfähige Trainingstechnik. In diesem Tutorial werden Sie ein vortrainiertes Modell mit einem Deep-Learning-Framework Ihrer Wahl feinabstimmen:
-
-* Feinabstimmung eines vorab trainierten Modells mit 🤗 Transformers [`Trainer`].
-* Feinabstimmung eines vorab trainierten Modells in TensorFlow mit Keras.
-* Feinabstimmung eines vorab trainierten Modells in nativem PyTorch.
-
-<a id='data-processing'></a>
-
-## Vorbereitung eines Datensatzes
-
-<Youtube id="_BZearw7f0w"/>
-
-Bevor Sie die Feinabstimmung eines vortrainierten Modells vornehmen können, müssen Sie einen Datensatz herunterladen und für das Training vorbereiten. Im vorangegangenen Leitfaden haben Sie gelernt, wie man Daten für das Training aufbereitet, und jetzt haben Sie die Gelegenheit, diese Fähigkeiten zu testen!
-
-Laden Sie zunächst den Datensatz [Yelp Reviews](https://huggingface.co/datasets/yelp_review_full):
-
-```py
->>> from datasets import load_dataset
-
->>> dataset = load_dataset("yelp_review_full")
->>> dataset["train"][100]
-{'label': 0,
- 'text': 'My expectations for McDonalds are t rarely high. But for one to still fail so spectacularly...that takes something special!\\nThe cashier took my friends\'s order, then promptly ignored me. I had to force myself in front of a cashier who opened his register to wait on the person BEHIND me. I waited over five minutes for a gigantic order that included precisely one kid\'s meal. After watching two people who ordered after me be handed their food, I asked where mine was. The manager started yelling at the cashiers for \\"serving off their orders\\" when they didn\'t have their food. But neither cashier was anywhere near those controls, and the manager was the one serving food to customers and clearing the boards.\\nThe manager was rude when giving me my order. She didn\'t make sure that I had everything ON MY RECEIPT, and never even had the decency to apologize that I felt I was getting poor service.\\nI\'ve eaten at various McDonalds restaurants for over 30 years. I\'ve worked at more than one location. I expect bad days, bad moods, and the occasional mistake. But I have yet to have a decent experience at this store. It will remain a place I avoid unless someone in my party needs to avoid illness from low blood sugar. Perhaps I should go back to the racially biased service of Steak n Shake instead!'}
-```
-
-Wie Sie nun wissen, benötigen Sie einen Tokenizer, um den Text zu verarbeiten und eine Auffüll- und Abschneidungsstrategie einzubauen, um mit variablen Sequenzlängen umzugehen. Um Ihren Datensatz in einem Schritt zu verarbeiten, verwenden Sie die 🤗 Methode Datasets [`map`](https://huggingface.co/docs/datasets/process.html#map), um eine Vorverarbeitungsfunktion auf den gesamten Datensatz anzuwenden:
-
-```py
->>> from transformers import AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
-
-
->>> def tokenize_function(examples):
-...     return tokenizer(examples["text"], padding="max_length", truncation=True)
-
-
->>> tokenized_datasets = dataset.map(tokenize_function, batched=True)
-```
-
-Wenn Sie möchten, können Sie eine kleinere Teilmenge des gesamten Datensatzes für die Feinabstimmung erstellen, um den Zeitaufwand zu verringern:
-
-```py
->>> small_train_dataset = tokenized_datasets["train"].shuffle(seed=42).select(range(1000))
->>> small_eval_dataset = tokenized_datasets["test"].shuffle(seed=42).select(range(1000))
-```
-
-<a id='trainer'></a>
-
-## Training
-
-An dieser Stelle sollten Sie dem Abschnitt folgen, der dem Rahmen entspricht, den Sie verwenden möchten. Sie können über die Links
-in der rechten Seitenleiste können Sie zu dem gewünschten Abschnitt springen - und wenn Sie den gesamten Inhalt eines bestimmten Frameworks ausblenden möchten,
-klicken Sie einfach auf die Schaltfläche oben rechts im Block des jeweiligen Frameworks!
-
-<frameworkcontent>
-<pt>
-<Youtube id="nvBXf7s7vTI"/>
-
-## Trainieren mit PyTorch Trainer
-
-🤗 Transformers bietet eine [`Trainer`]-Klasse, die für das Training von 🤗 Transformers-Modellen optimiert ist und es einfacher macht, mit dem Training zu beginnen, ohne manuell eine eigene Trainingsschleife zu schreiben. Die [`Trainer`]-API unterstützt eine breite Palette von Trainingsoptionen und Funktionen wie Logging, Gradientenakkumulation und gemischte Präzision.
-
-Beginnen Sie mit dem Laden Ihres Modells und geben Sie die Anzahl der erwarteten Labels an. Aus dem Yelp Review [dataset card](https://huggingface.co/datasets/yelp_review_full#data-fields) wissen Sie, dass es fünf Labels gibt:
-
-```py
->>> from transformers import AutoModelForSequenceClassification
-
->>> model = AutoModelForSequenceClassification.from_pretrained("bert-base-cased", num_labels=5)
-```
-
-<Tip>
-
-Es wird eine Warnung angezeigt, dass einige der trainierten Parameter nicht verwendet werden und einige Parameter zufällig
-initialisiert werden. Machen Sie sich keine Sorgen, das ist völlig normal! Der vorher trainierte Kopf des BERT-Modells wird verworfen und durch einen zufällig initialisierten Klassifikationskopf ersetzt. Sie werden diesen neuen Modellkopf in Ihrer Sequenzklassifizierungsaufgabe feinabstimmen, indem Sie das Wissen des vortrainierten Modells auf ihn übertragen.
-
-</Tip>
-
-### Hyperparameter für das Training
-
-Als Nächstes erstellen Sie eine Klasse [`TrainingArguments`], die alle Hyperparameter enthält, die Sie einstellen können, sowie Flags zur Aktivierung verschiedener Trainingsoptionen. Für dieses Lernprogramm können Sie mit den Standard- [Hyperparametern](https://huggingface.co/docs/transformers/main_classes/trainer#transformers.TrainingArguments) beginnen, aber Sie können mit diesen experimentieren, um Ihre optimalen Einstellungen zu finden.
-
-Geben Sie an, wo die Kontrollpunkte Ihres Trainings gespeichert werden sollen:
-
-```py
->>> from transformers import TrainingArguments
-
->>> training_args = TrainingArguments(output_dir="test_trainer")
-```
-
-### Auswerten
-
-Der [`Trainer`] wertet die Leistung des Modells während des Trainings nicht automatisch aus. Sie müssen [`Trainer`] eine Funktion übergeben, um Metriken zu berechnen und zu berichten. Die [🤗 Evaluate](https://huggingface.co/docs/evaluate/index) Bibliothek bietet eine einfache [`accuracy`](https://huggingface.co/spaces/evaluate-metric/accuracy) Funktion, die Sie mit der [`evaluate.load`] Funktion laden können (siehe diese [quicktour](https://huggingface.co/docs/evaluate/a_quick_tour) für weitere Informationen):
-
-```py
->>> import numpy as np
->>> import evaluate
-
->>> metric = evaluate.load("accuracy")
-```
-
-Rufen Sie [`~evaluate.compute`] auf `metric` auf, um die Genauigkeit Ihrer Vorhersagen zu berechnen. Bevor Sie Ihre Vorhersagen an `compute` übergeben, müssen Sie die Vorhersagen in Logits umwandeln (denken Sie daran, dass alle 🤗 Transformers-Modelle Logits zurückgeben):
-
-```py
->>> def compute_metrics(eval_pred):
-...     logits, labels = eval_pred
-...     predictions = np.argmax(logits, axis=-1)
-...     return metric.compute(predictions=predictions, references=labels)
-```
-
-Wenn Sie Ihre Bewertungsmetriken während der Feinabstimmung überwachen möchten, geben Sie den Parameter `evaluation_strategy` in Ihren Trainingsargumenten an, um die Bewertungsmetrik am Ende jeder Epoche zu ermitteln:
-
-```py
->>> from transformers import TrainingArguments, Trainer
-
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
-```
-
-### Trainer
-
-Erstellen Sie ein [`Trainer`]-Objekt mit Ihrem Modell, Trainingsargumenten, Trainings- und Testdatensätzen und einer Evaluierungsfunktion:
-
-```py
->>> trainer = Trainer(
-...     model=model,
-...     args=training_args,
-...     train_dataset=small_train_dataset,
-...     eval_dataset=small_eval_dataset,
-...     compute_metrics=compute_metrics,
-... )
-```
-
-Anschließend können Sie Ihr Modell durch den Aufruf von [`~transformers.Trainer.train`] optimieren:
-
-```py
->>> trainer.train()
-```
-</pt>
-<tf>
-<a id='keras'></a>
-
-<Youtube id="rnTGBy2ax1c"/>
-
-## Trainieren Sie ein TensorFlow-Modell mit Keras
-
-Sie können auch 🤗 Transformers Modelle in TensorFlow mit der Keras API trainieren!
-
-### Laden von Daten für Keras
-
-Wenn Sie ein 🤗 Transformers Modell mit der Keras API trainieren wollen, müssen Sie Ihren Datensatz in ein Format konvertieren, das
-Keras versteht. Wenn Ihr Datensatz klein ist, können Sie das Ganze einfach in NumPy-Arrays konvertieren und an Keras übergeben.
-Probieren wir das zuerst aus, bevor wir etwas Komplizierteres tun.
-
-Laden Sie zunächst ein Dataset. Wir werden den CoLA-Datensatz aus dem [GLUE-Benchmark](https://huggingface.co/datasets/glue) verwenden,
-da es sich um eine einfache Aufgabe zur Klassifizierung von binärem Text handelt, und nehmen vorerst nur den Trainingssplit.
-
-```py
-from datasets import load_dataset
-
-dataset = load_dataset("glue", "cola")
-dataset = dataset["train"]  # Just take the training split for now
-```
-
-Als nächstes laden Sie einen Tokenizer und tokenisieren die Daten als NumPy-Arrays. Beachten Sie, dass die Beschriftungen bereits eine Liste von 0 und 1en sind,
-Wir können sie also ohne Tokenisierung direkt in ein NumPy-Array konvertieren!
-
-```py
-from transformers import AutoTokenizer
-
-tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
-tokenized_data = tokenizer(dataset["text"], return_tensors="np", padding=True)
-
-labels = np.array(dataset["label"])  # Label is already an array of 0 and 1
-```
-
-Schließlich laden, [`compile`](https://keras.io/api/models/model_training_apis/#compile-method) und [`fit`](https://keras.io/api/models/model_training_apis/#fit-method) Sie das Modell:
-
-```py
-from transformers import TFAutoModelForSequenceClassification
-from tensorflow.keras.optimizers import Adam
-
-# Load and compile our model
-model = TFAutoModelForSequenceClassification.from_pretrained("bert-base-cased")
-# Lower learning rates are often better for fine-tuning transformers
-model.compile(optimizer=Adam(3e-5))
-
-model.fit(tokenized_data, labels)
-```
-
-<Tip>
-
-Sie müssen Ihren Modellen kein Verlustargument übergeben, wenn Sie sie `compile()`! Hugging-Face-Modelle wählen automatisch
-einen Loss, der für ihre Aufgabe und Modellarchitektur geeignet ist, wenn dieses Argument leer gelassen wird. Sie können jederzeit außer Kraft setzen, indem Sie selbst einen Loss angeben, wenn Sie das möchten!
-
-</Tip>
-
-Dieser Ansatz eignet sich hervorragend für kleinere Datensätze, aber bei größeren Datensätzen kann er zu einem Problem werden. Warum?
-Weil das tokenisierte Array und die Beschriftungen vollständig in den Speicher geladen werden müssten, und weil NumPy nicht mit
-"gezackte" Arrays nicht verarbeiten kann, so dass jedes tokenisierte Sample auf die Länge des längsten Samples im gesamten Datensatz aufgefüllt werden müsste.
-Datensatzes aufgefüllt werden. Dadurch wird das Array noch größer, und all die aufgefüllten Token verlangsamen auch das Training!
-
-### Laden von Daten als tf.data.Dataset
-
-Wenn Sie eine Verlangsamung des Trainings vermeiden wollen, können Sie Ihre Daten stattdessen als `tf.data.Dataset` laden. Sie können zwar Ihre eigene
-tf.data"-Pipeline schreiben können, wenn Sie wollen, haben wir zwei bequeme Methoden, um dies zu tun:
-
-- [`~TFPreTrainedModel.prepare_tf_dataset`]: Dies ist die Methode, die wir in den meisten Fällen empfehlen. Da es sich um eine Methode
-Ihres Modells ist, kann sie das Modell inspizieren, um automatisch herauszufinden, welche Spalten als Modelleingaben verwendet werden können, und
-verwirft die anderen, um einen einfacheren, leistungsfähigeren Datensatz zu erstellen.
-- [~datasets.Dataset.to_tf_dataset`]: Diese Methode ist eher auf niedriger Ebene angesiedelt und ist nützlich, wenn Sie genau kontrollieren wollen, wie
-Dataset erstellt wird, indem man genau angibt, welche `columns` und `label_cols` einbezogen werden sollen.
-
-Bevor Sie [~TFPreTrainedModel.prepare_tf_dataset`] verwenden können, müssen Sie die Tokenizer-Ausgaben als Spalten zu Ihrem Datensatz hinzufügen, wie in
-dem folgenden Codebeispiel:
-
-```py
-def tokenize_dataset(data):
-    # Keys of the returned dictionary will be added to the dataset as columns
-    return tokenizer(data["text"])
-
-
-dataset = dataset.map(tokenize_dataset)
-```
-
-Denken Sie daran, dass Hugging Face-Datensätze standardmäßig auf der Festplatte gespeichert werden, so dass dies nicht zu einem erhöhten Arbeitsspeicherbedarf führen wird! Sobald die
-Spalten hinzugefügt wurden, können Sie Batches aus dem Datensatz streamen und zu jedem Batch Auffüllungen hinzufügen, was die Anzahl der Auffüllungs-Token im Vergleich zum Auffüllen des gesamten Datensatzes reduziert.
-
-
-```py
->>> tf_dataset = model.prepare_tf_dataset(dataset, batch_size=16, shuffle=True, tokenizer=tokenizer)
-```
-
-Beachten Sie, dass Sie im obigen Codebeispiel den Tokenizer an `prepare_tf_dataset` übergeben müssen, damit die Stapel beim Laden korrekt aufgefüllt werden können.
-Wenn alle Stichproben in Ihrem Datensatz die gleiche Länge haben und kein Auffüllen erforderlich ist, können Sie dieses Argument weglassen.
-Wenn Sie etwas Komplexeres als nur das Auffüllen von Stichproben benötigen (z. B. das Korrumpieren von Token für die maskierte Sprachmodellierung), können Sie das Argument
-Modellierung), können Sie stattdessen das Argument `collate_fn` verwenden, um eine Funktion zu übergeben, die aufgerufen wird, um die
-Liste von Stichproben in einen Stapel umwandelt und alle gewünschten Vorverarbeitungen vornimmt. Siehe unsere
-[examples](https://github.com/huggingface/transformers/tree/main/examples) oder
-[notebooks](https://huggingface.co/docs/transformers/notebooks), um diesen Ansatz in Aktion zu sehen.
-
-Sobald Sie einen `tf.data.Dataset` erstellt haben, können Sie das Modell wie zuvor kompilieren und anpassen:
-
-```py
-model.compile(optimizer=Adam(3e-5))
-
-model.fit(tf_dataset)
-```
-
-</tf>
-</frameworkcontent>
-
-<a id='pytorch_native'></a>
-
-## Trainieren in nativem PyTorch
-
-<frameworkcontent>
-<pt>
-<Youtube id="Dh9CL8fyG80"/>
-
-[`Trainer`] kümmert sich um die Trainingsschleife und ermöglicht die Feinabstimmung eines Modells in einer einzigen Codezeile. Für Benutzer, die es vorziehen, ihre eigene Trainingsschleife zu schreiben, können Sie auch eine Feinabstimmung eines 🤗 Transformers-Modells in nativem PyTorch vornehmen.
-
-An diesem Punkt müssen Sie möglicherweise Ihr Notebook neu starten oder den folgenden Code ausführen, um etwas Speicher freizugeben:
-
-```py
-del model
-del pytorch_model
-del trainer
-torch.cuda.empty_cache()
-```
-
-Als Nächstes müssen Sie den Datensatz `tokenized_dataset` manuell nachbearbeiten, um ihn für das Training vorzubereiten.
-
-1. Entfernen Sie die Spalte "Text", da das Modell keinen Rohtext als Eingabe akzeptiert:
-
-    ```py
-    >>> tokenized_datasets = tokenized_datasets.remove_columns(["text"])
-    ```
-
-2. Benennen Sie die Spalte "Label" in "Labels" um, da das Modell erwartet, dass das Argument "Labels" genannt wird:
-
-    ```py
-    >>> tokenized_datasets = tokenized_datasets.rename_column("label", "labels")
-    ```
-
-3. Stellen Sie das Format des Datensatzes so ein, dass PyTorch-Tensoren anstelle von Listen zurückgegeben werden:
-
-    ```py
-    >>> tokenized_datasets.set_format("torch")
-    ```
-
-Erstellen Sie dann eine kleinere Teilmenge des Datensatzes, wie zuvor gezeigt, um die Feinabstimmung zu beschleunigen:
-
-```py
->>> small_train_dataset = tokenized_datasets["train"].shuffle(seed=42).select(range(1000))
->>> small_eval_dataset = tokenized_datasets["test"].shuffle(seed=42).select(range(1000))
-```
-
-### DataLoader
-
-Erstellen Sie einen `DataLoader` für Ihre Trainings- und Testdatensätze, damit Sie über die Datenstapel iterieren können:
-
-```py
->>> from torch.utils.data import DataLoader
-
->>> train_dataloader = DataLoader(small_train_dataset, shuffle=True, batch_size=8)
->>> eval_dataloader = DataLoader(small_eval_dataset, batch_size=8)
-```
-
-Laden Sie Ihr Modell mit der Anzahl der erwarteten Kennzeichnungen:
-
-```py
->>> from transformers import AutoModelForSequenceClassification
-
->>> model = AutoModelForSequenceClassification.from_pretrained("bert-base-cased", num_labels=5)
-```
-
-### Optimierer und Lernratensteuerung
-
-Erstellen Sie einen Optimierer und einen Scheduler für die Lernrate, um das Modell fein abzustimmen. Wir verwenden den Optimierer [`AdamW`](https://pytorch.org/docs/stable/generated/torch.optim.AdamW.html) aus PyTorch:
-
-```py
->>> from torch.optim import AdamW
-
->>> optimizer = AdamW(model.parameters(), lr=5e-5)
-```
-
-Erstellen Sie den Standard-Lernratenplaner aus [`Trainer`]:
-
-```py
->>> from transformers import get_scheduler
-
->>> num_epochs = 3
->>> num_training_steps = num_epochs * len(train_dataloader)
->>> lr_scheduler = get_scheduler(
-...     name="linear", optimizer=optimizer, num_warmup_steps=0, num_training_steps=num_training_steps
-... )
-```
-
-Geben Sie schließlich `device` an, um einen Grafikprozessor zu verwenden, wenn Sie Zugang zu einem solchen haben. Andernfalls kann das Training auf einer CPU mehrere Stunden statt ein paar Minuten dauern.
-
-```py
->>> import torch
-
->>> device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
->>> model.to(device)
-```
-
-<Tip>
-
-Holen Sie sich mit einem gehosteten Notebook wie [Colaboratory](https://colab.research.google.com/) oder [SageMaker StudioLab](https://studiolab.sagemaker.aws/) kostenlosen Zugang zu einem Cloud-GPU, wenn Sie noch keinen haben.
-
-</Tip>
-
-Großartig, Sie sind bereit für das Training! 🥳 
-
-### Trainingsschleife
-
-Um Ihren Trainingsfortschritt zu verfolgen, verwenden Sie die [tqdm](https://tqdm.github.io/) Bibliothek, um einen Fortschrittsbalken über die Anzahl der Trainingsschritte hinzuzufügen:
-
-```py
->>> from tqdm.auto import tqdm
-
->>> progress_bar = tqdm(range(num_training_steps))
-
->>> model.train()
->>> for epoch in range(num_epochs):
-...     for batch in train_dataloader:
-...         batch = {k: v.to(device) for k, v in batch.items()}
-...         outputs = model(**batch)
-...         loss = outputs.loss
-...         loss.backward()
-
-...         optimizer.step()
-...         lr_scheduler.step()
-...         optimizer.zero_grad()
-...         progress_bar.update(1)
-```
-
-### Auswertung
-
-Genauso wie Sie eine Bewertungsfunktion zu [`Trainer`] hinzugefügt haben, müssen Sie dasselbe tun, wenn Sie Ihre eigene Trainingsschleife schreiben. Aber anstatt die Metrik am Ende jeder Epoche zu berechnen und zu melden, werden Sie dieses Mal alle Stapel mit [`~evaluate.add_batch`] akkumulieren und die Metrik ganz am Ende berechnen.
-
-```py
->>> import evaluate
-
->>> metric = evaluate.load("accuracy")
->>> model.eval()
->>> for batch in eval_dataloader:
-...     batch = {k: v.to(device) for k, v in batch.items()}
-...     with torch.no_grad():
-...         outputs = model(**batch)
-
-...     logits = outputs.logits
-...     predictions = torch.argmax(logits, dim=-1)
-...     metric.add_batch(predictions=predictions, references=batch["labels"])
-
->>> metric.compute()
-```
-</pt>
-</frameworkcontent>
-
-<a id='additional-resources'></a>
-
-## Zusätzliche Ressourcen
-
-Weitere Beispiele für die Feinabstimmung finden Sie unter:
-
-- [🤗 Transformers Examples](https://github.com/huggingface/transformers/tree/main/examples) enthält Skripte
-  um gängige NLP-Aufgaben in PyTorch und TensorFlow zu trainieren.
-
-- [🤗 Transformers Notebooks](notebooks) enthält verschiedene Notebooks zur Feinabstimmung eines Modells für bestimmte Aufgaben in PyTorch und TensorFlow.

docs/source/debugging.rst

@@ -0,0 +1,295 @@
+..
+    Copyright 2021 The HuggingFace Team. All rights reserved.
+
+    Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+    the License. You may obtain a copy of the License at
+
+        http://www.apache.org/licenses/LICENSE-2.0
+
+    Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+    an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+    specific language governing permissions and limitations under the License.
+
+
+
+Debugging
+=======================================================================================================================
+
+Underflow and Overflow Detection
+-----------------------------------------------------------------------------------------------------------------------
+
+.. note::
+
+   This feature is currently available for PyTorch-only.
+
+.. note::
+
+   This feature can be used with any ``nn.Module``-based model
+
+If you start getting ``loss=NaN`` or the model inhibits some other abnormal behavior due to ``inf`` or ``nan`` in
+activations or weights one needs to discover where the first underflow or overflow happens and what led to it. Luckily
+you can accomplish that easily by activating a special module that will do the detection automatically.
+
+If you're using :class:`~transformers.Trainer`, you just need to add:
+
+.. code-block:: bash
+
+    --debug underflow_overflow
+
+to the normal command line arguments, or pass ``debug="underflow_overflow"`` when creating the
+:class:`~transformers.TrainingArguments` object.
+
+If you're using your own training loop or another Trainer you can accomplish the same with:
+
+.. code-block:: python
+
+    from .debug_utils import DebugUnderflowOverflow
+    debug_overflow = DebugUnderflowOverflow(model)
+
+:class:`~transformers.debug_utils.DebugUnderflowOverflow` inserts hooks into the model that immediately after each
+forward call will test input and output variables and also the corresponding module's weights. As soon as ``inf`` or
+``nan`` is detected in at least one element of the activations or weights, the program will assert and print a report
+like this (this was caught with ``google/mt5-small`` under fp16 mixed precision):
+
+.. code-block::
+
+    Detected inf/nan during batch_number=0
+    Last 21 forward frames:
+    abs min  abs max  metadata
+                      encoder.block.1.layer.1.DenseReluDense.dropout Dropout
+    0.00e+00 2.57e+02 input[0]
+    0.00e+00 2.85e+02 output
+    [...]
+                      encoder.block.2.layer.0 T5LayerSelfAttention
+    6.78e-04 3.15e+03 input[0]
+    2.65e-04 3.42e+03 output[0]
+                 None output[1]
+    2.25e-01 1.00e+04 output[2]
+                      encoder.block.2.layer.1.layer_norm T5LayerNorm
+    8.69e-02 4.18e-01 weight
+    2.65e-04 3.42e+03 input[0]
+    1.79e-06 4.65e+00 output
+                      encoder.block.2.layer.1.DenseReluDense.wi_0 Linear
+    2.17e-07 4.50e+00 weight
+    1.79e-06 4.65e+00 input[0]
+    2.68e-06 3.70e+01 output
+                      encoder.block.2.layer.1.DenseReluDense.wi_1 Linear
+    8.08e-07 2.66e+01 weight
+    1.79e-06 4.65e+00 input[0]
+    1.27e-04 2.37e+02 output
+                      encoder.block.2.layer.1.DenseReluDense.dropout Dropout
+    0.00e+00 8.76e+03 input[0]
+    0.00e+00 9.74e+03 output
+                      encoder.block.2.layer.1.DenseReluDense.wo Linear
+    1.01e-06 6.44e+00 weight
+    0.00e+00 9.74e+03 input[0]
+    3.18e-04 6.27e+04 output
+                      encoder.block.2.layer.1.DenseReluDense T5DenseGatedGeluDense
+    1.79e-06 4.65e+00 input[0]
+    3.18e-04 6.27e+04 output
+                      encoder.block.2.layer.1.dropout Dropout
+    3.18e-04 6.27e+04 input[0]
+    0.00e+00      inf output
+
+The example output has been trimmed in the middle for brevity.
+
+The second column shows the value of the absolute largest element, so if you have a closer look at the last few frames,
+the inputs and outputs were in the range of ``1e4``. So when this training was done under fp16 mixed precision the very
+last step overflowed (since under ``fp16`` the largest number before ``inf`` is ``64e3``). To avoid overflows under
+``fp16`` the activations must remain way below ``1e4``, because ``1e4 * 1e4 = 1e8`` so any matrix multiplication with
+large activations is going to lead to a numerical overflow condition.
+
+At the very start of the trace you can discover at which batch number the problem occurred (here ``Detected inf/nan
+during batch_number=0`` means the problem occurred on the first batch).
+
+Each reported frame starts by declaring the fully qualified entry for the corresponding module this frame is reporting
+for. If we look just at this frame:
+
+.. code-block::
+
+                      encoder.block.2.layer.1.layer_norm T5LayerNorm
+    8.69e-02 4.18e-01 weight
+    2.65e-04 3.42e+03 input[0]
+    1.79e-06 4.65e+00 output
+
+Here, ``encoder.block.2.layer.1.layer_norm`` indicates that it was a layer norm for the first layer, of the second
+block of the encoder. And the specific calls of the ``forward`` is ``T5LayerNorm``.
+
+Let's look at the last few frames of that report:
+
+.. code-block::
+
+        Detected inf/nan during batch_number=0
+        Last 21 forward frames:
+        abs min  abs max  metadata
+        [...]
+                          encoder.block.2.layer.1.DenseReluDense.wi_0 Linear
+        2.17e-07 4.50e+00 weight
+        1.79e-06 4.65e+00 input[0]
+        2.68e-06 3.70e+01 output
+                          encoder.block.2.layer.1.DenseReluDense.wi_1 Linear
+        8.08e-07 2.66e+01 weight
+        1.79e-06 4.65e+00 input[0]
+        1.27e-04 2.37e+02 output
+                          encoder.block.2.layer.1.DenseReluDense.wo Linear
+        1.01e-06 6.44e+00 weight
+        0.00e+00 9.74e+03 input[0]
+        3.18e-04 6.27e+04 output
+                          encoder.block.2.layer.1.DenseReluDense T5DenseGatedGeluDense
+        1.79e-06 4.65e+00 input[0]
+        3.18e-04 6.27e+04 output
+                          encoder.block.2.layer.1.dropout Dropout
+        3.18e-04 6.27e+04 input[0]
+        0.00e+00      inf output
+
+The last frame reports for ``Dropout.forward`` function with the first entry for the only input and the second for the
+only output. You can see that it was called from an attribute ``dropout`` inside ``DenseReluDense`` class. We can see
+that it happened during the first layer, of the 2nd block, during the very first batch. Finally, the absolute largest
+input elements was ``6.27e+04`` and same for the output was ``inf``.
+
+You can see here, that ``T5DenseGatedGeluDense.forward`` resulted in output activations, whose absolute max value was
+around 62.7K, which is very close to fp16's top limit of 64K. In the next frame we have ``Dropout`` which renormalizes
+the weights, after it zeroed some of the elements, which pushes the absolute max value to more than 64K, and we get an
+overlow (``inf``).
+
+As you can see it's the previous frames that we need to look into when the numbers start going into very large for fp16
+numbers.
+
+Let's match the report to the code from ``models/t5/modeling_t5.py``:
+
+.. code-block:: python
+
+    class T5DenseGatedGeluDense(nn.Module):
+        def __init__(self, config):
+            super().__init__()
+            self.wi_0 = nn.Linear(config.d_model, config.d_ff, bias=False)
+            self.wi_1 = nn.Linear(config.d_model, config.d_ff, bias=False)
+            self.wo = nn.Linear(config.d_ff, config.d_model, bias=False)
+            self.dropout = nn.Dropout(config.dropout_rate)
+            self.gelu_act = ACT2FN["gelu_new"]
+
+        def forward(self, hidden_states):
+            hidden_gelu = self.gelu_act(self.wi_0(hidden_states))
+            hidden_linear = self.wi_1(hidden_states)
+            hidden_states = hidden_gelu * hidden_linear
+            hidden_states = self.dropout(hidden_states)
+            hidden_states = self.wo(hidden_states)
+            return hidden_states
+
+Now it's easy to see the ``dropout`` call, and all the previous calls as well.
+
+Since the detection is happening in a forward hook, these reports are printed immediately after each ``forward``
+returns.
+
+Going back to the full report, to act on it and to fix the problem, we need to go a few frames up where the numbers
+started to go up and most likely switch to the ``fp32`` mode here, so that the numbers don't overflow when multiplied
+or summed up. Of course, there might be other solutions. For example, we could turn off ``amp`` temporarily if it's
+enabled, after moving the original ``forward`` into a helper wrapper, like so:
+
+.. code-block:: python
+
+    def _forward(self, hidden_states):
+        hidden_gelu = self.gelu_act(self.wi_0(hidden_states))
+        hidden_linear = self.wi_1(hidden_states)
+        hidden_states = hidden_gelu * hidden_linear
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.wo(hidden_states)
+        return hidden_states
+
+    import torch
+    def forward(self, hidden_states):
+        if torch.is_autocast_enabled():
+             with torch.cuda.amp.autocast(enabled=False):
+                 return self._forward(hidden_states)
+         else:
+             return self._forward(hidden_states)
+
+Since the automatic detector only reports on inputs and outputs of full frames, once you know where to look, you may
+want to analyse the intermediary stages of any specific ``forward`` function as well. In such a case you can use the
+``detect_overflow`` helper function to inject the detector where you want it, for example:
+
+.. code-block:: python
+
+    from debug_utils import detect_overflow
+
+    class T5LayerFF(nn.Module):
+        [...]
+        def forward(self, hidden_states):
+            forwarded_states = self.layer_norm(hidden_states)
+            detect_overflow(forwarded_states, "after layer_norm")
+            forwarded_states = self.DenseReluDense(forwarded_states)
+            detect_overflow(forwarded_states, "after DenseReluDense")
+            return hidden_states + self.dropout(forwarded_states)
+
+You can see that we added 2 of these and now we track if ``inf`` or ``nan`` for ``forwarded_states`` was detected
+somewhere in between.
+
+Actually, the detector already reports these because each of the calls in the example above is a `nn.Module``, but
+let's say if you had some local direct calculations this is how you'd do that.
+
+Additionally, if you're instantiating the debugger in your own code, you can adjust the number of frames printed from
+its default, e.g.:
+
+.. code-block:: python
+
+    from .debug_utils import DebugUnderflowOverflow
+    debug_overflow = DebugUnderflowOverflow(model, max_frames_to_save=100)
+
+Specific batch absolute mix and max value tracing
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The same debugging class can be used for per-batch tracing with the underflow/overflow detection feature turned off.
+
+Let's say you want to watch the absolute min and max values for all the ingredients of each ``forward`` call of a given
+batch, and only do that for batches 1 and 3. Then you instantiate this class as:
+
+.. code-block:: python
+
+    debug_overflow = DebugUnderflowOverflow(model, trace_batch_nums=[1,3])
+
+And now full batches 1 and 3 will be traced using the same format as the underflow/overflow detector does.
+
+Batches are 0-indexed.
+
+This is helpful if you know that the program starts misbehaving after a certain batch number, so you can fast-forward
+right to that area. Here is a sample truncated output for such configuration:
+
+.. code-block::
+
+                      *** Starting batch number=1 ***
+    abs min  abs max  metadata
+                      shared Embedding
+    1.01e-06 7.92e+02 weight
+    0.00e+00 2.47e+04 input[0]
+    5.36e-05 7.92e+02 output
+    [...]
+                      decoder.dropout Dropout
+    1.60e-07 2.27e+01 input[0]
+    0.00e+00 2.52e+01 output
+                      decoder T5Stack
+         not a tensor output
+                      lm_head Linear
+    1.01e-06 7.92e+02 weight
+    0.00e+00 1.11e+00 input[0]
+    6.06e-02 8.39e+01 output
+                       T5ForConditionalGeneration
+         not a tensor output
+
+                      *** Starting batch number=3 ***
+    abs min  abs max  metadata
+                      shared Embedding
+    1.01e-06 7.92e+02 weight
+    0.00e+00 2.78e+04 input[0]
+    5.36e-05 7.92e+02 output
+    [...]
+
+Here you will get a huge number of frames dumped - as many as there were forward calls in your model, so it may or may
+not what you want, but sometimes it can be easier to use for debugging purposes than a normal debugger. For example, if
+a problem starts happening at batch number 150. So you can dump traces for batches 149 and 150 and compare where
+numbers started to diverge.
+
+You can also specify the batch number after which to stop the training, with:
+
+.. code-block:: python
+
+    debug_overflow = DebugUnderflowOverflow(model, trace_batch_nums=[1,3], abort_after_batch_num=3)

docs/source/en/_config.py

@@ -1,14 +0,0 @@
-# docstyle-ignore
-INSTALL_CONTENT = """
-# Transformers installation
-! pip install transformers datasets
-# To install from source instead of the last release, comment the command above and uncomment the following one.
-# ! pip install git+https://github.com/huggingface/transformers.git
-"""
-
-notebook_first_cells = [{"type": "code", "content": INSTALL_CONTENT}]
-black_avoid_patterns = {
-    "{processor_class}": "FakeProcessorClass",
-    "{model_class}": "FakeModelClass",
-    "{object_class}": "FakeObjectClass",    
-}

docs/source/en/_toctree.yml

@@ -1,529 +0,0 @@
-- sections:
-  - local: index
-    title: 🤗 Transformers
-  - local: quicktour
-    title: Quick tour
-  - local: installation
-    title: Installation
-  title: Get started
-- sections:
-  - local: pipeline_tutorial
-    title: Pipelines for inference
-  - local: autoclass_tutorial
-    title: Load pretrained instances with an AutoClass
-  - local: preprocessing
-    title: Preprocess
-  - local: training
-    title: Fine-tune a pretrained model
-  - local: accelerate
-    title: Distributed training with 🤗 Accelerate
-  - local: model_sharing
-    title: Share a model
-  title: Tutorials
-- sections:
-  - sections:
-    - local: create_a_model
-      title: Create a custom architecture
-    - local: custom_models
-      title: Sharing custom models
-    - local: run_scripts
-      title: Train with a script
-    - local: sagemaker
-      title: Run training on Amazon SageMaker
-    - local: converting_tensorflow_models
-      title: Converting from TensorFlow checkpoints
-    - local: serialization
-      title: Export to ONNX
-    - local: torchscript
-      title: Export to TorchScript
-    - local: troubleshooting
-      title: Troubleshoot
-    title: General usage
-  - sections:
-    - local: fast_tokenizers
-      title: Use tokenizers from 🤗 Tokenizers
-    - local: multilingual
-      title: Inference for multilingual models
-    - sections:
-      - local: tasks/sequence_classification
-        title: Text classification
-      - local: tasks/token_classification
-        title: Token classification
-      - local: tasks/question_answering
-        title: Question answering
-      - local: tasks/language_modeling
-        title: Language modeling
-      - local: tasks/translation
-        title: Translation
-      - local: tasks/summarization
-        title: Summarization
-      - local: tasks/multiple_choice
-        title: Multiple choice
-      title: Task guides
-      isExpanded: false
-    title: Natural Language Processing
-  - sections:
-    - local: tasks/audio_classification
-      title: Audio classification
-    - local: tasks/asr
-      title: Automatic speech recognition
-    title: Audio
-  - sections:
-    - local: tasks/image_classification
-      title: Image classification
-    - local: tasks/semantic_segmentation
-      title: Semantic segmentation
-    title: Computer Vision
-  - sections:
-    - local: performance
-      title: Overview
-    - local: perf_train_gpu_one
-      title: Training on one GPU
-    - local: perf_train_gpu_many
-      title: Training on many GPUs
-    - local: perf_train_cpu
-      title: Training on CPU
-    - local: perf_train_cpu_many
-      title: Training on many CPUs
-    - local: perf_train_tpu
-      title: Training on TPUs
-    - local: perf_train_special
-      title: Training on Specialized Hardware
-    - local: perf_infer_cpu
-      title: Inference on CPU
-    - local: perf_infer_gpu_one
-      title: Inference on one GPU
-    - local: perf_infer_gpu_many
-      title: Inference on many GPUs
-    - local: perf_infer_special
-      title: Inference on Specialized Hardware
-    - local: perf_hardware
-      title: Custom hardware for training
-    - local: big_models
-      title: Instantiating a big model
-    - local: debugging
-      title: Debugging
-    - local: hpo_train
-      title: Hyperparameter Search using Trainer API
-    title: Performance and scalability
-  - sections:
-    - local: contributing
-      title: How to contribute to transformers?
-    - local: add_new_model
-      title: How to add a model to 🤗 Transformers?
-    - local: add_tensorflow_model
-      title: How to convert a 🤗 Transformers model to TensorFlow?
-    - local: add_new_pipeline
-      title: How to add a pipeline to 🤗 Transformers?
-    - local: testing
-      title: Testing
-    - local: pr_checks
-      title: Checks on a Pull Request
-    title: Contribute
-  - local: notebooks
-    title: 🤗 Transformers Notebooks
-  - local: community
-    title: Community resources
-  - local: benchmarks
-    title: Benchmarks
-  - local: migration
-    title: Migrating from previous packages
-  title: How-to guides
-- sections:
-  - local: philosophy
-    title: Philosophy
-  - local: glossary
-    title: Glossary
-  - local: task_summary
-    title: Summary of the tasks
-  - local: model_summary
-    title: Summary of the models
-  - local: tokenizer_summary
-    title: Summary of the tokenizers
-  - local: pad_truncation
-    title: Padding and truncation
-  - local: bertology
-    title: BERTology
-  - local: perplexity
-    title: Perplexity of fixed-length models
-  title: Conceptual guides
-- sections:
-  - sections:
-    - local: model_doc/auto
-      title: Auto Classes
-    - local: main_classes/callback
-      title: Callbacks
-    - local: main_classes/configuration
-      title: Configuration
-    - local: main_classes/data_collator
-      title: Data Collator
-    - local: main_classes/keras_callbacks
-      title: Keras callbacks
-    - local: main_classes/logging
-      title: Logging
-    - local: main_classes/model
-      title: Models
-    - local: main_classes/text_generation
-      title: Text Generation
-    - local: main_classes/onnx
-      title: ONNX
-    - local: main_classes/optimizer_schedules
-      title: Optimization
-    - local: main_classes/output
-      title: Model outputs
-    - local: main_classes/pipelines
-      title: Pipelines
-    - local: main_classes/processors
-      title: Processors
-    - local: main_classes/tokenizer
-      title: Tokenizer
-    - local: main_classes/trainer
-      title: Trainer
-    - local: main_classes/deepspeed
-      title: DeepSpeed Integration
-    - local: main_classes/feature_extractor
-      title: Feature Extractor
-    title: Main Classes
-  - sections:
-    - isExpanded: false
-      sections:
-      - local: model_doc/albert
-        title: ALBERT
-      - local: model_doc/bart
-        title: BART
-      - local: model_doc/barthez
-        title: BARThez
-      - local: model_doc/bartpho
-        title: BARTpho
-      - local: model_doc/bert
-        title: BERT
-      - local: model_doc/bert-generation
-        title: BertGeneration
-      - local: model_doc/bert-japanese
-        title: BertJapanese
-      - local: model_doc/bertweet
-        title: Bertweet
-      - local: model_doc/big_bird
-        title: BigBird
-      - local: model_doc/bigbird_pegasus
-        title: BigBirdPegasus
-      - local: model_doc/blenderbot
-        title: Blenderbot
-      - local: model_doc/blenderbot-small
-        title: Blenderbot Small
-      - local: model_doc/bloom
-        title: BLOOM
-      - local: model_doc/bort
-        title: BORT
-      - local: model_doc/byt5
-        title: ByT5
-      - local: model_doc/camembert
-        title: CamemBERT
-      - local: model_doc/canine
-        title: CANINE
-      - local: model_doc/codegen
-        title: CodeGen
-      - local: model_doc/convbert
-        title: ConvBERT
-      - local: model_doc/cpm
-        title: CPM
-      - local: model_doc/ctrl
-        title: CTRL
-      - local: model_doc/deberta
-        title: DeBERTa
-      - local: model_doc/deberta-v2
-        title: DeBERTa-v2
-      - local: model_doc/dialogpt
-        title: DialoGPT
-      - local: model_doc/distilbert
-        title: DistilBERT
-      - local: model_doc/dpr
-        title: DPR
-      - local: model_doc/electra
-        title: ELECTRA
-      - local: model_doc/encoder-decoder
-        title: Encoder Decoder Models
-      - local: model_doc/ernie
-        title: ERNIE
-      - local: model_doc/esm
-        title: ESM
-      - local: model_doc/flaubert
-        title: FlauBERT
-      - local: model_doc/fnet
-        title: FNet
-      - local: model_doc/fsmt
-        title: FSMT
-      - local: model_doc/funnel
-        title: Funnel Transformer
-      - local: model_doc/openai-gpt
-        title: GPT
-      - local: model_doc/gpt_neo
-        title: GPT Neo
-      - local: model_doc/gpt_neox
-        title: GPT NeoX
-      - local: model_doc/gpt_neox_japanese
-        title: GPT NeoX Japanese
-      - local: model_doc/gptj
-        title: GPT-J
-      - local: model_doc/gpt2
-        title: GPT2
-      - local: model_doc/herbert
-        title: HerBERT
-      - local: model_doc/ibert
-        title: I-BERT
-      - local: model_doc/layoutlm
-        title: LayoutLM
-      - local: model_doc/led
-        title: LED
-      - local: model_doc/lilt
-        title: LiLT
-      - local: model_doc/longformer
-        title: Longformer
-      - local: model_doc/longt5
-        title: LongT5
-      - local: model_doc/luke
-        title: LUKE
-      - local: model_doc/m2m_100
-        title: M2M100
-      - local: model_doc/marian
-        title: MarianMT
-      - local: model_doc/markuplm
-        title: MarkupLM
-      - local: model_doc/mbart
-        title: MBart and MBart-50
-      - local: model_doc/megatron-bert
-        title: MegatronBERT
-      - local: model_doc/megatron_gpt2
-        title: MegatronGPT2
-      - local: model_doc/mluke
-        title: mLUKE
-      - local: model_doc/mobilebert
-        title: MobileBERT
-      - local: model_doc/mpnet
-        title: MPNet
-      - local: model_doc/mt5
-        title: MT5
-      - local: model_doc/mvp
-        title: MVP
-      - local: model_doc/nezha
-        title: NEZHA
-      - local: model_doc/nllb
-        title: NLLB
-      - local: model_doc/nystromformer
-        title: Nyströmformer
-      - local: model_doc/opt
-        title: OPT
-      - local: model_doc/pegasus
-        title: Pegasus
-      - local: model_doc/pegasus_x
-        title: PEGASUS-X
-      - local: model_doc/phobert
-        title: PhoBERT
-      - local: model_doc/plbart
-        title: PLBart
-      - local: model_doc/prophetnet
-        title: ProphetNet
-      - local: model_doc/qdqbert
-        title: QDQBert
-      - local: model_doc/rag
-        title: RAG
-      - local: model_doc/realm
-        title: REALM
-      - local: model_doc/reformer
-        title: Reformer
-      - local: model_doc/rembert
-        title: RemBERT
-      - local: model_doc/retribert
-        title: RetriBERT
-      - local: model_doc/roberta
-        title: RoBERTa
-      - local: model_doc/roformer
-        title: RoFormer
-      - local: model_doc/splinter
-        title: Splinter
-      - local: model_doc/squeezebert
-        title: SqueezeBERT
-      - local: model_doc/t5
-        title: T5
-      - local: model_doc/t5v1.1
-        title: T5v1.1
-      - local: model_doc/tapas
-        title: TAPAS
-      - local: model_doc/tapex
-        title: TAPEX
-      - local: model_doc/transfo-xl
-        title: Transformer XL
-      - local: model_doc/ul2
-        title: UL2
-      - local: model_doc/xglm
-        title: XGLM
-      - local: model_doc/xlm
-        title: XLM
-      - local: model_doc/xlm-prophetnet
-        title: XLM-ProphetNet
-      - local: model_doc/xlm-roberta
-        title: XLM-RoBERTa
-      - local: model_doc/xlm-roberta-xl
-        title: XLM-RoBERTa-XL
-      - local: model_doc/xlnet
-        title: XLNet
-      - local: model_doc/yoso
-        title: YOSO
-      title: Text models
-    - isExpanded: false
-      sections:
-      - local: model_doc/beit
-        title: BEiT
-      - local: model_doc/conditional_detr
-        title: Conditional DETR
-      - local: model_doc/convnext
-        title: ConvNeXT
-      - local: model_doc/cvt
-        title: CvT
-      - local: model_doc/deformable_detr
-        title: Deformable DETR
-      - local: model_doc/deit
-        title: DeiT
-      - local: model_doc/detr
-        title: DETR
-      - local: model_doc/dit
-        title: DiT
-      - local: model_doc/dpt
-        title: DPT
-      - local: model_doc/glpn
-        title: GLPN
-      - local: model_doc/imagegpt
-        title: ImageGPT
-      - local: model_doc/levit
-        title: LeViT
-      - local: model_doc/maskformer
-        title: MaskFormer
-      - local: model_doc/mobilevit
-        title: MobileViT
-      - local: model_doc/poolformer
-        title: PoolFormer
-      - local: model_doc/regnet
-        title: RegNet
-      - local: model_doc/resnet
-        title: ResNet
-      - local: model_doc/segformer
-        title: SegFormer
-      - local: model_doc/swin
-        title: Swin Transformer
-      - local: model_doc/swinv2
-        title: Swin Transformer V2
-      - local: model_doc/van
-        title: VAN
-      - local: model_doc/videomae
-        title: VideoMAE
-      - local: model_doc/vit
-        title: Vision Transformer (ViT)
-      - local: model_doc/vit_mae
-        title: ViTMAE
-      - local: model_doc/vit_msn
-        title: ViTMSN
-      - local: model_doc/yolos
-        title: YOLOS
-      title: Vision models
-    - isExpanded: false
-      sections:
-      - local: model_doc/hubert
-        title: Hubert
-      - local: model_doc/mctct
-        title: MCTCT
-      - local: model_doc/sew
-        title: SEW
-      - local: model_doc/sew-d
-        title: SEW-D
-      - local: model_doc/speech_to_text
-        title: Speech2Text
-      - local: model_doc/speech_to_text_2
-        title: Speech2Text2
-      - local: model_doc/unispeech
-        title: UniSpeech
-      - local: model_doc/unispeech-sat
-        title: UniSpeech-SAT
-      - local: model_doc/wav2vec2
-        title: Wav2Vec2
-      - local: model_doc/wav2vec2-conformer
-        title: Wav2Vec2-Conformer
-      - local: model_doc/wav2vec2_phoneme
-        title: Wav2Vec2Phoneme
-      - local: model_doc/wavlm
-        title: WavLM
-      - local: model_doc/whisper
-        title: Whisper
-      - local: model_doc/xls_r
-        title: XLS-R
-      - local: model_doc/xlsr_wav2vec2
-        title: XLSR-Wav2Vec2
-      title: Audio models
-    - isExpanded: false
-      sections:
-      - local: model_doc/clip
-        title: CLIP
-      - local: model_doc/data2vec
-        title: Data2Vec
-      - local: model_doc/donut
-        title: Donut
-      - local: model_doc/flava
-        title: FLAVA
-      - local: model_doc/groupvit
-        title: GroupViT
-      - local: model_doc/layoutlmv2
-        title: LayoutLMV2
-      - local: model_doc/layoutlmv3
-        title: LayoutLMV3
-      - local: model_doc/layoutxlm
-        title: LayoutXLM
-      - local: model_doc/lxmert
-        title: LXMERT
-      - local: model_doc/owlvit
-        title: OWL-ViT
-      - local: model_doc/perceiver
-        title: Perceiver
-      - local: model_doc/speech-encoder-decoder
-        title: Speech Encoder Decoder Models
-      - local: model_doc/trocr
-        title: TrOCR
-      - local: model_doc/vilt
-        title: ViLT
-      - local: model_doc/vision-encoder-decoder
-        title: Vision Encoder Decoder Models
-      - local: model_doc/vision-text-dual-encoder
-        title: Vision Text Dual Encoder
-      - local: model_doc/visual_bert
-        title: VisualBERT
-      - local: model_doc/xclip
-        title: X-CLIP
-      title: Multimodal models
-    - isExpanded: false
-      sections:
-      - local: model_doc/decision_transformer
-        title: Decision Transformer
-      - local: model_doc/trajectory_transformer
-        title: Trajectory Transformer
-      title: Reinforcement learning models
-    - isExpanded: false
-      sections:
-      - local: model_doc/time_series_transformer
-        title: Time Series Transformer
-      title: Time series models
-    title: Models
-  - sections:
-    - local: internal/modeling_utils
-      title: Custom Layers and Utilities
-    - local: internal/pipelines_utils
-      title: Utilities for pipelines
-    - local: internal/tokenization_utils
-      title: Utilities for Tokenizers
-    - local: internal/trainer_utils
-      title: Utilities for Trainer
-    - local: internal/generation_utils
-      title: Utilities for Generation
-    - local: internal/image_processing_utils
-      title: Utilities for Image Processors
-    - local: internal/file_utils
-      title: General Utilities
-    title: Internal Helpers
-  title: API

docs/source/en/accelerate.mdx

@@ -1,132 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Distributed training with 🤗 Accelerate
-
-As models get bigger, parallelism has emerged as a strategy for training larger models on limited hardware and accelerating training speed by several orders of magnitude. At Hugging Face, we created the [🤗 Accelerate](https://huggingface.co/docs/accelerate) library to help users easily train a 🤗 Transformers model on any type of distributed setup, whether it is multiple GPU's on one machine or multiple GPU's across several machines. In this tutorial, learn how to customize your native PyTorch training loop to enable training in a distributed environment.
-
-## Setup
-
-Get started by installing 🤗 Accelerate:
-
-```bash
-pip install accelerate
-```
-
-Then import and create an [`~accelerate.Accelerator`] object. The [`~accelerate.Accelerator`] will automatically detect your type of distributed setup and initialize all the necessary components for training. You don't need to explicitly place your model on a device.
-
-```py
->>> from accelerate import Accelerator
-
->>> accelerator = Accelerator()
-```
-
-## Prepare to accelerate
-
-The next step is to pass all the relevant training objects to the [`~accelerate.Accelerator.prepare`] method. This includes your training and evaluation DataLoaders, a model and an optimizer:
-
-```py
->>> train_dataloader, eval_dataloader, model, optimizer = accelerator.prepare(
-...     train_dataloader, eval_dataloader, model, optimizer
-... )
-```
-
-## Backward
-
-The last addition is to replace the typical `loss.backward()` in your training loop with 🤗 Accelerate's [`~accelerate.Accelerator.backward`]method:
-
-```py
->>> for epoch in range(num_epochs):
-...     for batch in train_dataloader:
-...         outputs = model(**batch)
-...         loss = outputs.loss
-...         accelerator.backward(loss)
-
-...         optimizer.step()
-...         lr_scheduler.step()
-...         optimizer.zero_grad()
-...         progress_bar.update(1)
-```
-
-As you can see in the following code, you only need to add four additional lines of code to your training loop to enable distributed training!
-
-```diff
-+ from accelerate import Accelerator
-  from transformers import AdamW, AutoModelForSequenceClassification, get_scheduler
-
-+ accelerator = Accelerator()
-
-  model = AutoModelForSequenceClassification.from_pretrained(checkpoint, num_labels=2)
-  optimizer = AdamW(model.parameters(), lr=3e-5)
-
-- device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
-- model.to(device)
-
-+ train_dataloader, eval_dataloader, model, optimizer = accelerator.prepare(
-+     train_dataloader, eval_dataloader, model, optimizer
-+ )
-
-  num_epochs = 3
-  num_training_steps = num_epochs * len(train_dataloader)
-  lr_scheduler = get_scheduler(
-      "linear",
-      optimizer=optimizer,
-      num_warmup_steps=0,
-      num_training_steps=num_training_steps
-  )
-
-  progress_bar = tqdm(range(num_training_steps))
-
-  model.train()
-  for epoch in range(num_epochs):
-      for batch in train_dataloader:
--         batch = {k: v.to(device) for k, v in batch.items()}
-          outputs = model(**batch)
-          loss = outputs.loss
--         loss.backward()
-+         accelerator.backward(loss)
-
-          optimizer.step()
-          lr_scheduler.step()
-          optimizer.zero_grad()
-          progress_bar.update(1)
-```
-
-## Train
-
-Once you've added the relevant lines of code, launch your training in a script or a notebook like Colaboratory.
-
-### Train with a script
-
-If you are running your training from a script, run the following command to create and save a configuration file:
-
-```bash
-accelerate config
-```
-
-Then launch your training with:
-
-```bash
-accelerate launch train.py
-```
-
-### Train with a notebook
-
-🤗 Accelerate can also run in a notebook if you're planning on using Colaboratory's TPUs. Wrap all the code responsible for training in a function, and pass it to [`~accelerate.notebook_launcher`]:
-
-```py
->>> from accelerate import notebook_launcher
-
->>> notebook_launcher(training_function)
-```
-
-For more information about 🤗 Accelerate and it's rich features, refer to the [documentation](https://huggingface.co/docs/accelerate).

docs/source/en/add_new_pipeline.mdx

@@ -1,254 +0,0 @@
-<!--Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
--->
-
-# How to create a custom pipeline?
-
-In this guide, we will see how to create a custom pipeline and share it on the [Hub](hf.co/models) or add it to the
-Transformers library.
-
-First and foremost, you need to decide the raw entries the pipeline will be able to take. It can be strings, raw bytes,
-dictionaries or whatever seems to be the most likely desired input. Try to keep these inputs as pure Python as possible
-as it makes compatibility easier (even through other languages via JSON). Those will be the `inputs` of the
-pipeline (`preprocess`).
-
-Then define the `outputs`. Same policy as the `inputs`. The simpler, the better. Those will be the outputs of
-`postprocess` method.
-
-Start by inheriting the base class `Pipeline`. with the 4 methods needed to implement `preprocess`,
-`_forward`, `postprocess` and `_sanitize_parameters`.
-
-
-```python
-from transformers import Pipeline
-
-
-class MyPipeline(Pipeline):
-    def _sanitize_parameters(self, **kwargs):
-        preprocess_kwargs = {}
-        if "maybe_arg" in kwargs:
-            preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]
-        return preprocess_kwargs, {}, {}
-
-    def preprocess(self, inputs, maybe_arg=2):
-        model_input = Tensor(inputs["input_ids"])
-        return {"model_input": model_input}
-
-    def _forward(self, model_inputs):
-        # model_inputs == {"model_input": model_input}
-        outputs = self.model(**model_inputs)
-        # Maybe {"logits": Tensor(...)}
-        return outputs
-
-    def postprocess(self, model_outputs):
-        best_class = model_outputs["logits"].softmax(-1)
-        return best_class
-```
-
-The structure of this breakdown is to support relatively seamless support for CPU/GPU, while supporting doing
-pre/postprocessing on the CPU on different threads
-
-`preprocess` will take the originally defined inputs, and turn them into something feedable to the model. It might
-contain more information and is usually a `Dict`.
-
-`_forward` is the implementation detail and is not meant to be called directly. `forward` is the preferred
-called method as it contains safeguards to make sure everything is working on the expected device. If anything is
-linked to a real model it belongs in the `_forward` method, anything else is in the preprocess/postprocess.
-
-`postprocess` methods will take the output of `_forward` and turn it into the final output that were decided
-earlier.
-
-`_sanitize_parameters` exists to allow users to pass any parameters whenever they wish, be it at initialization
-time `pipeline(...., maybe_arg=4)` or at call time `pipe = pipeline(...); output = pipe(...., maybe_arg=4)`.
-
-The returns of `_sanitize_parameters` are the 3 dicts of kwargs that will be passed directly to `preprocess`,
-`_forward` and `postprocess`. Don't fill anything if the caller didn't call with any extra parameter. That
-allows to keep the default arguments in the function definition which is always more "natural".
-
-A classic example would be a `top_k` argument in the post processing in classification tasks.
-
-```python
->>> pipe = pipeline("my-new-task")
->>> pipe("This is a test")
-[{"label": "1-star", "score": 0.8}, {"label": "2-star", "score": 0.1}, {"label": "3-star", "score": 0.05}
-{"label": "4-star", "score": 0.025}, {"label": "5-star", "score": 0.025}]
-
->>> pipe("This is a test", top_k=2)
-[{"label": "1-star", "score": 0.8}, {"label": "2-star", "score": 0.1}]
-```
-
-In order to achieve that, we'll update our `postprocess` method with a default parameter to `5`. and edit
-`_sanitize_parameters` to allow this new parameter.
-
-
-```python
-def postprocess(self, model_outputs, top_k=5):
-    best_class = model_outputs["logits"].softmax(-1)
-    # Add logic to handle top_k
-    return best_class
-
-
-def _sanitize_parameters(self, **kwargs):
-    preprocess_kwargs = {}
-    if "maybe_arg" in kwargs:
-        preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]
-
-    postprocess_kwargs = {}
-    if "top_k" in kwargs:
-        postprocess_kwargs["top_k"] = kwargs["top_k"]
-    return preprocess_kwargs, {}, postprocess_kwargs
-```
-
-Try to keep the inputs/outputs very simple and ideally JSON-serializable as it makes the pipeline usage very easy
-without requiring users to understand new kind of objects. It's also relatively common to support many different types
-of arguments for ease of use (audio files, can be filenames, URLs or pure bytes)
-
-
-
-## Adding it to the list of supported tasks
-
-To register your `new-task` to the list of supported tasks, you have to add it to the `PIPELINE_REGISTRY`:
-
-```python
-from transformers.pipelines import PIPELINE_REGISTRY
-
-PIPELINE_REGISTRY.register_pipeline(
-    "new-task",
-    pipeline_class=MyPipeline,
-    pt_model=AutoModelForSequenceClassification,
-)
-```
-
-You can specify a default model if you want, in which case it should come with a specific revision (which can be the name of a branch or a commit hash, here we took `"abcdef"`) as well was the type:
-
-```python
-PIPELINE_REGISTRY.register_pipeline(
-    "new-task",
-    pipeline_class=MyPipeline,
-    pt_model=AutoModelForSequenceClassification,
-    default={"pt": ("user/awesome_model", "abcdef")},
-    type="text",  # current support type: text, audio, image, multimodal
-)
-```
-
-## Share your pipeline on the Hub
-
-To share your custom pipeline on the Hub, you just have to save the custom code of your `Pipeline` subclass in a
-python file. For instance, let's say we want to use a custom pipeline for sentence pair classification like this:
-
-```py
-import numpy as np
-
-from transformers import Pipeline
-
-
-def softmax(outputs):
-    maxes = np.max(outputs, axis=-1, keepdims=True)
-    shifted_exp = np.exp(outputs - maxes)
-    return shifted_exp / shifted_exp.sum(axis=-1, keepdims=True)
-
-
-class PairClassificationPipeline(Pipeline):
-    def _sanitize_parameters(self, **kwargs):
-        preprocess_kwargs = {}
-        if "second_text" in kwargs:
-            preprocess_kwargs["second_text"] = kwargs["second_text"]
-        return preprocess_kwargs, {}, {}
-
-    def preprocess(self, text, second_text=None):
-        return self.tokenizer(text, text_pair=second_text, return_tensors=self.framework)
-
-    def _forward(self, model_inputs):
-        return self.model(**model_inputs)
-
-    def postprocess(self, model_outputs):
-        logits = model_outputs.logits[0].numpy()
-        probabilities = softmax(logits)
-
-        best_class = np.argmax(probabilities)
-        label = self.model.config.id2label[best_class]
-        score = probabilities[best_class].item()
-        logits = logits.tolist()
-        return {"label": label, "score": score, "logits": logits}
-```
-
-The implementation is framework agnostic, and will work for PyTorch and TensorFlow models. If we have saved this in
-a file named `pair_classification.py`, we can then import it and register it like this:
-
-```py
-from pair_classification import PairClassificationPipeline
-from transformers.pipelines import PIPELINE_REGISTRY
-from transformers import AutoModelForSequenceClassification, TFAutoModelForSequenceClassification
-
-PIPELINE_REGISTRY.register_pipeline(
-    "pair-classification",
-    pipeline_class=PairClassificationPipeline,
-    pt_model=AutoModelForSequenceClassification,
-    tf_model=TFAutoModelForSequenceClassification,
-)
-```
-
-Once this is done, we can use it with a pretrained model. For instance `sgugger/finetuned-bert-mrpc` has been
-fine-tuned on the MRPC dataset, which classifies pairs of sentences as paraphrases or not.
-
-```py
-from transformers import pipeline
-
-classifier = pipeline("pair-classification", model="sgugger/finetuned-bert-mrpc")
-```
-
-Then we can share it on the Hub by using the `save_pretrained` method in a `Repository`:
-
-```py
-from huggingface_hub import Repository
-
-repo = Repository("test-dynamic-pipeline", clone_from="{your_username}/test-dynamic-pipeline")
-classifier.save_pretrained("test-dynamic-pipeline")
-repo.push_to_hub()
-```
-
-This will copy the file where you defined `PairClassificationPipeline` inside the folder `"test-dynamic-pipeline"`,
-along with saving the model and tokenizer of the pipeline, before pushing everything in the repository
-`{your_username}/test-dynamic-pipeline`. After that anyone can use it as long as they provide the option
-`trust_remote_code=True`:
-
-```py
-from transformers import pipeline
-
-classifier = pipeline(model="{your_username}/test-dynamic-pipeline", trust_remote_code=True)
-```
-
-## Add the pipeline to Transformers
-
-If you want to contribute your pipeline to Transformers, you will need to add a new module in the `pipelines` submodule
-with the code of your pipeline, then add it in the list of tasks defined in `pipelines/__init__.py`.
-
-Then you will need to add tests. Create a new file `tests/test_pipelines_MY_PIPELINE.py` with example with the other tests.
-
-The `run_pipeline_test` function will be very generic and run on small random models on every possible
-architecture as defined by `model_mapping` and `tf_model_mapping`.
-
-This is very important to test future compatibility, meaning if someone adds a new model for
-`XXXForQuestionAnswering` then the pipeline test will attempt to run on it. Because the models are random it's
-impossible to check for actual values, that's why There is a helper `ANY` that will simply attempt to match the
-output of the pipeline TYPE.
-
-You also *need* to implement 2 (ideally 4) tests.
-
-- `test_small_model_pt` : Define 1 small model for this pipeline (doesn't matter if the results don't make sense)
-  and test the pipeline outputs. The results should be the same as `test_small_model_tf`.
-- `test_small_model_tf` : Define 1 small model for this pipeline (doesn't matter if the results don't make sense)
-  and test the pipeline outputs. The results should be the same as `test_small_model_pt`.
-- `test_large_model_pt` (`optional`): Tests the pipeline on a real pipeline where the results are supposed to
-  make sense. These tests are slow and should be marked as such. Here the goal is to showcase the pipeline and to make
-  sure there is no drift in future releases
-- `test_large_model_tf` (`optional`): Tests the pipeline on a real pipeline where the results are supposed to
-  make sense. These tests are slow and should be marked as such. Here the goal is to showcase the pipeline and to make
-  sure there is no drift in future releases

docs/source/en/add_tensorflow_model.mdx

@@ -1,346 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
--->
-
-# How to convert a 🤗 Transformers model to TensorFlow?
-
-Having multiple frameworks available to use with 🤗 Transformers gives you flexibility to play their strengths when
-designing your application, but it implies that compatibility must be added on a per-model basis. The good news is that
-adding TensorFlow compatibility to an existing model is simpler than [adding a new model from scratch](add_new_model)!
-Whether you wish to have a deeper understanding of large TensorFlow models, make a major open-source contribution, or
-enable TensorFlow for your model of choice, this guide is for you.
-
-This guide empowers you, a member of our community, to contribute TensorFlow model weights and/or
-architectures to be used in 🤗 Transformers, with minimal supervision from the Hugging Face team. Writing a new model
-is no small feat, but hopefully this guide will make it less of a rollercoaster 🎢 and more of a walk in the park 🚶.
-Harnessing our collective experiences is absolutely critical to make this process increasingly easier, and thus we
-highly encourage that you suggest improvements to this guide!
-
-Before you dive deeper, it is recommended that you check the following resources if you're new to 🤗 Transformers:
-- [General overview of 🤗 Transformers](add_new_model#general-overview-of-transformers)
-- [Hugging Face's TensorFlow Philosophy](https://huggingface.co/blog/tensorflow-philosophy)
-
-In the remainder of this guide, you will learn what's needed to add a new TensorFlow model architecture, the
-procedure to convert PyTorch into TensorFlow model weights, and how to efficiently debug mismatches across ML
-frameworks. Let's get started!
-
-<Tip>
-
-Are you unsure whether the model you wish to use already has a corresponding TensorFlow architecture?
-
-&nbsp;
-
-Check the `model_type` field of the `config.json` of your model of choice
-([example](https://huggingface.co/bert-base-uncased/blob/main/config.json#L14)). If the corresponding model folder in
-🤗 Transformers has a file whose name starts with "modeling_tf", it means that it has a corresponding TensorFlow
-architecture ([example](https://github.com/huggingface/transformers/tree/main/src/transformers/models/bert)).
-
-</Tip>
-
-
-## Step-by-step guide to add TensorFlow model architecture code
-
-There are many ways to design a large model architecture, and multiple ways of implementing said design. However,
-you might recall from our [general overview of 🤗 Transformers](add_new_model#general-overview-of-transformers)
-that we are an opinionated bunch - the ease of use of 🤗 Transformers relies on consistent design choices. From
-experience, we can tell you a few important things about adding TensorFlow models:
-
-- Don't reinvent the wheel! More often that not, there are at least two reference implementations you should check: the
-PyTorch equivalent of the model you are implementing and other TensorFlow models for the same class of problems.
-- Great model implementations survive the test of time. This doesn't happen because the code is pretty, but rather
-because the code is clear, easy to debug and build upon. If you make the life of the maintainers easy with your
-TensorFlow implementation, by replicating the same patterns as in other TensorFlow models and minimizing the mismatch
-to the PyTorch implementation, you ensure your contribution will be long lived.
-- Ask for help when you're stuck! The 🤗 Transformers team is here to help, and we've probably found solutions to the same
-problems you're facing.
-
-Here's an overview of the steps needed to add a TensorFlow model architecture:
-1. Select the model you wish to convert
-2. Prepare transformers dev environment
-3. (Optional) Understand theoretical aspects and the existing implementation
-4. Implement the model architecture
-5. Implement model tests
-6. Submit the pull request
-7. (Optional) Build demos and share with the world
-
-### 1.-3. Prepare your model contribution
-
-**1. Select the model you wish to convert**
-
-Let's start off with the basics: the first thing you need to know is the architecture you want to convert. If you
-don't have your eyes set on a specific architecture, asking the 🤗 Transformers team for suggestions is a great way to
-maximize your impact - we will guide you towards the most prominent architectures that are missing on the TensorFlow
-side. If the specific model you want to use with TensorFlow already has a TensorFlow architecture implementation in
-🤗 Transformers but is lacking weights, feel free to jump straight into the
-[weight conversion section](#adding-tensorflow-weights-to-hub)
-of this page.
-
-For simplicity, the remainder of this guide assumes you've decided to contribute with the TensorFlow version of
-*BrandNewBert* (the same example as in the [guide](add_new_model) to add a new model from scratch).
-
-<Tip>
-
-Before starting the work on a TensorFlow model architecture, double-check that there is no ongoing effort to do so.
-You can search for `BrandNewBert` on the
-[pull request GitHub page](https://github.com/huggingface/transformers/pulls?q=is%3Apr) to confirm that there is no
-TensorFlow-related pull request.
-
-</Tip>
-
-
-**2. Prepare transformers dev environment**
-
-Having selected the model architecture, open an draft PR to signal your intention to work on it. Follow the
-instructions below to set up your environment and open a draft PR.
-
-1. Fork the [repository](https://github.com/huggingface/transformers) by clicking on the 'Fork' button on the
-   repository's page. This creates a copy of the code under your GitHub user account.
-
-2. Clone your `transformers` fork to your local disk, and add the base repository as a remote:
-
-```bash
-git clone https://github.com/[your Github handle]/transformers.git
-cd transformers
-git remote add upstream https://github.com/huggingface/transformers.git
-```
-
-3. Set up a development environment, for instance by running the following command:
-
-```bash
-python -m venv .env
-source .env/bin/activate
-pip install -e ".[dev]"
-```
-
-**Note:** You don't need to have CUDA installed. Making the new model work on CPU is sufficient.
-
-4. Create a branch with a descriptive name from your main branch
-
-```bash
-git checkout -b add_tf_brand_new_bert
-```
-
-5. Fetch and rebase to current main
-
-```bash
-git fetch upstream
-git rebase upstream/main
-```
-
-6. Add an empty `.py` file in `transformers/src/models/brandnewbert/` named `modeling_tf_brandnewbert.py`. This will
-be your TensorFlow model file.
-
-7. Push the changes to your account using:
-
-```bash
-git add .
-git commit -m "initial commit"
-git push -u origin add_tf_brand_new_bert
-```
-
-8. Once you are satisfied, go to the webpage of your fork on GitHub. Click on “Pull request”. Make sure to add the
-   GitHub handle of some members of the Hugging Face team as reviewers, so that the Hugging Face team gets notified for
-   future changes.
-
-9. Change the PR into a draft by clicking on “Convert to draft” on the right of the GitHub pull request web page.
-
-
-Now you have set up a development environment to port *BrandNewBert* to TensorFlow in 🤗 Transformers.
-
-
-**3. (Optional) Understand theoretical aspects and the existing implementation**
-
-You should take some time to read *BrandNewBert's* paper, if such descriptive work exists. There might be large
-sections of the paper that are difficult to understand. If this is the case, this is fine - don't worry! The goal is
-not to get a deep theoretical understanding of the paper, but to extract the necessary information required to
-effectively re-implement the model in 🤗 Transformers using TensorFlow. That being said, you don't have to spend too
-much time on the theoretical aspects, but rather focus on the practical ones, namely the existing model documentation
-page (e.g. [model docs for BERT](model_doc/bert)).
-
-After you've grasped the basics of the models you are about to implement, it's important to understand the existing
-implementation. This is a great chance to confirm that a working implementation matches your expectations for the
-model, as well as to foresee technical challenges on the TensorFlow side.
-
-It's perfectly natural that you feel overwhelmed with the amount of information that you've just absorbed. It is
-definitely not a requirement that you understand all facets of the model at this stage. Nevertheless, we highly
-encourage you to clear any pressing questions in our [forum](https://discuss.huggingface.co/).
-
-
-### 4. Model implementation
-
-Now it's time to finally start coding. Our suggested starting point is the PyTorch file itself: copy the contents of
-`modeling_brand_new_bert.py` inside `src/transformers/models/brand_new_bert/` into
-`modeling_tf_brand_new_bert.py`. The goal of this section is to modify the file and update the import structure of
-🤗 Transformers such that you can import `TFBrandNewBert` and
-`TFBrandNewBert.from_pretrained(model_repo, from_pt=True)` sucessfully loads a working TensorFlow *BrandNewBert* model.
-
-Sadly, there is no prescription to convert a PyTorch model into TensorFlow. You can, however, follow our selection of
-tips to make the process as smooth as possible:
-- Prepend `TF` to the name of all classes (e.g. `BrandNewBert` becomes `TFBrandNewBert`).
-- Most PyTorch operations have a direct TensorFlow replacement. For example, `torch.nn.Linear` corresponds to
-  `tf.keras.layers.Dense`, `torch.nn.Dropout` corresponds to `tf.keras.layers.Dropout`, etc. If you're not sure
-  about a specific operation, you can use the [TensorFlow documentation](https://www.tensorflow.org/api_docs/python/tf)
-  or the [PyTorch documentation](https://pytorch.org/docs/stable/).
-- Look for patterns in the 🤗 Transformers codebase. If you come across a certain operation that doesn't have a direct
-   replacement, the odds are that someone else already had the same problem.
-- By default, keep the same variable names and structure as in PyTorch. This will make it easier to debug, track
-   issues, and add fixes down the line.
-- Some layers have different default values in each framework. A notable example is the batch normalization layer's
-   epsilon (`1e-5` in [PyTorch](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html#torch.nn.BatchNorm2d)
-   and `1e-3` in [TensorFlow](https://www.tensorflow.org/api_docs/python/tf/keras/layers/BatchNormalization)).
-   Double-check the documentation!
-- PyTorch's `nn.Parameter` variables typically need to be initialized within TF Layer's `build()`. See the following
-   example: [PyTorch](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_vit_mae.py#L212) /
-   [TensorFlow](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_tf_vit_mae.py#L220)
-- If the PyTorch model has a `#copied from ...` on top of a function, the odds are that your TensorFlow model can also
-   borrow that function from the architecture it was copied from, assuming it has a TensorFlow architecture.
-- Assigning the `name` attribute correctly in TensorFlow functions is critical to do the `from_pt=True` weight
-   cross-loading. `name` is almost always the name of the corresponding variable in the PyTorch code. If `name` is not
-   properly set, you will see it in the error message when loading the model weights.
-- The logic of the base model class, `BrandNewBertModel`, will actually reside in `TFBrandNewBertMainLayer`, a Keras
-   layer subclass ([example](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L719)).
-   `TFBrandNewBertModel` will simply be a wrapper around this layer.
-- Keras models need to be built in order to load pretrained weights. For that reason, `TFBrandNewBertPreTrainedModel`
-   will need to hold an example of inputs to the model, the `dummy_inputs`
-   ([example](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L916)).
-- If you get stuck, ask for help - we're here to help you! 🤗
-
-In addition to the model file itself, you will also need to add the pointers to the model classes and related
-documentation pages. You can complete this part entirely following the patterns in other PRs
-([example](https://github.com/huggingface/transformers/pull/18020/files)). Here's a list of the needed manual
-changes:
-- Include all public classes of *BrandNewBert* in `src/transformers/__init__.py`
-- Add *BrandNewBert* classes to the corresponing Auto classes in `src/transformers/models/auto/modeling_tf_auto.py`
-- Include the modeling file in the documentation test file list in `utils/documentation_tests.txt`
-- Add the lazy loading classes related to *BrandNewBert* in `src/transformers/utils/dummy_tf_objects.py`
-- Update the import structures for the public classes in `src/transformers/models/brand_new_bert/__init__.py`
-- Add the documentation pointers to the public methods of *BrandNewBert* in `docs/source/en/model_doc/brand_new_bert.mdx`
-- Add yourself to the list of contributors to *BrandNewBert* in `docs/source/en/model_doc/brand_new_bert.mdx`
-- Finally, add a green tick ✅ to the TensorFlow column of *BrandNewBert* in `docs/source/en/index.mdx`
-
-When you're happy with your implementation, run the following checklist to confirm that your model architecture is
-ready:
-1. All layers that behave differently at train time (e.g. Dropout) are called with a `training` argument, which is
-propagated all the way from the top-level classes
-2. You have used `#copied from ...` whenever possible
-3. `TFBrandNewBertMainLayer` and all classes that use it have their `call` function decorated with `@unpack_inputs`
-4. `TFBrandNewBertMainLayer` is decorated with `@keras_serializable`
-5. A TensorFlow model can be loaded from PyTorch weights using `TFBrandNewBert.from_pretrained(model_repo, from_pt=True)`
-6. You can call the TensorFlow model using the expected input format
-
-
-### 5. Add model tests
-
-Hurray, you've implemented a TensorFlow model! Now it's time to add tests to make sure that your model behaves as
-expected. As in the previous section, we suggest you start by copying the `test_modeling_brand_new_bert.py` file in
-`tests/models/brand_new_bert/` into `test_modeling_tf_brand_new_bert.py`, and continue by making the necessary
-TensorFlow replacements. For now, in all `.from_pretrained()` calls, you should use the `from_pt=True` flag to load
-the existing PyTorch weights.
-
-After you're done, it's time for the moment of truth: run the tests! 😬
-
-```bash
-NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
-py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
-```
-
-The most likely outcome is that you'll see a bunch of errors. Don't worry, this is expected! Debugging ML models is
-notoriously hard, and the key ingredient to success is patience (and `breakpoint()`). In our experience, the hardest
-problems arise from subtle mismatches between ML frameworks, for which we have a few pointers at the end of this guide.
-In other cases, a general test might not be directly applicable to your model, in which case we suggest an override
-at the model test class level. Regardless of the issue, don't hesitate to ask for help in your draft pull request if
-you're stuck.
-
-When all tests pass, congratulations, your model is nearly ready to be added to the 🤗 Transformers library! 🎉
-
-### 6.-7. Ensure everyone can use your model
-
-**6. Submit the pull request**
-
-Once you're done with the implementation and the tests, it's time to submit a pull request. Before pushing your code,
-run our code formatting utility, `make fixup` 🪄. This will automatically fix any formatting issues, which would cause
-our automatic checks to fail.
-
-It's now time to convert your draft pull request into a real pull request. To do so, click on the "Ready for
-review" button and add Joao (`@gante`) and Matt (`@Rocketknight1`) as reviewers. A model pull request will need
-at least 3 reviewers, but they will take care of finding appropriate additional reviewers for your model.
-
-After all reviewers are happy with the state of your PR, the final action point is to remove the `from_pt=True` flag in
-`.from_pretrained()` calls. Since there are no TensorFlow weights, you will have to add them! Check the section
-below for instructions on how to do it.
-
-Finally, when the TensorFlow weights get merged, you have at least 3 reviewer approvals, and all CI checks are
-green, double-check the tests locally one last time
-
-```bash
-NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
-py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
-```
-
-and we will merge your PR! Congratulations on the milestone 🎉
-
-**7. (Optional) Build demos and share with the world**
-
-One of the hardest parts about open-source is discovery. How can the other users learn about the existence of your
-fabulous TensorFlow contribution? With proper communication, of course! 📣
-
-There are two main ways to share your model with the community:
-- Build demos. These include Gradio demos, notebooks, and other fun ways to show off your model. We highly
-   encourage you to add a notebook to our [community-driven demos](https://huggingface.co/docs/transformers/community).
-- Share stories on social media like Twitter and LinkedIn. You should be proud of your work and share
-   your achievement with the community - your model can now be used by thousands of engineers and researchers around
-   the world 🌍! We will be happy to retweet your posts and help you share your work with the community.
-
-
-## Adding TensorFlow weights to 🤗 Hub
-
-Assuming that the TensorFlow model architecture is available in 🤗 Transformers, converting PyTorch weights into
-TensorFlow weights is a breeze!
-
-Here's how to do it:
-1. Make sure you are logged into your Hugging Face account in your terminal. You can log in using the command
-   `huggingface-cli login` (you can find your access tokens [here](https://huggingface.co/settings/tokens))
-2. Run `transformers-cli pt-to-tf --model-name foo/bar`, where `foo/bar` is the name of the model repository
-   containing the PyTorch weights you want to convert
-3. Tag `@joaogante` and `@Rocketknight1` in the 🤗 Hub PR the command above has just created
-
-That's it! 🎉
-
-
-## Debugging mismatches across ML frameworks 🐛
-
-At some point, when adding a new architecture or when creating TensorFlow weights for an existing architecture, you
-might come across errors compaining about mismatches between PyTorch and TensorFlow. You might even decide to open the
-model architecture code for the two frameworks, and find that they look identical. What's going on? 🤔
-
-First of all, let's talk about why understanding these mismatches matters. Many community members will use 🤗
-Transformers models out of the box, and trust that our models behave as expected. When there is a large mismatch
-between the two frameworks, it implies that the model is not following the reference implementation for at least one
-of the frameworks. This might lead to silent failures, in which the model runs but has poor performance. This is
-arguably worse than a model that fails to run at all! To that end, we aim at having a framework mismatch smaller than
-`1e-5` at all stages of the model.
-
-As in other numerical problems, the devil is in the details. And as in any detail-oriented craft, the secret
-ingredient here is patience. Here is our suggested workflow for when you come across this type of issues:
-1. Locate the source of mismatches. The model you're converting probably has near identical inner variables up to a
-   certain point. Place `breakpoint()` statements in the two frameworks' architectures, and compare the values of the
-   numerical variables in a top-down fashion until you find the source of the problems.
-2. Now that you've pinpointed the source of the issue, get in touch with the 🤗 Transformers team. It is possible
-   that we've seen a similar problem before and can promptly provide a solution. As a fallback, scan popular pages
-   like StackOverflow and GitHub issues.
-3. If there is no solution in sight, it means you'll have to go deeper. The good news is that you've located the
-   issue, so you can focus on the problematic instruction, abstracting away the rest of the model! The bad news is
-   that you'll have to venture into the source implementation of said instruction. In some cases, you might find an
-   issue with a reference implementation - don't abstain from opening an issue in the upstream repository.
-
-In some cases, in dicussion with the 🤗 Transformers team, we might find that the fixing the mismatch is infeasible.
-When the mismatch is very small in the output layers of the model (but potentially large in the hidden states), we
-might decide to ignore it in favor of distributing the model. The `pt-to-tf` CLI mentioned above has a `--max-error`
-flag to override the error message at weight conversion time.

docs/source/en/autoclass_tutorial.mdx

@@ -1,127 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Load pretrained instances with an AutoClass
-
-With so many different Transformer architectures, it can be challenging to create one for your checkpoint. As a part of 🤗 Transformers core philosophy to make the library easy, simple and flexible to use, an `AutoClass` automatically infer and load the correct architecture from a given checkpoint. The `from_pretrained()` method lets you quickly load a pretrained model for any architecture so you don't have to devote time and resources to train a model from scratch. Producing this type of checkpoint-agnostic code means if your code works for one checkpoint, it will work with another checkpoint - as long as it was trained for a similar task - even if the architecture is different.
-
-<Tip>
-
-Remember, architecture refers to the skeleton of the model and checkpoints are the weights for a given architecture. For example, [BERT](https://huggingface.co/bert-base-uncased) is an architecture, while `bert-base-uncased` is a checkpoint. Model is a general term that can mean either architecture or checkpoint.
-
-</Tip>
-
-In this tutorial, learn to:
-
-* Load a pretrained tokenizer.
-* Load a pretrained feature extractor.
-* Load a pretrained processor.
-* Load a pretrained model.
-
-## AutoTokenizer
-
-Nearly every NLP task begins with a tokenizer. A tokenizer converts your input into a format that can be processed by the model.
-
-Load a tokenizer with [`AutoTokenizer.from_pretrained`]:
-
-```py
->>> from transformers import AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
-```
-
-Then tokenize your input as shown below:
-
-```py
->>> sequence = "In a hole in the ground there lived a hobbit."
->>> print(tokenizer(sequence))
-{'input_ids': [101, 1999, 1037, 4920, 1999, 1996, 2598, 2045, 2973, 1037, 7570, 10322, 4183, 1012, 102], 
- 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
- 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
-```
-
-## AutoFeatureExtractor
-
-For audio and vision tasks, a feature extractor processes the audio signal or image into the correct input format.
-
-Load a feature extractor with [`AutoFeatureExtractor.from_pretrained`]:
-
-```py
->>> from transformers import AutoFeatureExtractor
-
->>> feature_extractor = AutoFeatureExtractor.from_pretrained(
-...     "ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition"
-... )
-```
-
-## AutoProcessor
-
-Multimodal tasks require a processor that combines two types of preprocessing tools. For example, the [LayoutLMV2](model_doc/layoutlmv2) model requires a feature extractor to handle images and a tokenizer to handle text; a processor combines both of them.
-
-Load a processor with [`AutoProcessor.from_pretrained`]:
-
-```py
->>> from transformers import AutoProcessor
-
->>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased")
-```
-
-## AutoModel
-
-<frameworkcontent>
-<pt>
-Finally, the `AutoModelFor` classes let you load a pretrained model for a given task (see [here](model_doc/auto) for a complete list of available tasks). For example, load a model for sequence classification with [`AutoModelForSequenceClassification.from_pretrained`]:
-
-```py
->>> from transformers import AutoModelForSequenceClassification
-
->>> model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-Easily reuse the same checkpoint to load an architecture for a different task:
-
-```py
->>> from transformers import AutoModelForTokenClassification
-
->>> model = AutoModelForTokenClassification.from_pretrained("distilbert-base-uncased")
-```
-
-<Tip warning={true}>
-
-For PyTorch models, the `from_pretrained()` method uses `torch.load()` which internally uses `pickle` and is known to be insecure. In general, never load a model that could have come from an untrusted source, or that could have been tampered with. This security risk is partially mitigated for public models hosted on the Hugging Face Hub, which are [scanned for malware](https://huggingface.co/docs/hub/security-malware) at each commit. See the [Hub documentation](https://huggingface.co/docs/hub/security) for best practices like [signed commit verification](https://huggingface.co/docs/hub/security-gpg#signing-commits-with-gpg) with GPG.
-
-TensorFlow and Flax checkpoints are not affected, and can be loaded within PyTorch architectures using the `from_tf` and `from_flax` kwargs for the `from_pretrained` method to circumvent this issue.
-
-</Tip>
-
-Generally, we recommend using the `AutoTokenizer` class and the `AutoModelFor` class to load pretrained instances of models. This will ensure you load the correct architecture every time. In the next [tutorial](preprocessing), learn how to use your newly loaded tokenizer, feature extractor and processor to preprocess a dataset for fine-tuning.
-</pt>
-<tf>
-Finally, the `TFAutoModelFor` classes let you load a pretrained model for a given task (see [here](model_doc/auto) for a complete list of available tasks). For example, load a model for sequence classification with [`TFAutoModelForSequenceClassification.from_pretrained`]:
-
-```py
->>> from transformers import TFAutoModelForSequenceClassification
-
->>> model = TFAutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-Easily reuse the same checkpoint to load an architecture for a different task:
-
-```py
->>> from transformers import TFAutoModelForTokenClassification
-
->>> model = TFAutoModelForTokenClassification.from_pretrained("distilbert-base-uncased")
-```
-
-Generally, we recommend using the `AutoTokenizer` class and the `TFAutoModelFor` class to load pretrained instances of models. This will ensure you load the correct architecture every time. In the next [tutorial](preprocessing), learn how to use your newly loaded tokenizer, feature extractor and processor to preprocess a dataset for fine-tuning.
-</tf>
-</frameworkcontent>

docs/source/en/benchmarks.mdx

@@ -1,383 +0,0 @@
-<!--Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Benchmarks
-
-<Tip warning={true}>
-
-Hugging Face's Benchmarking tools are deprecated and it is advised to use external Benchmarking libraries to measure the speed 
-and memory complexity of Transformer models.
-
-</Tip>
-
-[[open-in-colab]]
-
-Let's take a look at how 🤗 Transformers models can be benchmarked, best practices, and already available benchmarks.
-
-A notebook explaining in more detail how to benchmark 🤗 Transformers models can be found [here](https://github.com/huggingface/notebooks/tree/main/examples/benchmark.ipynb).
-
-## How to benchmark 🤗 Transformers models
-
-The classes [`PyTorchBenchmark`] and [`TensorFlowBenchmark`] allow to flexibly benchmark 🤗 Transformers models. The benchmark classes allow us to measure the _peak memory usage_ and _required time_ for both _inference_ and _training_.
-
-<Tip>
-
-Hereby, _inference_ is defined by a single forward pass, and _training_ is defined by a single forward pass and
-backward pass.
-
-</Tip>
-
-The benchmark classes [`PyTorchBenchmark`] and [`TensorFlowBenchmark`] expect an object of type [`PyTorchBenchmarkArguments`] and
-[`TensorFlowBenchmarkArguments`], respectively, for instantiation. [`PyTorchBenchmarkArguments`] and [`TensorFlowBenchmarkArguments`] are data classes and contain all relevant configurations for their corresponding benchmark class. In the following example, it is shown how a BERT model of type _bert-base-cased_ can be benchmarked.
-
-<frameworkcontent>
-<pt>
-```py
->>> from transformers import PyTorchBenchmark, PyTorchBenchmarkArguments
-
->>> args = PyTorchBenchmarkArguments(models=["bert-base-uncased"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512])
->>> benchmark = PyTorchBenchmark(args)
-```
-</pt>
-<tf>
-```py
->>> from transformers import TensorFlowBenchmark, TensorFlowBenchmarkArguments
-
->>> args = TensorFlowBenchmarkArguments(
-...     models=["bert-base-uncased"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512]
-... )
->>> benchmark = TensorFlowBenchmark(args)
-```
-</tf>
-</frameworkcontent>
-
-Here, three arguments are given to the benchmark argument data classes, namely `models`, `batch_sizes`, and
-`sequence_lengths`. The argument `models` is required and expects a `list` of model identifiers from the
-[model hub](https://huggingface.co/models) The `list` arguments `batch_sizes` and `sequence_lengths` define
-the size of the `input_ids` on which the model is benchmarked. There are many more parameters that can be configured
-via the benchmark argument data classes. For more detail on these one can either directly consult the files
-`src/transformers/benchmark/benchmark_args_utils.py`, `src/transformers/benchmark/benchmark_args.py` (for PyTorch)
-and `src/transformers/benchmark/benchmark_args_tf.py` (for Tensorflow). Alternatively, running the following shell
-commands from root will print out a descriptive list of all configurable parameters for PyTorch and Tensorflow
-respectively.
-
-<frameworkcontent>
-<pt>
-```bash
-python examples/pytorch/benchmarking/run_benchmark.py --help
-```
-
-An instantiated benchmark object can then simply be run by calling `benchmark.run()`.
-
-```py
->>> results = benchmark.run()
->>> print(results)
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length     Time in s                  
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             0.006     
-bert-base-uncased          8               32            0.006     
-bert-base-uncased          8              128            0.018     
-bert-base-uncased          8              512            0.088     
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length    Memory in MB 
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             1227
-bert-base-uncased          8               32            1281
-bert-base-uncased          8              128            1307
-bert-base-uncased          8              512            1539
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: PyTorch
-- use_torchscript: False
-- framework_version: 1.4.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 08:58:43.371351
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</pt>
-<tf>
-```bash
-python examples/tensorflow/benchmarking/run_benchmark_tf.py --help
-```
-
-An instantiated benchmark object can then simply be run by calling `benchmark.run()`.
-
-```py
->>> results = benchmark.run()
->>> print(results)
->>> results = benchmark.run()
->>> print(results)
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length     Time in s                  
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             0.005
-bert-base-uncased          8               32            0.008
-bert-base-uncased          8              128            0.022
-bert-base-uncased          8              512            0.105
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length    Memory in MB 
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             1330
-bert-base-uncased          8               32            1330
-bert-base-uncased          8              128            1330
-bert-base-uncased          8              512            1770
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: Tensorflow
-- use_xla: False
-- framework_version: 2.2.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 09:26:35.617317
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</tf>
-</frameworkcontent>
-
-By default, the _time_ and the _required memory_ for _inference_ are benchmarked. In the example output above the first
-two sections show the result corresponding to _inference time_ and _inference memory_. In addition, all relevant
-information about the computing environment, _e.g._ the GPU type, the system, the library versions, etc... are printed
-out in the third section under _ENVIRONMENT INFORMATION_. This information can optionally be saved in a _.csv_ file
-when adding the argument `save_to_csv=True` to [`PyTorchBenchmarkArguments`] and
-[`TensorFlowBenchmarkArguments`] respectively. In this case, every section is saved in a separate
-_.csv_ file. The path to each _.csv_ file can optionally be defined via the argument data classes.
-
-Instead of benchmarking pre-trained models via their model identifier, _e.g._ `bert-base-uncased`, the user can
-alternatively benchmark an arbitrary configuration of any available model class. In this case, a `list` of
-configurations must be inserted with the benchmark args as follows.
-
-<frameworkcontent>
-<pt>
-```py
->>> from transformers import PyTorchBenchmark, PyTorchBenchmarkArguments, BertConfig
-
->>> args = PyTorchBenchmarkArguments(
-...     models=["bert-base", "bert-384-hid", "bert-6-lay"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512]
-... )
->>> config_base = BertConfig()
->>> config_384_hid = BertConfig(hidden_size=384)
->>> config_6_lay = BertConfig(num_hidden_layers=6)
-
->>> benchmark = PyTorchBenchmark(args, configs=[config_base, config_384_hid, config_6_lay])
->>> benchmark.run()
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length       Time in s                  
---------------------------------------------------------------------------------
-bert-base                  8              128            0.006
-bert-base                  8              512            0.006
-bert-base                  8              128            0.018     
-bert-base                  8              512            0.088     
-bert-384-hid              8               8             0.006     
-bert-384-hid              8               32            0.006     
-bert-384-hid              8              128            0.011     
-bert-384-hid              8              512            0.054     
-bert-6-lay                 8               8             0.003     
-bert-6-lay                 8               32            0.004     
-bert-6-lay                 8              128            0.009     
-bert-6-lay                 8              512            0.044
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length      Memory in MB 
---------------------------------------------------------------------------------
-bert-base                  8               8             1277
-bert-base                  8               32            1281
-bert-base                  8              128            1307     
-bert-base                  8              512            1539     
-bert-384-hid              8               8             1005     
-bert-384-hid              8               32            1027     
-bert-384-hid              8              128            1035     
-bert-384-hid              8              512            1255     
-bert-6-lay                 8               8             1097     
-bert-6-lay                 8               32            1101     
-bert-6-lay                 8              128            1127     
-bert-6-lay                 8              512            1359
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: PyTorch
-- use_torchscript: False
-- framework_version: 1.4.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 09:35:25.143267
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</pt>
-<tf>
-```py
->>> from transformers import TensorFlowBenchmark, TensorFlowBenchmarkArguments, BertConfig
-
->>> args = TensorFlowBenchmarkArguments(
-...     models=["bert-base", "bert-384-hid", "bert-6-lay"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512]
-... )
->>> config_base = BertConfig()
->>> config_384_hid = BertConfig(hidden_size=384)
->>> config_6_lay = BertConfig(num_hidden_layers=6)
-
->>> benchmark = TensorFlowBenchmark(args, configs=[config_base, config_384_hid, config_6_lay])
->>> benchmark.run()
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length       Time in s                  
---------------------------------------------------------------------------------
-bert-base                  8               8             0.005
-bert-base                  8               32            0.008
-bert-base                  8              128            0.022
-bert-base                  8              512            0.106
-bert-384-hid              8               8             0.005
-bert-384-hid              8               32            0.007
-bert-384-hid              8              128            0.018
-bert-384-hid              8              512            0.064
-bert-6-lay                 8               8             0.002
-bert-6-lay                 8               32            0.003
-bert-6-lay                 8              128            0.0011
-bert-6-lay                 8              512            0.074
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length      Memory in MB 
---------------------------------------------------------------------------------
-bert-base                  8               8             1330
-bert-base                  8               32            1330
-bert-base                  8              128            1330
-bert-base                  8              512            1770
-bert-384-hid              8               8             1330
-bert-384-hid              8               32            1330
-bert-384-hid              8              128            1330
-bert-384-hid              8              512            1540
-bert-6-lay                 8               8             1330
-bert-6-lay                 8               32            1330
-bert-6-lay                 8              128            1330
-bert-6-lay                 8              512            1540
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: Tensorflow
-- use_xla: False
-- framework_version: 2.2.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 09:38:15.487125
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</tf>
-</frameworkcontent>
-
-Again, _inference time_ and _required memory_ for _inference_ are measured, but this time for customized configurations
-of the `BertModel` class. This feature can especially be helpful when deciding for which configuration the model
-should be trained.
-
-
-## Benchmark best practices
-
-This section lists a couple of best practices one should be aware of when benchmarking a model.
-
-- Currently, only single device benchmarking is supported. When benchmarking on GPU, it is recommended that the user
-  specifies on which device the code should be run by setting the `CUDA_VISIBLE_DEVICES` environment variable in the
-  shell, _e.g._ `export CUDA_VISIBLE_DEVICES=0` before running the code.
-- The option `no_multi_processing` should only be set to `True` for testing and debugging. To ensure accurate
-  memory measurement it is recommended to run each memory benchmark in a separate process by making sure
-  `no_multi_processing` is set to `True`.
-- One should always state the environment information when sharing the results of a model benchmark. Results can vary
-  heavily between different GPU devices, library versions, etc., so that benchmark results on their own are not very
-  useful for the community.
-
-
-## Sharing your benchmark
-
-Previously all available core models (10 at the time) have been benchmarked for _inference time_, across many different
-settings: using PyTorch, with and without TorchScript, using TensorFlow, with and without XLA. All of those tests were
-done across CPUs (except for TensorFlow XLA) and GPUs.
-
-The approach is detailed in the [following blogpost](https://medium.com/huggingface/benchmarking-transformers-pytorch-and-tensorflow-e2917fb891c2) and the results are
-available [here](https://docs.google.com/spreadsheets/d/1sryqufw2D0XlUH4sq3e9Wnxu5EAQkaohzrJbd5HdQ_w/edit?usp=sharing).
-
-With the new _benchmark_ tools, it is easier than ever to share your benchmark results with the community
-
-- [PyTorch Benchmarking Results](https://github.com/huggingface/transformers/tree/main/examples/pytorch/benchmarking/README.md).
-- [TensorFlow Benchmarking Results](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/benchmarking/README.md).

docs/source/en/bertology.mdx

@@ -1,36 +0,0 @@
-<!--Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# BERTology
-
-There is a growing field of study concerned with investigating the inner working of large-scale transformers like BERT
-(that some call "BERTology"). Some good examples of this field are:
-
-
-- BERT Rediscovers the Classical NLP Pipeline by Ian Tenney, Dipanjan Das, Ellie Pavlick:
-  https://arxiv.org/abs/1905.05950
-- Are Sixteen Heads Really Better than One? by Paul Michel, Omer Levy, Graham Neubig: https://arxiv.org/abs/1905.10650
-- What Does BERT Look At? An Analysis of BERT's Attention by Kevin Clark, Urvashi Khandelwal, Omer Levy, Christopher D.
-  Manning: https://arxiv.org/abs/1906.04341
-
-In order to help this new field develop, we have included a few additional features in the BERT/GPT/GPT-2 models to
-help people access the inner representations, mainly adapted from the great work of Paul Michel
-(https://arxiv.org/abs/1905.10650):
-
-
-- accessing all the hidden-states of BERT/GPT/GPT-2,
-- accessing all the attention weights for each head of BERT/GPT/GPT-2,
-- retrieving heads output values and gradients to be able to compute head importance score and prune head as explained
-  in https://arxiv.org/abs/1905.10650.
-
-To help you understand and use these features, we have added a specific example script: [bertology.py](https://github.com/huggingface/transformers/tree/main/examples/research_projects/bertology/run_bertology.py) while extract information and prune a model pre-trained on
-GLUE.

docs/source/en/big_models.mdx

@@ -1,119 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Instantiating a big model
-
-When you want to use a very big pretrained model, one challenge is to minimize the use of the RAM. The usual workflow
-from PyTorch is:
-
-1. Create your model with random weights.
-2. Load your pretrained weights.
-3. Put those pretrained weights in your random model.
-
-Step 1 and 2 both require a full version of the model in memory, which is not a problem in most cases, but if your model starts weighing several GigaBytes, those two copies can make you got our of RAM. Even worse, if you are using `torch.distributed` to launch a distributed training, each process will load the pretrained model and store these two copies in RAM.
-
-<Tip>
-
-Note that the randomly created model is initialized with "empty" tensors, which take the space in memory without filling it (thus the random values are whatever was in this chunk of memory at a given time). The random initialization following the appropriate distribution for the kind of model/parameters instatiated (like a normal distribution for instance) is only performed after step 3 on the non-initialized weights, to be as fast as possible! 
-
-</Tip>
-
-In this guide, we explore the solutions Transformers offer to deal with this issue. Note that this is an area of active development, so the APIs explained here may change slightly in the future.
-
-## Sharded checkpoints
-
-Since version 4.18.0, model checkpoints that end up taking more than 10GB of space are automatically sharded in smaller pieces. In terms of having one single checkpoint when you do `model.save_pretrained(save_dir)`, you will end up with several partial checkpoints (each of which being of size < 10GB) and an index that maps parameter names to the files they are stored in.
-
-You can control the maximum size before sharding with the `max_shard_size` parameter, so for the sake of an example, we'll use a normal-size models with a small shard size: let's take a traditional BERT model.
-
-```py
-from transformers import AutoModel
-
-model = AutoModel.from_pretrained("bert-base-cased")
-```
-
-If you save it using [`~PreTrainedModel.save_pretrained`], you will get a new folder with two files: the config of the model and its weights:
-
-```py
->>> import os
->>> import tempfile
-
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir)
-...     print(sorted(os.listdir(tmp_dir)))
-['config.json', 'pytorch_model.bin']
-```
-
-Now let's use a maximum shard size of 200MB:
-
-```py
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     print(sorted(os.listdir(tmp_dir)))
-['config.json', 'pytorch_model-00001-of-00003.bin', 'pytorch_model-00002-of-00003.bin', 'pytorch_model-00003-of-00003.bin', 'pytorch_model.bin.index.json']
-```
-
-On top of the configuration of the model, we see three different weights files, and an `index.json` file which is our index. A checkpoint like this can be fully reloaded using the [`~PreTrainedModel.from_pretrained`] method:
-
-```py
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     new_model = AutoModel.from_pretrained(tmp_dir)
-```
-
-The main advantage of doing this for big models is that during step 2 of the workflow shown above, each shard of the checkpoint is loaded after the previous one, capping the memory usage in RAM to the model size plus the size of the biggest shard.
-
-Beind the scenes, the index file is used to determine which keys are in the checkpoint, and where the corresponding weights are stored. We can load that index like any json and get a dictionary:
-
-```py
->>> import json
-
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     with open(os.path.join(tmp_dir, "pytorch_model.bin.index.json"), "r") as f:
-...         index = json.load(f)
-
->>> print(index.keys())
-dict_keys(['metadata', 'weight_map'])
-```
-
-The metadata just consists of the total size of the model for now. We plan to add several other informations in the future:
-
-```py
->>> index["metadata"]
-{'total_size': 433245184}
-```
-
-The weights map is the main part of this index, which maps each parameter name (as usually found in a PyTorch model `state_dict`) to the file it's stored in:
-
-```py
->>> index["weight_map"]
-{'embeddings.LayerNorm.bias': 'pytorch_model-00001-of-00003.bin',
- 'embeddings.LayerNorm.weight': 'pytorch_model-00001-of-00003.bin',
- ...
-```
-
-If you want to directly load such a sharded checkpoint inside a model without using [`~PreTrainedModel.from_pretrained`] (like you would do `model.load_state_dict()` for a full checkpoint) you should use [`~modeling_utils.load_sharded_checkpoint`]:
-
-```py
->>> from transformers.modeling_utils import load_sharded_checkpoint
-
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     load_sharded_checkpoint(model, tmp_dir)
-```
-
-## Low memory loading
-
-Sharded checkpoints reduce the memory usage during step 2 of the workflow mentioned above, but in order to use that model in a low memory setting, we recommend leveraging our tools based on the Accelerate library.
-
-Please read the following guide for more information: [Large model loading using Accelerate](./main_classes/model#large-model-loading)

docs/source/en/contributing.md

@@ -1 +0,0 @@
-../../../CONTRIBUTING.md

docs/source/en/converting_tensorflow_models.mdx

@@ -1,162 +0,0 @@
-<!--Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Converting From Tensorflow Checkpoints
-
-A command-line interface is provided to convert original Bert/GPT/GPT-2/Transformer-XL/XLNet/XLM checkpoints to models
-that can be loaded using the `from_pretrained` methods of the library.
-
-<Tip>
-
-Since 2.3.0 the conversion script is now part of the transformers CLI (**transformers-cli**) available in any
-transformers >= 2.3.0 installation.
-
-The documentation below reflects the **transformers-cli convert** command format.
-
-</Tip>
-
-## BERT
-
-You can convert any TensorFlow checkpoint for BERT (in particular [the pre-trained models released by Google](https://github.com/google-research/bert#pre-trained-models)) in a PyTorch save file by using the
-[convert_bert_original_tf_checkpoint_to_pytorch.py](https://github.com/huggingface/transformers/tree/main/src/transformers/models/bert/convert_bert_original_tf_checkpoint_to_pytorch.py) script.
-
-This CLI takes as input a TensorFlow checkpoint (three files starting with `bert_model.ckpt`) and the associated
-configuration file (`bert_config.json`), and creates a PyTorch model for this configuration, loads the weights from
-the TensorFlow checkpoint in the PyTorch model and saves the resulting model in a standard PyTorch save file that can
-be imported using `from_pretrained()` (see example in [quicktour](quicktour) , [run_glue.py](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification/run_glue.py) ).
-
-You only need to run this conversion script **once** to get a PyTorch model. You can then disregard the TensorFlow
-checkpoint (the three files starting with `bert_model.ckpt`) but be sure to keep the configuration file (\
-`bert_config.json`) and the vocabulary file (`vocab.txt`) as these are needed for the PyTorch model too.
-
-To run this specific conversion script you will need to have TensorFlow and PyTorch installed (`pip install tensorflow`). The rest of the repository only requires PyTorch.
-
-Here is an example of the conversion process for a pre-trained `BERT-Base Uncased` model:
-
-```bash
-export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12
-
-transformers-cli convert --model_type bert \
-  --tf_checkpoint $BERT_BASE_DIR/bert_model.ckpt \
-  --config $BERT_BASE_DIR/bert_config.json \
-  --pytorch_dump_output $BERT_BASE_DIR/pytorch_model.bin
-```
-
-You can download Google's pre-trained models for the conversion [here](https://github.com/google-research/bert#pre-trained-models).
-
-## ALBERT
-
-Convert TensorFlow model checkpoints of ALBERT to PyTorch using the
-[convert_albert_original_tf_checkpoint_to_pytorch.py](https://github.com/huggingface/transformers/tree/main/src/transformers/models/albert/convert_albert_original_tf_checkpoint_to_pytorch.py) script.
-
-The CLI takes as input a TensorFlow checkpoint (three files starting with `model.ckpt-best`) and the accompanying
-configuration file (`albert_config.json`), then creates and saves a PyTorch model. To run this conversion you will
-need to have TensorFlow and PyTorch installed.
-
-Here is an example of the conversion process for the pre-trained `ALBERT Base` model:
-
-```bash
-export ALBERT_BASE_DIR=/path/to/albert/albert_base
-
-transformers-cli convert --model_type albert \
-  --tf_checkpoint $ALBERT_BASE_DIR/model.ckpt-best \
-  --config $ALBERT_BASE_DIR/albert_config.json \
-  --pytorch_dump_output $ALBERT_BASE_DIR/pytorch_model.bin
-```
-
-You can download Google's pre-trained models for the conversion [here](https://github.com/google-research/albert#pre-trained-models).
-
-## OpenAI GPT
-
-Here is an example of the conversion process for a pre-trained OpenAI GPT model, assuming that your NumPy checkpoint
-save as the same format than OpenAI pretrained model (see [here](https://github.com/openai/finetune-transformer-lm)\
-)
-
-```bash
-export OPENAI_GPT_CHECKPOINT_FOLDER_PATH=/path/to/openai/pretrained/numpy/weights
-
-transformers-cli convert --model_type gpt \
-  --tf_checkpoint $OPENAI_GPT_CHECKPOINT_FOLDER_PATH \
-  --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-  [--config OPENAI_GPT_CONFIG] \
-  [--finetuning_task_name OPENAI_GPT_FINETUNED_TASK] \
-```
-
-## OpenAI GPT-2
-
-Here is an example of the conversion process for a pre-trained OpenAI GPT-2 model (see [here](https://github.com/openai/gpt-2))
-
-```bash
-export OPENAI_GPT2_CHECKPOINT_PATH=/path/to/gpt2/pretrained/weights
-
-transformers-cli convert --model_type gpt2 \
-  --tf_checkpoint $OPENAI_GPT2_CHECKPOINT_PATH \
-  --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-  [--config OPENAI_GPT2_CONFIG] \
-  [--finetuning_task_name OPENAI_GPT2_FINETUNED_TASK]
-```
-
-## Transformer-XL
-
-Here is an example of the conversion process for a pre-trained Transformer-XL model (see [here](https://github.com/kimiyoung/transformer-xl/tree/master/tf#obtain-and-evaluate-pretrained-sota-models))
-
-```bash
-export TRANSFO_XL_CHECKPOINT_FOLDER_PATH=/path/to/transfo/xl/checkpoint
-
-transformers-cli convert --model_type transfo_xl \
-  --tf_checkpoint $TRANSFO_XL_CHECKPOINT_FOLDER_PATH \
-  --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-  [--config TRANSFO_XL_CONFIG] \
-  [--finetuning_task_name TRANSFO_XL_FINETUNED_TASK]
-```
-
-## XLNet
-
-Here is an example of the conversion process for a pre-trained XLNet model:
-
-```bash
-export TRANSFO_XL_CHECKPOINT_PATH=/path/to/xlnet/checkpoint
-export TRANSFO_XL_CONFIG_PATH=/path/to/xlnet/config
-
-transformers-cli convert --model_type xlnet \
-  --tf_checkpoint $TRANSFO_XL_CHECKPOINT_PATH \
-  --config $TRANSFO_XL_CONFIG_PATH \
-  --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-  [--finetuning_task_name XLNET_FINETUNED_TASK] \
-```
-
-## XLM
-
-Here is an example of the conversion process for a pre-trained XLM model:
-
-```bash
-export XLM_CHECKPOINT_PATH=/path/to/xlm/checkpoint
-
-transformers-cli convert --model_type xlm \
-  --tf_checkpoint $XLM_CHECKPOINT_PATH \
-  --pytorch_dump_output $PYTORCH_DUMP_OUTPUT
- [--config XML_CONFIG] \
- [--finetuning_task_name XML_FINETUNED_TASK]
-```
-
-## T5
-
-Here is an example of the conversion process for a pre-trained T5 model:
-
-```bash
-export T5=/path/to/t5/uncased_L-12_H-768_A-12
-
-transformers-cli convert --model_type t5 \
-  --tf_checkpoint $T5/t5_model.ckpt \
-  --config $T5/t5_config.json \
-  --pytorch_dump_output $T5/pytorch_model.bin
-```