Release: v4.15.0

* Revert "Revert "Added support for other features for already supported models (#14358)" (#14679)"

This reverts commit 0f4e39c5599523c110cd713f60a3bfa145dad807.

* is_torch_available test to avoid failing imports

* sorting parameterize parameters to solve ERROR gw0 gw1

* tests fix

* tests fix

* GPT2 with past fix

* Fixed stateful class attribute change that was breaking things when converting multiple models sequentially

* Removed onnx file

* Implemented suggestions

* Fixed __init__ to resolve conflict with master

* Remove commented import

.circleci/TROUBLESHOOT.md

@@ -0,0 +1,7 @@
+# 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

@@ -65,7 +65,7 @@ jobs:
     run_tests_torch_and_tf:
         working_directory: ~/transformers
         docker:
-            - image: circleci/python:3.6
+            - image: circleci/python:3.7
         environment:
             OMP_NUM_THREADS: 1
             RUN_PT_TF_CROSS_TESTS: yes
@@ -78,15 +78,56 @@ jobs:
                   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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - run: pip install --upgrade pip
-            - 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
+            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install tensorflow_probability
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf $(cat test_list.txt) -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_torch_and_tf_all:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.7
+        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 espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install tensorflow_probability
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
+            - 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_output.txt
             - store_artifacts:
@@ -108,15 +149,54 @@ jobs:
                   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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - run: pip install --upgrade pip
-            - 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
+            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_flax $(cat test_list.txt) -m is_pt_flax_cross_test --durations=0 | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_torch_and_flax_all:
+        working_directory: ~/transformers
+        docker:
+            - 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 espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
+            - 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: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -137,15 +217,53 @@ jobs:
                   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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - 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
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 3 --dist=loadfile -s --make-reports=tests_torch $(cat test_list.txt) | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_torch_all:
+        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 espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
+            - 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:
@@ -166,13 +284,53 @@ jobs:
                   keys:
                       - v0.4-tf-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]
+            - run: pip install tensorflow_probability
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_tf $(cat test_list.txt) | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_tf_all:
+        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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]
+            - run: pip install tensorflow_probability
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
+            - 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:
@@ -193,13 +351,51 @@ jobs:
                 keys:
                     - v0.4-flax-{{ checksum "setup.py" }}
                     - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - run: pip install --upgrade pip
-            - run: sudo pip install .[flax,testing,sentencepiece]
+            - run: pip install .[flax,testing,sentencepiece,flax-speech,vision]
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_flax $(cat test_list.txt) | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_flax_all:
+        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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[flax,testing,sentencepiece,vision,flax-speech]
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
+            - 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:
@@ -221,15 +417,54 @@ jobs:
                   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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - 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
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch -m is_pipeline_test $(cat test_list.txt) | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_pipelines_torch_all:
+        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 espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.10.0+cpu.html
+            - run: pip install https://github.com/kpu/kenlm/archive/master.zip
+            - 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:
@@ -253,11 +488,48 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
             - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
+            - run: pip install tensorflow_probability
             - 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
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf $(cat test_list.txt) -m is_pipeline_test | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_pipelines_tf_all:
+        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]
+            - run: pip install tensorflow_probability
+            - 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:
@@ -283,7 +555,10 @@ jobs:
                   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
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -s --make-reports=tests_custom_tokenizers ./tests/test_tokenization_bert_japanese.py | tee tests_output.txt
+                  fi
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -304,19 +579,119 @@ jobs:
                   keys:
                       - v0.4-torch_examples-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,sentencepiece,testing]
+            - run: pip install .[sklearn,torch,sentencepiece,testing,torch-speech]
             - 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
+            - run: python utils/tests_fetcher.py --filters examples tests | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_torch ./examples/pytorch/ | tee tests_output.txt
+                  fi
             - store_artifacts:
                   path: ~/transformers/examples_output.txt
             - store_artifacts:
                   path: ~/transformers/reports
 
+    run_examples_torch_all:
+        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: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng
+            - run: pip install --upgrade pip
+            - run: pip install .[sklearn,torch,sentencepiece,testing,torch-speech]
+            - 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_examples_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_examples-{{ checksum "setup.py" }}
+                    - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: sudo pip install .[flax,testing,sentencepiece]
+            - run: pip install -r examples/flax/_tests_requirements.txt
+            - save_cache:
+                  key: v0.4-flax_examples-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python utils/tests_fetcher.py --filters examples tests | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_flax ./examples/flax/ | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/flax_examples_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_examples_flax_all:
+        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_examples-{{ checksum "setup.py" }}
+                    - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: sudo pip install .[flax,testing,sentencepiece]
+            - run: pip install -r examples/flax/_tests_requirements.txt
+            - save_cache:
+                  key: v0.4-flax_examples-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: |
+                  TRANSFORMERS_IS_CI=1 python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_flax ./examples/flax/ | tee examples_output.txt
+            - store_artifacts:
+                  path: ~/transformers/flax_examples_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_tests_hub:
         working_directory: ~/transformers
         docker:
@@ -343,56 +718,117 @@ jobs:
                   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"
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
             - store_artifacts:
-                path: ./docs/_build
-
-    deploy_doc:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -sv --make-reports=tests_hub $(cat test_list.txt) -m is_staging_test | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_hub_all:
         working_directory: ~/transformers
         docker:
-            - image: circleci/python:3.6
+            - 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:
-            - 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-hub-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
-            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+            - 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 ."[docs]"
+            - run: pip install .[torch,sentencepiece,testing]
             - save_cache:
-                  key: v0.4-deploy_doc-{{ checksum "setup.py" }}
+                  key: v0.4-hub-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: ./.circleci/deploy.sh
+            - run: |
+                  python -m pytest -sv --make-reports=tests_hub tests -m is_staging_test | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
+    run_tests_onnxruntime:
+        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: pip install --upgrade pip
+            - run: pip install .[torch,testing,sentencepiece,onnxruntime]
+            - save_cache:
+                  key: v0.4-onnx-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_onnx $(cat test_list.txt) -k onnx | tee tests_output.txt
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+    
+    run_tests_onnxruntime_all:
+        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: pip install --upgrade pip
+            - run: pip install .[torch,testing,sentencepiece,onnxruntime]
+            - save_cache:
+                  key: v0.4-onnx-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: |
+                  python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_onnx tests -k onnx | tee tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
 
     check_code_quality:
         working_directory: ~/transformers
         docker:
             - image: circleci/python:3.6
-        resource_class: medium
+        resource_class: large
         environment:
             TRANSFORMERS_IS_CI: yes
         parallelism: 1
@@ -403,7 +839,6 @@ jobs:
                       - 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: v0.4-code_quality-{{ checksum "setup.py" }}
@@ -413,23 +848,73 @@ jobs:
             - run: isort --check-only examples tests src utils
             - run: python utils/custom_init_isort.py --check_only
             - run: flake8 examples tests src utils
-            - 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/style_doc.py src/transformers docs/source --max_len 119 --check_only
 
     check_repository_consistency:
         working_directory: ~/transformers
         docker:
             - image: circleci/python:3.6
-        resource_class: small
+        resource_class: large
+        environment:
+            TRANSFORMERS_IS_CI: yes
         parallelism: 1
         steps:
             - checkout
-            - run: pip install requests
-            - run: python ./utils/link_tester.py
+            - restore_cache:
+                  keys:
+                      - v0.4-repository_consistency-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
+            - run: pip install --upgrade pip
+            - run: pip install .[all,quality]
+            - save_cache:
+                  key: v0.4-repository_consistency-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - 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: make deps_table_check_updated
+            - run: python utils/tests_fetcher.py --sanity_check
+
+    run_tests_layoutlmv2:
+        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 .[torch,testing,vision]
+            - run: pip install torchvision
+            - run: python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
+            - run: sudo apt install tesseract-ocr
+            - run: pip install pytesseract
+            - save_cache:
+                  key: v0.4-torch-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python utils/tests_fetcher.py | tee test_preparation.txt
+            - store_artifacts:
+                  path: ~/transformers/test_preparation.txt
+            - run: |
+                  if [ -f test_list.txt ]; then
+                    python -m pytest -n 1 tests/*layoutlmv2* --dist=loadfile -s --make-reports=tests_layoutlmv2 --durations=100
+                  fi
+            - store_artifacts:
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
 
 # TPU JOBS
     run_examples_tpu:
@@ -474,6 +959,7 @@ workflows:
             - check_code_quality
             - check_repository_consistency
             - run_examples_torch
+            - run_examples_flax
             - run_tests_custom_tokenizers
             - run_tests_torch_and_tf
             - run_tests_torch_and_flax
@@ -482,9 +968,30 @@ workflows:
             - run_tests_flax
             - run_tests_pipelines_torch
             - run_tests_pipelines_tf
+            - run_tests_onnxruntime
             - run_tests_hub
-            - build_doc
-            - deploy_doc: *workflow_filters
+            - run_tests_layoutlmv2
+    nightly:
+        triggers:
+            - schedule:
+                cron: "0 0 * * *"
+                filters:
+                    branches:
+                        only:
+                            - master
+        jobs:
+            - run_examples_torch_all
+            - run_examples_flax_all
+            - run_tests_torch_and_tf_all
+            - run_tests_torch_and_flax_all
+            - run_tests_torch_all
+            - run_tests_tf_all
+            - run_tests_flax_all
+            - run_tests_pipelines_torch_all
+            - run_tests_pipelines_tf_all
+            - run_tests_onnxruntime_all
+            - run_tests_hub_all
+
 #    tpu_testing_jobs:
 #        triggers:
 #            - schedule:

.circleci/deploy.sh

@@ -1,67 +0,0 @@
-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

.github/ISSUE_TEMPLATE/bug-report.md

@@ -27,30 +27,39 @@ assignees: ''
 
 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
+- ALBERT, BERT, XLM, DeBERTa, DeBERTa-v2, ELECTRA, MobileBert, SqueezeBert: @LysandreJik
+- T5, BART, Marian, Pegasus, EncoderDecoder: @patrickvonplaten
+- Blenderbot, MBART: @patil-suraj
+- Longformer, Reformer, TransfoXL, XLNet, FNet, BigBird: @patrickvonplaten
+- FSMT: @stas00
+- Funnel: @sgugger
+- GPT-2, GPT: @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, SpeechEncoderDecoder, UniSpeech, UniSpeechSAT, SEW, SEW-D, Speech2Text: @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
-- tokenizers: @LysandreJik
-- trainer: @sgugger
-- pipelines: @LysandreJik
+- Benchmarks: @patrickvonplaten
+- Deepspeed: @stas00
+- Ray/raytune: @richardliaw, @amogkam
+- Text generation: @patrickvonplaten @narsil
+- Tokenizers: @LysandreJik
+- Trainer: @sgugger
+- Pipelines: @Narsil
+- Speech: @patrickvonplaten, @anton-l
+- Vision: @NielsRogge, @sgugger
 
 Documentation: @sgugger
 
 Model hub:
 
-- for issues with a model report at https://discuss.huggingface.co/ and tag the model's creator.
+- for issues with a model, report at https://discuss.huggingface.co/ and tag the model's creator.
 
 HF projects:
 
@@ -60,6 +69,9 @@ HF projects:
 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
 

.github/conda/meta.yaml

@@ -26,7 +26,7 @@ requirements:
     - regex !=2019.12.17
     - protobuf
     - tokenizers >=0.10.1,<0.11.0
-    - pyyaml
+    - pyyaml >=5.1
   run:
     - python
     - numpy >=1.17
@@ -41,7 +41,7 @@ requirements:
     - regex !=2019.12.17
     - protobuf
     - tokenizers >=0.10.1,<0.11.0
-    - pyyaml
+    - pyyaml >=5.1
 
 test:
   imports:

.github/workflows/TROUBLESHOOT.md

@@ -0,0 +1,9 @@
+# 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/build_doc_test.yml

@@ -0,0 +1,50 @@
+name: Documentation test build
+
+on:
+  pull_request:
+    paths:
+      - "src/**"
+      - "docs/**"
+      - ".github/**"
+
+jobs:
+  build_and_package:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        shell: bash -l {0}
+
+    steps:
+      - uses: actions/checkout@v2
+      
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: ~/.cache/pip
+          key: v1-test_build_doc
+          restore-keys: |
+            v1-test_build_doc-${{ hashFiles('setup.py') }}
+            v1-test_build_doc
+
+      - name: Setup environment
+        run: |
+          pip install --upgrade pip
+          sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+
+          pip install git+https://github.com/huggingface/doc-builder
+          pip install .[dev]
+
+          export TORCH_VERSION=$(python -c "from torch import version; print(version.__version__.split('+')[0])")
+          pip install torch-scatter -f https://data.pyg.org/whl/torch-${TORCH_VERSION}+cpu.html
+
+          pip install torchvision
+          python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
+
+          sudo apt install tesseract-ocr
+          pip install pytesseract
+          pip install pytorch-quantization --extra-index-url https://pypi.ngc.nvidia.com
+
+      - name: Make documentation
+        run: |
+          doc-builder build transformers ./docs/source

.github/workflows/build_documentation.yml

@@ -0,0 +1,102 @@
+name: Build documentation
+
+on:
+  push:
+    branches:
+      - master
+      - doc-builder*
+      - v*-release
+
+jobs:
+  build_and_package:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        shell: bash -l {0}
+
+    steps:
+      - uses: actions/checkout@v2
+        with:
+          repository: 'huggingface/doc-builder'
+          path: doc-builder
+          token: ${{ secrets.HUGGINGFACE_PUSH }}
+      
+      - uses: actions/checkout@v2
+        with:
+          repository: 'huggingface/transformers'
+          path: transformers
+      
+      - uses: actions/checkout@v2
+        with:
+          repository: 'huggingface/notebooks'
+          path: notebooks
+          token: ${{ secrets.HUGGINGFACE_PUSH }}
+      
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: ~/.cache/pip
+          key: v1-test_build_doc
+          restore-keys: |
+            v1-test_build_doc-${{ hashFiles('setup.py') }}
+            v1-test_build_doc
+
+      - name: Setup environment
+        run: |
+          sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+
+          pip install git+https://github.com/huggingface/doc-builder
+          cd transformers
+          pip install .[dev]
+          cd ..
+
+          export TORCH_VERSION=$(python -c "from torch import version; print(version.__version__.split('+')[0])")
+          pip install torch-scatter -f https://data.pyg.org/whl/torch-${TORCH_VERSION}+cpu.html
+
+          pip install torchvision
+          python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
+
+          sudo apt install tesseract-ocr
+          pip install pytesseract
+          pip install pytorch-quantization --extra-index-url https://pypi.ngc.nvidia.com
+          pip install https://github.com/kpu/kenlm/archive/master.zip
+
+      - name: Setup git
+        run: |
+          git config --global user.name "Hugging Face"
+          git config --global user.email transformers@huggingface.co
+          
+          cd doc-builder
+          git pull origin main
+          cd ..
+
+          cd notebooks
+          git pull origin master
+          cd ..
+
+      - name: Make documentation
+        run: |
+          doc-builder build transformers transformers/docs/source --build_dir doc-builder/build --notebook_dir notebooks/transformers_doc --clean
+
+      - name: Push to repositories
+        run: |
+          cd doc-builder
+          if [[ `git status --porcelain` ]]; then
+            git add build
+            git commit -m "Updated with commit ${{ github.sha }} \n\nSee: https://github.com/huggingface/transformers/commit/${{ github.sha }}"
+            git push origin main
+          else
+            echo "No diff in the documentation."
+          fi
+          cd ..
+
+          cd notebooks
+          if [[ `git status --porcelain` ]]; then
+            git add transformers_doc
+            git commit -m "Updated Transformer doc notebooks with commit ${{ github.sha }} \n\nSee: https://github.com/huggingface/transformers/commit/${{ github.sha }}"
+            git push origin master
+          else
+            echo "No diff in the notebooks."
+          fi
+          cd ..

.github/workflows/doctests.yml

@@ -0,0 +1,42 @@
+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
+  PYTEST_TIMEOUT: 600
+
+jobs:
+  run_doctests:
+    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: 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 .[dev]
+
+      - name: Run doctests
+        run: |
+          pytest --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure

.github/workflows/model-templates.yml

@@ -36,7 +36,7 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip install --upgrade pip
+          pip install --upgrade pip!=21.3
           sudo apt -y update && sudo apt install -y libsndfile1-dev
           pip install .[dev]
       - name: Create model files
@@ -47,6 +47,8 @@ jobs:
           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
@@ -59,7 +61,7 @@ jobs:
       - name: Run style changes
         run: |
           git fetch origin master:master
-          make fixup
+          make style && make quality
 
       - name: Failure short reports
         if: ${{ always() }}

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

@@ -0,0 +1,246 @@
+name: Self-hosted runner; Nightly (scheduled)
+
+on:
+    push:
+        branches:
+            - nightly_ci*
+    repository_dispatch:
+    schedule:
+        - cron: "0 0 */3 * *"
+
+env:
+    HF_HOME: /mnt/cache
+    TRANSFORMERS_IS_CI: yes
+    RUN_SLOW: yes
+    OMP_NUM_THREADS: 16
+    MKL_NUM_THREADS: 16
+    PYTEST_TIMEOUT: 600
+    SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
+
+jobs:
+    run_all_tests_torch_gpu:
+        runs-on: [self-hosted, docker-gpu, single-gpu]
+        container:
+            image: pytorch/pytorch:1.10.0-cuda11.3-cudnn8-runtime
+            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 libsndfile1-dev git
+                  pip install --upgrade pip
+                  pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+                  pip install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu113/torch_nightly.html -U
+
+            - name: Are GPUs recognized by our DL frameworks
+              run: |
+                utils/print_env_pt.py
+
+            - name: Run all tests on GPU
+              run: |
+                  python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_torch_gpu tests
+
+            - name: Failure short reports
+              if: ${{ always() }}
+              run: cat reports/tests_torch_gpu_failures_short.txt
+
+            - name: Run examples tests on GPU
+              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
+                  python -m pytest -n 1 -v --dist=loadfile --make-reports=examples_torch_gpu examples
+
+            - name: Failure short reports
+              if: ${{ always() }}
+              run: cat reports/examples_torch_gpu_failures_short.txt
+
+            - name: Run all pipeline tests on GPU
+              if: ${{ always() }}
+              env:
+                  RUN_PIPELINE_TESTS: yes
+              run: |
+                  python -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_gpu tests
+
+            - name: Failure short reports
+              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: run_all_tests_torch_gpu_test_reports
+                  path: reports
+
+    run_all_tests_torch_multi_gpu:
+        runs-on: [self-hosted, docker-gpu, multi-gpu]
+        container:
+            image: pytorch/pytorch:1.10.0-cuda11.3-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
+              continue-on-error: true
+              run: |
+                  nvidia-smi
+
+            - name: Install dependencies
+              run: |
+                  apt -y update && apt install -y libsndfile1-dev git
+                  pip install --upgrade pip
+                  pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+                  pip install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu113/torch_nightly.html -U
+
+            - name: Are GPUs recognized by our DL frameworks
+              run: |
+                utils/print_env_pt.py
+
+            - name: Run all tests on GPU
+              env:
+                  MKL_SERVICE_FORCE_INTEL: 1
+              run: |
+                  python -m pytest -n 1 -v --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 -v --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_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 --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu113/torch_nightly.html -U
+                  pip install .[testing,deepspeed]
+                  pip install git+https://github.com/microsoft/DeepSpeed
+
+            - name: Are GPUs recognized by our DL frameworks
+              run: |
+                utils/print_env_pt.py
+
+            - name: Run all tests on GPU
+              run: |
+                  python -m pytest -n 1 -v --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
+              continue-on-error: true
+              run: |
+                  nvidia-smi
+
+            - name: Install dependencies
+              run: |
+                  apt -y update && apt install -y libaio-dev
+                  pip install --upgrade pip
+                  pip install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu113/torch_nightly.html -U
+                  rm -rf ~/.cache/torch_extensions/ # shared between conflicting builds
+                  pip install .[testing,fairscale]
+                  pip install git+https://github.com/microsoft/DeepSpeed # testing bleeding edge
+
+            - name: Are GPUs recognized by our DL frameworks
+              run: |
+                utils/print_env_pt.py
+
+            - name: Run all tests on GPU
+              run: |
+                  python -m pytest -n 1 -v --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: [
+                run_all_tests_torch_gpu,
+                run_all_tests_torch_multi_gpu,
+                run_all_tests_torch_cuda_extensions_gpu,
+                run_all_tests_torch_cuda_extensions_multi_gpu
+        ]
+        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 }}
+                  CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
+                  CI_SLACK_CHANNEL_ID_PAST_FUTURE: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
+
+              run: |
+                  pip install slack_sdk
+                  python utils/notification_service.py scheduled nightly-torch

.github/workflows/self-push.yml

@@ -11,6 +11,7 @@ on:
       - "tests/**"
       - ".github/**"
       - "templates/**"
+      - "utils/**"
   repository_dispatch:
 
 env:
@@ -18,6 +19,7 @@ env:
   TRANSFORMERS_IS_CI: yes
   OMP_NUM_THREADS: 8
   MKL_NUM_THREADS: 8
+  PYTEST_TIMEOUT: 60
 
 jobs:
   run_tests_torch_gpu:
@@ -26,32 +28,45 @@ jobs:
       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: Install dependencies
+        run: |
+          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
+          apt install -y libsndfile1-dev espeak-ng
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
+
       - name: Launcher docker
         uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
 
       - 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())"
+          utils/print_env_pt.py
+
+      - name: Fetch the tests to run
+        run: |
+          python 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: test_preparation.txt
 
       - name: Run all non-slow tests on GPU
         run: |
-          python -m pytest -n 2 --dist=loadfile --make-reports=tests_torch_gpu tests
+          if [ -f test_list.txt ]; then
+            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_torch_gpu $(cat test_list.txt)
+          fi
 
       - name: Failure short reports
-        if: ${{ always() }}
+        if: ${{ failure() }}
         run: cat reports/tests_torch_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
@@ -61,48 +76,120 @@ jobs:
           name: run_all_tests_torch_gpu_test_reports
           path: reports
 
-  run_tests_tf_gpu:
-    runs-on: [self-hosted, docker-gpu, single-gpu]
-    timeout-minutes: 120
+  run_tests_flax_gpu:
+    runs-on: [self-hosted, docker-gpu-test, single-gpu]
     container:
       image: tensorflow/tensorflow:2.4.1-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git espeak-ng
+          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+          pip install --upgrade pip
+          pip install .[sklearn,testing,sentencepiece,flax,flax-speech,vision]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
+
       - name: Launcher docker
         uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
 
       - name: NVIDIA-SMI
+        continue-on-error: true
         run: |
           nvidia-smi
 
-      - name: Install dependencies
-        run: |
-          pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,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')))"
+          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+
+      - name: Fetch the tests to run
+        run: |
+          python 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: test_preparation.txt
 
       - 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
+          if [ -f test_list.txt ]; then
+            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_gpu $(cat test_list.txt)
+          fi
 
       - name: Failure short reports
-        if: ${{ always() }}
-        run: cat reports/tests_tf_gpu_failures_short.txt
+        if: ${{ failure() }}
+        run: cat reports/tests_flax_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
+          name: run_all_tests_flax_gpu_test_reports
           path: reports
 
+#  run_tests_tf_gpu:
+#    runs-on: [self-hosted, docker-gpu, single-gpu]
+#    timeout-minutes: 120
+#    container:
+#      image: tensorflow/tensorflow:2.4.1-gpu
+#      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+#    steps:
+#      - name: Install dependencies
+#        run: |
+#          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git espeak-ng
+#          pip install --upgrade pip
+#          pip install .[sklearn,testing,onnxruntime,sentencepiece,tf-speech]
+#          pip install https://github.com/kpu/kenlm/archive/master.zip
+#
+#      - name: Launcher docker
+#        uses: actions/checkout@v2
+#        with:
+#          fetch-depth: 2
+#
+#      - name: NVIDIA-SMI
+#        run: |
+#          nvidia-smi
+#
+#      - 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: Fetch the tests to run
+#        run: |
+#          python 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: test_preparation.txt
+#
+#      - name: Run all non-slow tests on GPU
+#        env:
+#          TF_NUM_INTRAOP_THREADS: 8
+#          TF_NUM_INTEROP_THREADS: 1
+#        run: |
+#          if [ -f test_list.txt ]; then
+#            python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu $(cat test_list.txt)
+#          fi
+#
+#      - name: Failure short reports
+#        if: ${{ failure() }}
+#        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]
@@ -110,34 +197,47 @@ jobs:
       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: Install dependencies
+        run: |
+          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git espeak-ng
+          apt install -y libsndfile1-dev espeak-ng
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
       - name: Launcher docker
         uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
 
       - name: NVIDIA-SMI
+        continue-on-error: true
         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())"
+          utils/print_env_pt.py
+
+      - name: Fetch the tests to run
+        run: |
+          python 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: test_preparation.txt
 
       - name: Run all non-slow tests on GPU
         env:
           MKL_SERVICE_FORCE_INTEL: 1
         run: |
-          python -m pytest -n 2 --dist=loadfile --make-reports=tests_torch_multi_gpu tests
+          if [ -f test_list.txt ]; then
+            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_torch_multi_gpu $(cat test_list.txt)
+          fi
 
       - name: Failure short reports
-        if: ${{ always() }}
+        if: ${{ failure() }}
         run: cat reports/tests_torch_multi_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
@@ -147,47 +247,119 @@ jobs:
           name: run_all_tests_torch_multi_gpu_test_reports
           path: reports
 
-  run_tests_tf_multi_gpu:
-    runs-on: [self-hosted, docker-gpu, multi-gpu]
-    timeout-minutes: 120
-    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
+#  run_tests_flax_multi_gpu:
+#    runs-on: [self-hosted, docker-gpu, multi-gpu]
+#    container:
+#      image: tensorflow/tensorflow:2.4.1-gpu
+#      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+#    steps:
+#      - name: Install dependencies
+#        run: |
+#          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git espeak-ng
+#          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+#          pip install --upgrade pip
+#          pip install .[sklearn,testing,sentencepiece,flax,flax-speech,vision]
+#          pip install https://github.com/kpu/kenlm/archive/master.zip
+#
+#      - name: Launcher docker
+#        uses: actions/checkout@v2
+#        with:
+#          fetch-depth: 2
+#
+#      - name: NVIDIA-SMI
+#        continue-on-error: true
+#        run: |
+#          nvidia-smi
+#
+#      - name: Are GPUs recognized by our DL frameworks
+#        run: |
+#          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+#          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+#
+#      - name: Fetch the tests to run
+#        run: |
+#          python 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: test_preparation.txt
+#
+#      - name: Run all non-slow tests on GPU
+#        run: |
+#          if [ -f test_list.txt ]; then
+#            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_multi_gpu $(cat test_list.txt)
+#          fi
+#
+#      - name: Failure short reports
+#        if: ${{ failure() }}
+#        run: cat reports/tests_flax_multi_gpu_failures_short.txt
+#
+#      - name: Test suite reports artifacts
+#        if: ${{ always() }}
+#        uses: actions/upload-artifact@v2
+#        with:
+#          name: run_all_tests_flax_multi_gpu_test_reports
+#          path: reports
 
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Install dependencies
-        run: |
-          pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,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 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: ${{ 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: run_all_tests_tf_multi_gpu_test_reports
-          path: reports
+#  run_tests_tf_multi_gpu:
+#    runs-on: [self-hosted, docker-gpu, multi-gpu]
+#    timeout-minutes: 120
+#    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: Install dependencies
+#        run: |
+#          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git espeak-ng
+#          pip install --upgrade pip
+#          pip install .[sklearn,testing,onnxruntime,sentencepiece,tf-speech]
+#          pip install https://github.com/kpu/kenlm/archive/master.zip
+#
+#      - name: Launcher docker
+#        uses: actions/checkout@v2
+#        with:
+#          fetch-depth: 2
+#
+#      - name: NVIDIA-SMI
+#        run: |
+#          nvidia-smi
+#
+#      - 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: Fetch the tests to run
+#        run: |
+#          python 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: test_preparation.txt
+#
+#      - name: Run all non-slow tests on GPU
+#        env:
+#          TF_NUM_INTRAOP_THREADS: 8
+#          TF_NUM_INTEROP_THREADS: 1
+#        run: |
+#          if [ -f test_list.txt ]; then
+#            python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu $(cat test_list.txt)
+#          fi
+#
+#      - name: Failure short reports
+#        if: ${{ failure() }}
+#        run: cat reports/tests_tf_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_tests_torch_cuda_extensions_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
@@ -197,6 +369,8 @@ jobs:
     steps:
       - name: Launcher docker
         uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
 
       - name: NVIDIA-SMI
         run: |
@@ -210,17 +384,26 @@ jobs:
 
       - 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())"
+          utils/print_env_pt.py
+
+      - name: Fetch the tests to run
+        run: |
+          python utils/tests_fetcher.py --diff_with_last_commit --filters tests/deepspeed tests/extended | tee test_preparation.txt
+
+      - name: Report fetched tests
+        uses: actions/upload-artifact@v2
+        with:
+          name: test_fetched
+          path: test_preparation.txt
 
       - 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
+          if [ -f test_list.txt ]; then
+            python -m pytest -n 1 --dist=loadfile -v --make-reports=tests_torch_cuda_extensions_gpu $(cat test_list.txt)
+          fi
 
       - name: Failure short reports
-        if: ${{ always() }}
+        if: ${{ failure() }}
         run: cat reports/tests_torch_cuda_extensions_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
@@ -238,8 +421,11 @@ jobs:
     steps:
       - name: Launcher docker
         uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
 
       - name: NVIDIA-SMI
+        continue-on-error: true
         run: |
           nvidia-smi
 
@@ -247,21 +433,31 @@ jobs:
         run: |
           apt -y update && apt install -y libaio-dev
           pip install --upgrade pip
+          rm -rf ~/.cache/torch_extensions/ # shared between conflicting builds
           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())"
+          utils/print_env_pt.py
+
+      - name: Fetch the tests to run
+        run: |
+          python utils/tests_fetcher.py --diff_with_last_commit --filters tests/deepspeed tests/extended | tee test_preparation.txt
+
+      - name: Report fetched tests
+        uses: actions/upload-artifact@v2
+        with:
+          name: test_fetched
+          path: test_preparation.txt
 
       - 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
+          if [ -f test_list.txt ]; then
+            python -m pytest -n 1 --dist=loadfile -v --make-reports=tests_torch_cuda_extensions_multi_gpu $(cat test_list.txt)
+          fi
 
       - name: Failure short reports
-        if: ${{ always() }}
+        if: ${{ failure() }}
         run: cat reports/tests_torch_cuda_extensions_multi_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
@@ -278,9 +474,9 @@ jobs:
     if: always()
     needs: [
         run_tests_torch_gpu,
-        run_tests_tf_gpu,
+#        run_tests_tf_gpu,
         run_tests_torch_multi_gpu,
-        run_tests_tf_multi_gpu,
+#        run_tests_tf_multi_gpu,
         run_tests_torch_cuda_extensions_gpu,
         run_tests_torch_cuda_extensions_multi_gpu
     ]

.github/workflows/self-scheduled.yml

@@ -14,6 +14,8 @@ env:
   RUN_SLOW: yes
   OMP_NUM_THREADS: 16
   MKL_NUM_THREADS: 16
+  PYTEST_TIMEOUT: 600
+  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
 
 jobs:
   run_all_tests_torch_gpu:
@@ -31,20 +33,18 @@ jobs:
 
       - name: Install dependencies
         run: |
-          apt -y update && apt install -y libsndfile1-dev
+          apt -y update && apt install -y libsndfile1-dev git espeak-ng
           pip install --upgrade pip
-          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
 
       - 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())"
+          utils/print_env_pt.py
 
       - name: Run all tests on GPU
         run: |
-          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_gpu tests
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_torch_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -60,7 +60,7 @@ jobs:
           TRANSFORMERS_IS_CI: yes
         run: |
           pip install -r examples/pytorch/_tests_requirements.txt
-          python -m pytest -n 1 --dist=loadfile --make-reports=examples_torch_gpu examples
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=examples_torch_gpu examples
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -71,7 +71,7 @@ jobs:
         env:
           RUN_PIPELINE_TESTS: yes
         run: |
-          python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_gpu tests
+          python -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -84,6 +84,47 @@ jobs:
           name: run_all_tests_torch_gpu_test_reports
           path: reports
 
+  run_all_tests_flax_gpu:
+    runs-on: [self-hosted, docker-gpu-test, single-gpu]
+    container:
+      image: tensorflow/tensorflow:2.4.1-gpu
+      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
+        continue-on-error: true
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          pip install --upgrade pip
+          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+          pip install .[flax,integrations,sklearn,testing,sentencepiece,flax-speech,vision]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_flax_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_flax_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_flax_gpu_test_reports
+          path: reports
+
   run_all_tests_tf_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
@@ -99,8 +140,11 @@ jobs:
 
       - name: Install dependencies
         run: |
+          apt -y update && apt install -y libsndfile1-dev git espeak-ng
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnx,sentencepiece]
+          pip install .[sklearn,testing,onnx,sentencepiece,tf-speech,vision]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
+
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -112,7 +156,7 @@ jobs:
           TF_NUM_INTEROP_THREADS: 1
           TF_NUM_INTRAOP_THREADS: 16
         run: |
-          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu tests
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_tf_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -125,7 +169,7 @@ jobs:
           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
+          python -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=tests_tf_pipeline_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -138,6 +182,45 @@ jobs:
           name: run_all_tests_tf_gpu_test_reports
           path: reports
 
+  run_all_examples_torch_xla_tpu:
+    runs-on: [self-hosted, docker-tpu-test, tpu-v3-8]
+    container:
+      image: gcr.io/tpu-pytorch/xla:nightly_3.8_tpuvm
+      options: --privileged -v "/lib/libtpu.so:/lib/libtpu.so" -v /mnt/cache/.cache/huggingface:/mnt/cache/ --shm-size 16G
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: Install dependencies
+        run: |
+          pip install --upgrade pip
+          pip install .[testing]
+
+      - name: Are TPUs recognized by our DL frameworks
+        env:
+          XRT_TPU_CONFIG: localservice;0;localhost:51011
+        run: |
+          python -c "import torch_xla.core.xla_model as xm; print(xm.xla_device())"
+
+      - name: Run example tests on TPU
+        env:
+          XRT_TPU_CONFIG: "localservice;0;localhost:51011"
+          MKL_SERVICE_FORCE_INTEL: "1"  # See: https://github.com/pytorch/pytorch/issues/37377
+
+        run: |
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_torch_xla_tpu examples/pytorch/test_xla_examples.py
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_xla_tpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_examples_torch_xla_tpu
+          path: reports
+
   run_all_tests_torch_multi_gpu:
     runs-on: [self-hosted, docker-gpu, multi-gpu]
     container:
@@ -148,27 +231,26 @@ jobs:
         uses: actions/checkout@v2
 
       - name: NVIDIA-SMI
+        continue-on-error: true
         run: |
           nvidia-smi
 
       - name: Install dependencies
         run: |
-          apt -y update && apt install -y libsndfile1-dev
+          apt -y update && apt install -y libsndfile1-dev git espeak-ng
           pip install --upgrade pip
-          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
 
       - 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())"
+          utils/print_env_pt.py
 
       - 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
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_torch_multi_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -179,7 +261,7 @@ jobs:
         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
+          python -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_multi_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -202,13 +284,16 @@ jobs:
         uses: actions/checkout@v2
 
       - name: NVIDIA-SMI
+        continue-on-error: true
         run: |
           nvidia-smi
 
       - name: Install dependencies
         run: |
+          apt -y update && apt install -y libsndfile1-dev git
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnx,sentencepiece]
+          pip install .[sklearn,testing,onnx,sentencepiece,tf-speech,vision]
+          pip install https://github.com/kpu/kenlm/archive/master.zip
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -220,7 +305,7 @@ jobs:
           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
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_tf_multi_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -233,7 +318,7 @@ jobs:
           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
+          python -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=tests_tf_pipeline_multi_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -246,6 +331,45 @@ jobs:
           name: run_all_tests_tf_multi_gpu_test_reports
           path: reports
 
+#  run_all_tests_flax_multi_gpu:
+#    runs-on: [self-hosted, docker-gpu, multi-gpu]
+#    container:
+#      image: tensorflow/tensorflow:2.4.1-gpu
+#      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: |
+#          pip install --upgrade pip
+#          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+#          pip install .[flax,integrations,sklearn,testing,sentencepiece,flax-speech,vision]
+#
+#      - name: Are GPUs recognized by our DL frameworks
+#        run: |
+#          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+#          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+#
+#      - name: Run all tests on GPU
+#        run: |
+#          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_flax_gpu tests
+#
+#      - name: Failure short reports
+#        if: ${{ always() }}
+#        run: cat reports/tests_flax_gpu_failures_short.txt
+#
+#      - name: Test suite reports artifacts
+#        if: ${{ always() }}
+#        uses: actions/upload-artifact@v2
+#        with:
+#          name: run_all_tests_flax_gpu_test_reports
+#          path: reports
+
   run_all_tests_torch_cuda_extensions_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
@@ -267,14 +391,11 @@ jobs:
 
       - 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())"
+          utils/print_env_pt.py
 
       - 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
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -297,6 +418,7 @@ jobs:
         uses: actions/checkout@v2
 
       - name: NVIDIA-SMI
+        continue-on-error: true
         run: |
           nvidia-smi
 
@@ -304,18 +426,16 @@ jobs:
         run: |
           apt -y update && apt install -y libaio-dev
           pip install --upgrade pip
+          rm -rf ~/.cache/torch_extensions/ # shared between conflicting builds
           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())"
+          utils/print_env_pt.py
 
       - 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
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_torch_cuda_extensions_multi_gpu tests/deepspeed tests/extended
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -349,6 +469,7 @@ jobs:
         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 }}
 
 
         run: |

.github/workflows/update_metdata.yml

@@ -0,0 +1,36 @@
+name: Update Transformers metadata
+
+on:
+  push:
+    branches:
+      - master
+      - update_transformers_metadata
+
+jobs:
+  build_and_package:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        shell: bash -l {0}
+
+    steps:
+      - uses: actions/checkout@v2
+      
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: ~/.cache/pip
+          key: v1-metadata
+          restore-keys: |
+            v1-metadata-${{ hashFiles('setup.py') }}
+            v1-metadata
+
+      - name: Setup environment
+        run: |
+          pip install git+https://github.com/huggingface/transformers#egg=transformers[dev]
+
+      - name: Update metadata
+        run: |
+          python utils/update_metadata.py --token ${{ secrets.SYLVAIN_HF_TOKEN }} --commit_sha ${{ github.sha }}
+

CITATION.cff

@@ -0,0 +1,82 @@
+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

@@ -124,7 +124,7 @@ 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.
 
@@ -175,34 +175,26 @@ Follow these steps to start contributing:
 5. Develop the features on your branch.
 
    As you work on the features, you should make sure that the test suite
-   passes:
+   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:
 
    ```bash
    $ make test
    ```
 
-   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:
+   For more information about tests, check out the
+   [dedicated documentation](https://huggingface.co/docs/transformers/testing)
 
-   ```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:
+   🤗 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
    $ make fixup
@@ -210,16 +202,55 @@ Follow these steps to start contributing:
 
    This target is also optimized to only work with files modified by the PR you're working on.
 
-   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, by
-   running `pip install .[tf,torch,docs]` once from the root of this repository
-   and then run:
+   If you prefer to run the checks one after the other, the following command apply the
+   style corrections:
 
    ```bash
-   $ make docs
+   $ 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:
+
+   ```bash
+   $ doc-builder build transformers docs/source/ --build_dir ~/tmp/test-build
+   ```
+
+   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:
 
@@ -273,8 +304,15 @@ Follow these steps to start contributing:
    - 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_ctrl.py` for an
+6. All public methods must have informative docstrings that work nicely with sphinx. See `modeling_bert.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
 
@@ -326,7 +364,7 @@ $ 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).
+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.
 

ISSUES.md

@@ -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 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.
+   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.
 
    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

@@ -21,35 +21,43 @@ 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 source code meets quality standards
+# Check that the repo is in a good state
 
-extra_quality_checks:
+repo-consistency:
 	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/tests_fetcher.py --sanity_check
 
 # this target runs checks on all files
+
 quality:
 	black --check $(check_dirs)
 	isort --check-only $(check_dirs)
 	python utils/custom_init_isort.py --check_only
 	flake8 $(check_dirs)
-	${MAKE} extra_quality_checks
+#	python utils/style_doc.py src/transformers docs/source --max_len 119 --check_only
 
 # 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/style_doc.py src/transformers docs/source --max_len 119
+#	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 $(check_dirs)
 	isort $(check_dirs)
@@ -58,7 +66,7 @@ style:
 
 # 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 extra_quality_checks
+fixup: modified_only_fixup extra_style_checks autogenerate_code repo-consistency
 
 # Make marked copies of snippets of codes conform to the original
 
@@ -83,11 +91,6 @@ 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

@@ -16,7 +16,7 @@ limitations under the License.
 
 <p align="center">
     <br>
-    <img src="https://raw.githubusercontent.com/huggingface/transformers/master/docs/source/imgs/transformers_logo_name.png" width="400"/>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
 <p>
 <p align="center">
@@ -26,8 +26,8 @@ limitations under the 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/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 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">
@@ -38,15 +38,32 @@ limitations under the License.
     <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/master/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_ko.md">한국어</a>
+    <p>
+</h4>
+
 <h3 align="center">
-    <p>State-of-the-art Natural Language Processing for Jax, PyTorch and TensorFlow</p>
+    <p>State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow</p>
 </h3>
 
 <h3 align="center">
-    <a href="https://hf.co/course"><img src="https://raw.githubusercontent.com/huggingface/transformers/master/docs/source/imgs/course_banner.png"></a>
+    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
 </h3>
 
-🤗 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 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 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.
 
@@ -57,6 +74,8 @@ limitations under the License.
 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+)
@@ -65,6 +84,15 @@ 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)
+- [Image 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)
+
 **[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
@@ -75,7 +103,7 @@ Here are a few examples:
 
 ## Quick tour
 
-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:
+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:
 
 ```python
 >>> from transformers import pipeline
@@ -103,7 +131,7 @@ Many NLP tasks have a pre-trained `pipeline` ready to go. For example, we can ea
 
 ```
 
-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).
+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/docs/transformers/task_summary).
 
 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
@@ -128,12 +156,12 @@ 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 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.
+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/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.
 
 ## Why should I use transformers?
 
 1. Easy-to-use state-of-the-art models:
-    - High performance on NLU and NLG tasks.
+    - High performance on natural language understanding & generation, computer vision, and audio 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.
@@ -141,11 +169,11 @@ 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 2,000 pretrained models, some in more than 100 languages.
+    - Dozens of architectures with over 20,000 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 frameworks at will.
+    - Move a single model between TF2.0/PyTorch/JAX frameworks at will.
     - Seamlessly pick the right framework for training, evaluation and production.
 
 1. Easily customize a model or an example to your needs:
@@ -178,7 +206,7 @@ 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/transformers/installation.html#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/docs/transformers/installation#installing-from-source).
 
 ### With conda
 
@@ -198,90 +226,118 @@ Follow the installation pages of Flax, PyTorch or TensorFlow to see how to insta
 
 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/transformers/model_summary.html) for a high-level summary of each them):
+🤗 Transformers currently provides the following architectures (see [here](https://huggingface.co/docs/transformers/model_summary) for a high-level summary of each them):
 
-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
+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. **[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. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bertgeneration)** (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/docs/transformers/model_doc/bigbird)** (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/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. **[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. **[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. **[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. **[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. **[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. **[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/master/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/research_projects/distillation) and a German version of DistilBERT.
+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. **[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. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoderdecoder)** (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. **[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. **[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. **[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. **[GPT](https://huggingface.co/docs/transformers/model_doc/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-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. **[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. **[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/master/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. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transformerxl)** (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. **[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. **[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. **[WavLM](https://huggingface.co/docs/transformers/master/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. **[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. **[Wav2Vec2Phoneme](https://huggingface.co/docs/master/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. **[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/xlmprophetnet)** (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/xlmroberta)** (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/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. **[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. **[XLS-R](https://huggingface.co/docs/master/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. 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/transformers/index.html#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/docs/transformers/index#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://huggingface.co/transformers/examples.html).
+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/docs/transformers/examples).
 
 
 ## Learn more
 
 | Section | Description |
 |-|-|
-| [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 |
+| [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/docstransformers/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/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` |
+| [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` |
 
 ## Citation
 

README_ko.md

@@ -0,0 +1,334 @@
+<!---
+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/master">
+    </a>
+    <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>
+    <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/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>
+        <a href="https://github.com/huggingface/transformers/">English</a> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_zh-hant.md">繁體中文</a> |
+        <b>한국어</b>
+    <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/master/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/bertgeneration)** (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/bigbird)** (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. **[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. **[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. **[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. **[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. **[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/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/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. **[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/encoderdecoder)** (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](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. **[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. **[GPT](https://huggingface.co/docs/transformers/model_doc/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-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. **[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/master/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. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transformerxl)** (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. **[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. **[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. **[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. **[Wav2Vec2Phoneme](https://huggingface.co/docs/master/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/master/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. **[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/xlmprophetnet)** (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/xlmroberta)** (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/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/master/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. 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/master/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

@@ -0,0 +1,359 @@
+<!---
+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/master">
+    </a>
+    <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>
+    <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/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>
+        <a href="https://github.com/huggingface/transformers/">English</a> |
+        <b>简体中文</b> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_ko.md">한국어</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：
+    - 研究人员可以分享亿训练的模型而非次次从头开始训练
+    - 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/master/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/bertgeneration)** (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/bigbird)** (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. **[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. **[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. **[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. **[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. **[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/master/examples/distillation), RoBERTa 到 [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/distillation), Multilingual BERT 到 [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) 和德语版 DistilBERT。
+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. **[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/encoderdecoder)** (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. **[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. **[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. **[GPT](https://huggingface.co/docs/transformers/model_doc/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-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. **[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/master/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. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transformerxl)** (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. **[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. **[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. **[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. **[Wav2Vec2Phoneme](https://huggingface.co/docs/master/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/master/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. **[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/xlmprophetnet)** (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/xlmroberta)** (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. **[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/master/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. 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/docstransformers/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/master/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

@@ -0,0 +1,371 @@
+<!---
+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/master">
+    </a>
+    <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>
+    <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/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>
+        <a href="https://github.com/huggingface/transformers/">English</a> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_zh-hans.md">简体中文</a> |
+        <b>繁體中文</b> |
+        <a href="https://github.com/huggingface/transformers/blob/master/README_ko.md">한국어</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：
+    - 研究人員可以分享預訓練的模型而非從頭開始訓練
+    - 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/master/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/bertgeneration)** (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/bigbird)** (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. **[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. **[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. **[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. **[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. **[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/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/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. **[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/encoderdecoder)** (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](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. **[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. **[GPT](https://huggingface.co/docs/transformers/model_doc/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-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. **[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/master/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. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transformerxl)** (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. **[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. **[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. **[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. **[Wav2Vec2Phoneme](https://huggingface.co/docs/master/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/master/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. **[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/xlmprophetnet)** (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/xlmroberta)** (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/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/master/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. 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/master/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"
+}
+```

docs/Makefile

@@ -1,19 +0,0 @@
-# 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

@@ -23,6 +23,12 @@ you can install them with the following command, at the root of the code reposit
 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**
 
@@ -31,88 +37,79 @@ check how they look like before committing for instance). You don't have to comm
 
 ---
 
-## 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 `sphinx`, you can build the documentation by running the following command in the `/docs` folder:
+Once you have setup the `doc-builder` and additional packages, you can generate the documentation by typing th
+folowwing command:
 
 ```bash
-make html
+doc-builder build transformers docs/source/ --build_dir ~/tmp/test-build
 ```
 
-A folder called ``_build/html`` should have been created. You can now open the file ``_build/html/index.html`` in your
-browser.
+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.
 
 ---
 **NOTE**
 
-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
-```
+It's not possible to see locally how the final documentation will look like for now. We are working on solutions to
+enable this, but any pre-visualiser of Markdown file should already give you a good idea of the result!
 
 ---
 
-It should build the static app that will be available under `/docs/_build/html`
+## Adding a new element to the navigation bar
 
-## Adding a new element to the tree (toc-tree)
+Accepted files are reStructuredText (.rst) and Markdown (.md or .mdx). We are progressively moving away from rst so you should
+create any new documentation file in the .mdx format. 
+
+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/master/docs/source/_toctree.yml) file.
+
+## Renaming section headers and moving sections
+
+It helps to keep the old links working when renaming 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 be 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 sections set please see the very end of [the Trainer doc](https://github.com/huggingface/transformers/blob/master/docs/source/main_classes/trainer.mdx).
 
-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.
 
 ## Preview the documentation in a pull request
 
-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.
+Coming soon!
 
 ## Writing Documentation - Specification
 
 The `huggingface/transformers` documentation follows the
-[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)).
-
+[Google documentation](https://sphinxcontrib-napoleon.readthedocs.io/en/latest/example_google.html) style for docstrings,
+although we can write them directly in Markdown. Parts of it are written in ReStructuredText
+([Sphinx simple documentation](https://www.sphinx-doc.org/en/master/usage/restructuredtext/index.html) but we are
+updating those.
 
 ### 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/index.rst` on the correct toc-tree.
+- Link that file in `./source/_toctree.yml` 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 section two, three or
@@ -122,8 +119,8 @@ four.
 
 When adding a new model:
 
-- 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.
+- 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`.
 - Write a short overview of the model:
     - Overview with paper & authors
     - Paper abstract
@@ -138,63 +135,79 @@ When adding a new model:
     - TensorFlow base model
     - TensorFlow head models
 
-These classes should be added using the RST syntax. Usually as follows:
-```
-XXXConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+These classes should be added using our Markdown syntax. Usually as follows:
 
-.. autoclass:: transformers.XXXConfig
-    :members:
+```
+## XXXConfig
+
+[[autodoc]] XXXConfig
 ```
 
 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
 
-.. autoclass:: transformers.XXXTokenizer
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
+[[autodoc]] XXXTokenizer
+    - 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 method like `__call__` are not documented
+byt 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 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`.
+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`.
 
-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\`
+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 mentioning a function, it is recommended to use the :func: syntax as the mentioned function will be automatically
-linked by Sphinx: :func:\`~transformers.function\`.
+If you want to create a link to some internal class or function, you need to
+provide its path. For instance: \[\`file_utils.ModelOutput\`\]. This will be converted into a link with
+`file_utils.ModelOutput` in the description. To get rid of the path and only keep the name of the object you are
+linking to, add a ~: \[\`~file_utils.ModelOutput\`\] will generate a link with `ModelOutput` in the description.
 
-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>\`__
+The same wroks for methods so you can either use \[\`XXXClass.method\`\] or \[~\`XXXClass.method\`\].
 
 #### Defining arguments in a method
 
-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.
+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 th argument.
 
 Here's an example showcasing everything so far:
 
 ```
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`):
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
             Indices of input sequence tokens in the vocabulary.
 
-            Indices can be obtained using :class:`~transformers.AlbertTokenizer`.
-            See :meth:`~transformers.PreTrainedTokenizer.encode` and
-            :meth:`~transformers.PreTrainedTokenizer.__call__` for details.
+            Indices can be obtained using [`AlbertTokenizer`]. See [`~PreTrainedTokenizer.encode`] and
+            [`~PreTrainedTokenizer.__call__`] for details.
 
-            `What are input IDs? <../glossary.html#input-ids>`__
+            [What are input IDs?](../glossary#input-ids)
 ```
 
 For optional arguments or arguments with defaults we follow the following syntax: imagine we have a function with the
@@ -208,93 +221,61 @@ then its documentation should look like this:
 
 ```
     Args:
-        x (:obj:`str`, `optional`):
+        x (`str`, *optional*):
             This argument controls ...
-        a (:obj:`float`, `optional`, defaults to 1):
+        a (`float`, *optional*, defaults to 1):
             This argument is used to ...
 ```
 
-Note that we always omit the "defaults to :obj:\`None\`" when None is the default for any argument. Also note that even
+Note that we always omit the "defaults to \`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 like so:
+Multi-line code blocks can be useful for displaying examples. They are done between two lines of three backticks as usual in Markdown:
 
+
+````
 ```
-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 stay consistent with the library.
 
 #### Writing a return block
 
-Arguments should be defined with the `Args:` prefix, followed by a line return and an indentation.
+The return block should be introduced with the `Returns:` 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 for tuple return, comprising several objects:
-
-```
-    Returns:
-        :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 for a single value return:
 
 ```
     Returns:
-        :obj:`List[int]`: A list of integers in the range [0, 1] --- 1 for a special token, 0 for a sequence token.
+        `List[int]`: A list of integers in the range [0, 1] --- 1 for a special token, 0 for a sequence token.
 ```
 
-#### Adding a new section
-
-In ReST section headers are designated as such with the help of a line of underlying characters, e.g.,:
+Here's an example for tuple return, comprising several objects:
 
 ```
-Section 1
-^^^^^^^^^^^^^^^^^^
-
-Sub-section 1
-~~~~~~~~~~~~~~~~~~
+    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 as 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).
 ```
 
-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.
+#### Adding an image
 
-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.
+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.
 
-But how do you know which characters you must use for an already existing level or when adding a new level?
-
-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 "
-```
-
-This tells you which characters have already been assigned for each level.
-
-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.
-
-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.
-
-Here is the full list of characters that can be used in this context: `= - ` : ' " ~ ^ _ * + # < >`

docs/source/_config.py

@@ -0,0 +1,9 @@
+# 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}]

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

@@ -1,16 +0,0 @@
-
-.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

@@ -1,350 +0,0 @@
-/* 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/_toctree.yml

@@ -0,0 +1,320 @@
+- sections: 
+  - local: index
+    title: 🤗 Transformers
+  - local: quicktour
+    title: Quick tour
+  - local: installation
+    title: Installation
+  - local: philosophy
+    title: Philosophy
+  - local: glossary
+    title: Glossary
+  title: Get started
+- sections:
+  - local: task_summary
+    title: Summary of the tasks
+  - local: model_summary
+    title: Summary of the models
+  - local: preprocessing
+    title: Preprocessing data
+  - local: training
+    title: Fine-tuning a pretrained model
+  - local: model_sharing
+    title: Model sharing and uploading
+  - local: tokenizer_summary
+    title: Summary of the tokenizers
+  - local: multilingual
+    title: Multi-lingual models
+  title: "Using 🤗 Transformers"
+- sections:
+  - local: examples
+    title: Examples
+  - local: troubleshooting
+    title: Troubleshooting
+  - local: custom_datasets
+    title: Fine-tuning with custom datasets
+  - local: notebooks
+    title: "🤗 Transformers Notebooks"
+  - local: sagemaker
+    title: Run training on Amazon SageMaker
+  - local: community
+    title: Community
+  - local: converting_tensorflow_models
+    title: Converting Tensorflow Checkpoints
+  - local: migration
+    title: Migrating from previous packages
+  - local: contributing
+    title: How to contribute to transformers?
+  - local: add_new_model
+    title: "How to add a model to 🤗 Transformers?"
+  - local: add_new_pipeline
+    title: "How to add a pipeline to 🤗 Transformers?"
+  - local: fast_tokenizers
+    title: "Using tokenizers from 🤗 Tokenizers"
+  - local: performance
+    title: 'Performance and Scalability: How To Fit a Bigger Model and Train It Faster'
+  - local: parallelism
+    title: Model Parallelism
+  - local: testing
+    title: Testing
+  - local: debugging
+    title: Debugging
+  - local: serialization
+    title: Exporting transformers models
+  - local: pr_checks
+    title: Checks on a Pull Request
+  title: Advanced guides
+- sections:
+  - local: bertology
+    title: BERTology
+  - local: perplexity
+    title: Perplexity of fixed-length models
+  - local: benchmarks
+    title: Benchmarks
+  title: Research
+- sections:
+  - sections:
+    - 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/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:
+    - local: model_doc/albert
+      title: ALBERT
+    - local: model_doc/auto
+      title: Auto Classes
+    - local: model_doc/bart
+      title: BART
+    - local: model_doc/barthez
+      title: BARThez
+    - local: model_doc/bartpho
+      title: BARTpho
+    - local: model_doc/beit
+      title: BEiT
+    - local: model_doc/bert
+      title: BERT
+    - local: model_doc/bertweet
+      title: Bertweet
+    - local: model_doc/bertgeneration
+      title: BertGeneration
+    - local: model_doc/bert_japanese
+      title: BertJapanese
+    - local: model_doc/bigbird
+      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/bort
+      title: BORT
+    - local: model_doc/byt5
+      title: ByT5
+    - local: model_doc/camembert
+      title: CamemBERT
+    - local: model_doc/canine
+      title: CANINE
+    - local: model_doc/clip
+      title: CLIP
+    - 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/deit
+      title: DeiT
+    - local: model_doc/detr
+      title: DETR
+    - 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/encoderdecoder
+      title: Encoder Decoder Models
+    - 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/herbert
+      title: herBERT
+    - local: model_doc/ibert
+      title: I-BERT
+    - local: model_doc/imagegpt
+      title: ImageGPT
+    - local: model_doc/layoutlm
+      title: LayoutLM
+    - local: model_doc/layoutlmv2
+      title: LayoutLMV2
+    - local: model_doc/layoutxlm
+      title: LayoutXLM
+    - local: model_doc/led
+      title: LED
+    - local: model_doc/longformer
+      title: Longformer
+    - local: model_doc/luke
+      title: LUKE
+    - local: model_doc/lxmert
+      title: LXMERT
+    - local: model_doc/marian
+      title: MarianMT
+    - local: model_doc/m2m_100
+      title: M2M100
+    - 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/mluke
+      title: mLUKE
+    - local: model_doc/mpnet
+      title: MPNet
+    - local: model_doc/mt5
+      title: MT5
+    - local: model_doc/gpt
+      title: OpenAI GPT
+    - local: model_doc/gpt2
+      title: OpenAI GPT2
+    - local: model_doc/gptj
+      title: GPT-J
+    - local: model_doc/gpt_neo
+      title: GPT Neo
+    - local: model_doc/hubert
+      title: Hubert
+    - local: model_doc/perceiver
+      title: Perceiver
+    - local: model_doc/pegasus
+      title: Pegasus
+    - local: model_doc/phobert
+      title: PhoBERT
+    - local: model_doc/prophetnet
+      title: ProphetNet
+    - local: model_doc/qdqbert
+      title: QDQBert
+    - local: model_doc/rag
+      title: RAG
+    - 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/segformer
+      title: SegFormer
+    - local: model_doc/sew
+      title: SEW
+    - local: model_doc/sew_d
+      title: SEW-D
+    - local: model_doc/speechencoderdecoder
+      title: Speech Encoder Decoder Models
+    - local: model_doc/speech_to_text
+      title: Speech2Text
+    - local: model_doc/speech_to_text_2
+      title: Speech2Text2
+    - 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/transformerxl
+      title: Transformer XL
+    - local: model_doc/trocr
+      title: TrOCR
+    - local: model_doc/unispeech
+      title: UniSpeech
+    - local: model_doc/unispeech_sat
+      title: UniSpeech-SAT
+    - local: model_doc/visionencoderdecoder
+      title: Vision Encoder Decoder Models
+    - local: model_doc/vision_text_dual_encoder
+      title: Vision Text Dual Encoder
+    - local: model_doc/vit
+      title: Vision Transformer (ViT)
+    - local: model_doc/visual_bert
+      title: VisualBERT
+    - local: model_doc/wav2vec2
+      title: Wav2Vec2
+    - local: model_doc/wav2vec2_phoneme
+      title: Wav2Vec2Phoneme
+    - local: model_doc/wavlm
+      title: WavLM
+    - local: model_doc/xlm
+      title: XLM
+    - local: model_doc/xlmprophetnet
+      title: XLM-ProphetNet
+    - local: model_doc/xlmroberta
+      title: XLM-RoBERTa
+    - local: model_doc/xlnet
+      title: XLNet
+    - local: model_doc/xlsr_wav2vec2
+      title: XLSR-Wav2Vec2
+    - local: model_doc/xls_r
+      title: XLS-R
+    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/file_utils
+      title: General Utilities
+    title: Internal Helpers
+  title: API

docs/source/add_new_model.rst

@@ -72,11 +72,11 @@ call the model to be added to 🤗 Transformers ``BrandNewBert``.
 
 Let's take a look:
 
-.. image:: ./imgs/transformers_overview.png
+.. image:: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_overview.png
 
 As you can see, we do make use of inheritance in 🤗 Transformers, but we keep the level of abstraction to an absolute
 minimum. There are never more than two levels of abstraction for any model in the library. :obj:`BrandNewBertModel`
-inherits from :obj:`BrandNewBertPreTrainedModel` which in turn inherits from :class:`~transformres.PreTrainedModel` and
+inherits from :obj:`BrandNewBertPreTrainedModel` which in turn inherits from :class:`~transformers.PreTrainedModel` and
 that's it. As a general rule, we want to make sure that a new model only depends on
 :class:`~transformers.PreTrainedModel`. The important functionalities that are automatically provided to every new
 model are :meth:`~transformers.PreTrainedModel.from_pretrained` and
@@ -271,7 +271,7 @@ logical components from one another and to have faster debugging cycles as inter
 notebooks are often easier to share with other contributors, which might be very helpful if you want to ask the Hugging
 Face team for help. If you are familiar with Jupiter notebooks, we strongly recommend you to work with them.
 
-The obvious disadvantage of Jupyther notebooks is that if you are not used to working with them you will have to spend
+The obvious disadvantage of Jupyter notebooks is that if you are not used to working with them you will have to spend
 some time adjusting to the new programming environment and that you might not be able to use your known debugging tools
 anymore, like ``ipdb``.
 
@@ -674,7 +674,7 @@ the ``input_ids`` (usually the word embeddings) are identical. And then work you
 network. At some point, you will notice a difference between the two implementations, which should point you to the bug
 in the 🤗 Transformers implementation. From our experience, a simple and efficient way is to add many print statements
 in both the original implementation and 🤗 Transformers implementation, at the same positions in the network
-respectively, and to successively remove print statements showing the same values for intermediate presentions.
+respectively, and to successively remove print statements showing the same values for intermediate presentations.
 
 When you're confident that both implementations yield the same output, verifying the outputs with
 ``torch.allclose(original_output, output, atol=1e-3)``, you're done with the most difficult part! Congratulations - the

docs/source/add_new_pipeline.rst

@@ -0,0 +1,143 @@
+.. 
+    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 add a pipeline to 🤗 Transformers?
+=======================================================================================================================
+
+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 :obj:`inputs` of the
+pipeline (:obj:`preprocess`).
+
+Then define the :obj:`outputs`. Same policy as the :obj:`inputs`. The simpler, the better. Those will be the outputs of
+:obj:`postprocess` method.
+
+Start by inheriting the base class :obj:`Pipeline`. with the 4 methods needed to implement :obj:`preprocess`,
+:obj:`_forward`, :obj:`postprocess` and :obj:`_sanitize_parameters`.
+
+
+.. code-block::
+
+    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(....)
+            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
+
+:obj:`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 :obj:`Dict`.
+
+:obj:`_forward` is the implementation detail and is not meant to be called directly. :obj:`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 :obj:`_forward` method, anything else is in the preprocess/postprocess.
+
+:obj:`postprocess` methods will take the output of :obj:`_forward` and turn it into the final output that were decided
+earlier.
+
+:obj:`_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 :obj:`_sanitize_parameters` are the 3 dicts of kwargs that will be passed directly to :obj:`preprocess`,
+:obj:`_forward` and :obj:`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 :obj:`top_k` argument in the post processing in classification tasks.
+
+.. code-block::
+
+    >>> 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 :obj:`postprocess` method with a default parameter to :obj:`5`. and edit
+:obj:`_sanitize_parameters` to allow this new parameter.
+
+
+.. code-block::
+
+
+        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:
+                preprocess_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
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Go to ``src/transformers/pipelines/__init__.py`` and fill in :obj:`SUPPORTED_TASKS` with your newly created pipeline.
+If possible it should provide a default model.
+
+Adding tests
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Create a new file ``tests/test_pipelines_MY_PIPELINE.py`` with example with the other tests.
+
+The :obj:`run_pipeline_test` function will be very generic and run on small random models on every possible
+architecture as defined by :obj:`model_mapping` and :obj:`tf_model_mapping`.
+
+This is very important to test future compatibility, meaning if someone adds a new model for
+:obj:`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 :obj:`ANY` that will simply attempt to match the
+output of the pipeline TYPE.
+
+You also *need* to implement 2 (ideally 4) tests.
+
+- :obj:`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 :obj:`test_small_model_tf`.
+- :obj:`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 :obj:`test_small_model_pt`.
+- :obj:`test_large_model_pt` (:obj:`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
+- :obj:`test_large_model_tf` (:obj:`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/benchmarks.mdx

@@ -0,0 +1,347 @@
+<!--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
+
+[[open-in-colab]]
+
+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 [here](https://github.com/huggingface/notebooks/tree/master/examples/benchmark.ipynb).
+
+## How to benchmark 🤗 Transformer models
+
+The classes [`PyTorchBenchmark`] and [`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_.
+
+<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.
+
+```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-SPLIT===
+>>> 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 `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.
+
+```bash
+python examples/pytorch/benchmarking/run_benchmark.py --help
+
+===PT-TF-SPLIT===
+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)
+====================       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-SPLIT===
+>>> 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 `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.
+
+```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-SPLIT===
+>>> 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 `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/master/examples/pytorch/benchmarking/README.md).
+- [TensorFlow Benchmarking Results](https://github.com/huggingface/transformers/tree/master/examples/tensorflow/benchmarking/README.md).

docs/source/benchmarks.rst

@@ -1,363 +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
-=======================================================================================================================
-
-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/community.md

@@ -1,4 +1,4 @@
-# Community
+# Community
 
 This page regroups resources around 🤗 Transformers developed by the community.
 
@@ -6,12 +6,13 @@ 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](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/) |
+| [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/) |
 
 ## 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) |
@@ -35,7 +36,7 @@ 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 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)|
+|[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)|
 |[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)|

docs/source/conf.py

@@ -1,218 +0,0 @@
-# -*- 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/converting_tensorflow_models.rst

@@ -13,8 +13,8 @@
 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.
+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.
 
 .. note::
     Since 2.3.0 the conversion script is now part of the transformers CLI (**transformers-cli**) available in any
@@ -26,22 +26,22 @@ 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
+<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>` \ ).
+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.
+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.
+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:
 
@@ -64,9 +64,9 @@ 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.
+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:
 
@@ -104,7 +104,7 @@ 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>`__\ )
+<https://github.com/openai/gpt-2>`__)
 
 .. code-block:: shell
 
@@ -120,7 +120,7 @@ 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>`__\ )
+<https://github.com/kimiyoung/transformer-xl/tree/master/tf#obtain-and-evaluate-pretrained-sota-models>`__)
 
 .. code-block:: shell
 

docs/source/custom_datasets.mdx

@@ -0,0 +1,681 @@
+<!--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.
+-->
+
+# How to fine-tune a model for common downstream tasks
+
+[[open-in-colab]]
+
+This guide will show you how to fine-tune 🤗 Transformers models for common downstream tasks. You will use the 🤗
+Datasets library to quickly load and preprocess the datasets, getting them ready for training with PyTorch and
+TensorFlow.
+
+Before you begin, make sure you have the 🤗 Datasets library installed. For more detailed installation instructions,
+refer to the 🤗 Datasets [installation page](https://huggingface.co/docs/datasets/installation.html). All of the
+examples in this guide will use 🤗 Datasets to load and preprocess a dataset.
+
+```bash
+pip install datasets
+```
+
+Learn how to fine-tune a model for:
+
+- [seq_imdb](#seq_imdb)
+- [tok_ner](#tok_ner)
+- [qa_squad](#qa_squad)
+
+<a id='seq_imdb'></a>
+
+## Sequence classification with IMDb reviews
+
+Sequence classification refers to the task of classifying sequences of text according to a given number of classes. In
+this example, learn how to fine-tune a model on the [IMDb dataset](https://huggingface.co/datasets/imdb) to determine
+whether a review is positive or negative.
+
+<Tip>
+
+For a more in-depth example of how to fine-tune a model for text classification, take a look at the corresponding
+[PyTorch notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/text_classification.ipynb)
+or [TensorFlow notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/text_classification-tf.ipynb).
+
+</Tip>
+
+### Load IMDb dataset
+
+The 🤗 Datasets library makes it simple to load a dataset:
+
+```python
+from datasets import load_dataset
+imdb = load_dataset("imdb")
+```
+
+This loads a `DatasetDict` object which you can index into to view an example:
+
+```python
+imdb["train"][0]
+{'label': 1,
+ 'text': 'Bromwell High is a cartoon comedy. It ran at the same time as some other programs about school life, such as "Teachers". My 35 years in the teaching profession lead me to believe that Bromwell High\'s satire is much closer to reality than is "Teachers". The scramble to survive financially, the insightful students who can see right through their pathetic teachers\' pomp, the pettiness of the whole situation, all remind me of the schools I knew and their students. When I saw the episode in which a student repeatedly tried to burn down the school, I immediately recalled ......... at .......... High. A classic line: INSPECTOR: I\'m here to sack one of your teachers. STUDENT: Welcome to Bromwell High. I expect that many adults of my age think that Bromwell High is far fetched. What a pity that it isn\'t!'
+}
+```
+
+### Preprocess
+
+The next step is to tokenize the text into a readable format by the model. It is important to load the same tokenizer a
+model was trained with to ensure appropriately tokenized words. Load the DistilBERT tokenizer with the
+[`AutoTokenizer`] because we will eventually train a classifier using a pretrained [DistilBERT](https://huggingface.co/distilbert-base-uncased) model:
+
+```python
+from transformers import AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
+```
+
+Now that you have instantiated a tokenizer, create a function that will tokenize the text. You should also truncate
+longer sequences in the text to be no longer than the model's maximum input length:
+
+```python
+def preprocess_function(examples):
+    return tokenizer(examples["text"], truncation=True)
+```
+
+Use 🤗 Datasets `map` function to apply the preprocessing function to the entire dataset. You can also set
+`batched=True` to apply the preprocessing function to multiple elements of the dataset at once for faster
+preprocessing:
+
+```python
+tokenized_imdb = imdb.map(preprocess_function, batched=True)
+```
+
+Lastly, pad your text so they are a uniform length. While it is possible to pad your text in the `tokenizer` function
+by setting `padding=True`, it is more efficient to only pad the text to the length of the longest element in its
+batch. This is known as **dynamic padding**. You can do this with the `DataCollatorWithPadding` function:
+
+```python
+from transformers import DataCollatorWithPadding
+data_collator = DataCollatorWithPadding(tokenizer=tokenizer)
+```
+
+### Fine-tune with the Trainer API
+
+Now load your model with the [`AutoModelForSequenceClassification`] class along with the number of expected labels:
+
+```python
+from transformers import AutoModelForSequenceClassification
+model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased", num_labels=2)
+```
+
+At this point, only three steps remain:
+
+1. Define your training hyperparameters in [`TrainingArguments`].
+2. Pass the training arguments to a [`Trainer`] along with the model, dataset, tokenizer, and data collator.
+3. Call [`Trainer.train()`] to fine-tune your model.
+
+```python
+from transformers import TrainingArguments, Trainer
+
+training_args = TrainingArguments(
+    output_dir='./results',
+    learning_rate=2e-5,
+    per_device_train_batch_size=16,
+    per_device_eval_batch_size=16,
+    num_train_epochs=5,
+    weight_decay=0.01,
+)
+
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_imdb["train"],
+    eval_dataset=tokenized_imdb["test"],
+    tokenizer=tokenizer,
+    data_collator=data_collator,
+)
+
+trainer.train()
+```
+
+### Fine-tune with TensorFlow
+
+Fine-tuning with TensorFlow is just as easy, with only a few differences.
+
+Start by batching the processed examples together with dynamic padding using the [`DataCollatorWithPadding`] function.
+Make sure you set `return_tensors="tf"` to return `tf.Tensor` outputs instead of PyTorch tensors!
+
+```python
+from transformers import DataCollatorWithPadding
+data_collator = DataCollatorWithPadding(tokenizer, return_tensors="tf")
+```
+
+Next, convert your datasets to the `tf.data.Dataset` format with `to_tf_dataset`. Specify inputs and labels in the
+`columns` argument:
+
+```python
+tf_train_dataset = tokenized_imdb["train"].to_tf_dataset(
+    columns=['attention_mask', 'input_ids', 'label'],
+    shuffle=True,
+    batch_size=16,
+    collate_fn=data_collator,
+)
+
+tf_validation_dataset = tokenized_imdb["train"].to_tf_dataset(
+    columns=['attention_mask', 'input_ids', 'label'],
+    shuffle=False,
+    batch_size=16,
+    collate_fn=data_collator,
+)
+```
+
+Set up an optimizer function, learning rate schedule, and some training hyperparameters:
+
+```python
+from transformers import create_optimizer
+import tensorflow as tf
+
+batch_size = 16
+num_epochs = 5
+batches_per_epoch = len(tokenized_imdb["train"]) // batch_size
+total_train_steps = int(batches_per_epoch * num_epochs)
+optimizer, schedule = create_optimizer(
+    init_lr=2e-5, 
+    num_warmup_steps=0, 
+    num_train_steps=total_train_steps
+)
+```
+
+Load your model with the [`TFAutoModelForSequenceClassification`] class along with the number of expected labels:
+
+```python
+from transformers import TFAutoModelForSequenceClassification
+model = TFAutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased", num_labels=2)
+```
+
+Compile the model:
+
+```python
+import tensorflow as tf
+model.compile(optimizer=optimizer)
+```
+
+Finally, fine-tune the model by calling `model.fit`:
+
+```python
+model.fit(
+    tf_train_set,
+    validation_data=tf_validation_set,
+    epochs=num_train_epochs,
+)
+```
+
+<a id='tok_ner'></a>
+
+## Token classification with WNUT emerging entities
+
+Token classification refers to the task of classifying individual tokens in a sentence. One of the most common token
+classification tasks is Named Entity Recognition (NER). NER attempts to find a label for each entity in a sentence,
+such as a person, location, or organization. In this example, learn how to fine-tune a model on the [WNUT 17](https://huggingface.co/datasets/wnut_17) dataset to detect new entities.
+
+<Tip>
+
+For a more in-depth example of how to fine-tune a model for token classification, take a look at the corresponding
+[PyTorch notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/token_classification.ipynb)
+or [TensorFlow notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/token_classification-tf.ipynb).
+
+</Tip>
+
+### Load WNUT 17 dataset
+
+Load the WNUT 17 dataset from the 🤗 Datasets library:
+
+```python
+from datasets import load_dataset
+wnut = load_dataset("wnut_17")
+```
+
+A quick look at the dataset shows the labels associated with each word in the sentence:
+
+```python
+wnut["train"][0]
+{'id': '0',
+ 'ner_tags': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7, 8, 8, 0, 7, 0, 0, 0, 0, 0, 0, 0, 0],
+ 'tokens': ['@paulwalk', 'It', "'s", 'the', 'view', 'from', 'where', 'I', "'m", 'living', 'for', 'two', 'weeks', '.', 'Empire', 'State', 'Building', '=', 'ESB', '.', 'Pretty', 'bad', 'storm', 'here', 'last', 'evening', '.']
+}
+```
+
+View the specific NER tags by:
+
+```python
+label_list = wnut["train"].features[f"ner_tags"].feature.names
+label_list
+['O',
+ 'B-corporation',
+ 'I-corporation',
+ 'B-creative-work',
+ 'I-creative-work',
+ 'B-group',
+ 'I-group',
+ 'B-location',
+ 'I-location',
+ 'B-person',
+ 'I-person',
+ 'B-product',
+ 'I-product'
+]
+```
+
+A letter prefixes each NER tag which can mean:
+
+- `B-` indicates the beginning of an entity.
+- `I-` indicates a token is contained inside the same entity (e.g., the `State` token is a part of an entity like
+  `Empire State Building`).
+- `0` indicates the token doesn't correspond to any entity.
+
+### Preprocess
+
+Now you need to tokenize the text. Load the DistilBERT tokenizer with an [`AutoTokenizer`]:
+
+```python
+from transformers import AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
+```
+
+Since the input has already been split into words, set `is_split_into_words=True` to tokenize the words into
+subwords:
+
+```python
+tokenized_input = tokenizer(example["tokens"], is_split_into_words=True)
+tokens = tokenizer.convert_ids_to_tokens(tokenized_input["input_ids"])
+tokens
+['[CLS]', '@', 'paul', '##walk', 'it', "'", 's', 'the', 'view', 'from', 'where', 'i', "'", 'm', 'living', 'for', 'two', 'weeks', '.', 'empire', 'state', 'building', '=', 'es', '##b', '.', 'pretty', 'bad', 'storm', 'here', 'last', 'evening', '.', '[SEP]']
+```
+
+The addition of the special tokens `[CLS]` and `[SEP]` and subword tokenization creates a mismatch between the
+input and labels. Realign the labels and tokens by:
+
+1. Mapping all tokens to their corresponding word with the `word_ids` method.
+2. Assigning the label `-100` to the special tokens `[CLS]` and ``[SEP]``` so the PyTorch loss function ignores
+   them.
+3. Only labeling the first token of a given word. Assign `-100` to the other subtokens from the same word.
+
+Here is how you can create a function that will realign the labels and tokens:
+
+```python
+def tokenize_and_align_labels(examples):
+    tokenized_inputs = tokenizer(examples["tokens"], truncation=True, is_split_into_words=True)
+
+    labels = []
+    for i, label in enumerate(examples[f"ner_tags"]):
+        word_ids = tokenized_inputs.word_ids(batch_index=i)  # Map tokens to their respective word.
+        previous_word_idx = None
+        label_ids = []
+        for word_idx in word_ids:                            # Set the special tokens to -100.
+            if word_idx is None:
+                label_ids.append(-100)
+            elif word_idx != previous_word_idx:              # Only label the first token of a given word.
+                label_ids.append(label[word_idx])
+
+        labels.append(label_ids)
+
+    tokenized_inputs["labels"] = labels
+    return tokenized_inputs
+```
+
+Now tokenize and align the labels over the entire dataset with 🤗 Datasets `map` function:
+
+```python
+tokenized_wnut = wnut.map(tokenize_and_align_labels, batched=True)
+```
+
+Finally, pad your text and labels, so they are a uniform length:
+
+```python
+from transformers import DataCollatorForTokenClassification
+data_collator = DataCollatorForTokenClassification(tokenizer)
+```
+
+### Fine-tune with the Trainer API
+
+Load your model with the [`AutoModelForTokenClassification`] class along with the number of expected labels:
+
+```python
+from transformers import AutoModelForTokenClassification, TrainingArguments, Trainer
+model = AutoModelForTokenClassification.from_pretrained("distilbert-base-uncased", num_labels=len(label_list))
+```
+
+Gather your training arguments in [`TrainingArguments`]:
+
+```python
+training_args = TrainingArguments(
+    output_dir='./results',
+    evaluation_strategy="epoch",
+    learning_rate=2e-5,
+    per_device_train_batch_size=16,
+    per_device_eval_batch_size=16,
+    num_train_epochs=3,
+    weight_decay=0.01,
+)
+```
+
+Collect your model, training arguments, dataset, data collator, and tokenizer in [`Trainer`]:
+
+```python
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_wnut["train"],
+    eval_dataset=tokenized_wnut["test"],
+    data_collator=data_collator,
+    tokenizer=tokenizer,
+)
+```
+
+Fine-tune your model:
+
+```python
+trainer.train()
+```
+
+### Fine-tune with TensorFlow
+
+Batch your examples together and pad your text and labels, so they are a uniform length:
+
+```python
+from transformers import DataCollatorForTokenClassification
+data_collator = DataCollatorForTokenClassification(tokenizer, return_tensors="tf")
+```
+
+Convert your datasets to the `tf.data.Dataset` format with `to_tf_dataset`:
+
+```python
+tf_train_set = tokenized_wnut["train"].to_tf_dataset(
+    columns=["attention_mask", "input_ids", "labels"],
+    shuffle=True,
+    batch_size=16,
+    collate_fn=data_collator,
+)
+
+tf_validation_set = tokenized_wnut["validation"].to_tf_dataset(
+    columns=["attention_mask", "input_ids", "labels"],
+    shuffle=False,
+    batch_size=16,
+    collate_fn=data_collator,
+)
+```
+
+Load the model with the [`TFAutoModelForTokenClassification`] class along with the number of expected labels:
+
+```python
+from transformers import TFAutoModelForTokenClassification
+model = TFAutoModelForTokenClassification.from_pretrained("distilbert-base-uncased", num_labels=len(label_list))
+```
+
+Set up an optimizer function, learning rate schedule, and some training hyperparameters:
+
+```python
+from transformers import create_optimizer
+
+batch_size = 16
+num_train_epochs = 3
+num_train_steps = (len(tokenized_datasets["train"]) // batch_size) * num_train_epochs
+optimizer, lr_schedule = create_optimizer(
+    init_lr=2e-5,
+    num_train_steps=num_train_steps,
+    weight_decay_rate=0.01,
+    num_warmup_steps=0,
+)
+```
+
+Compile the model:
+
+```python
+import tensorflow as tf
+model.compile(optimizer=optimizer)
+```
+
+Call `model.fit` to fine-tune your model:
+
+```python
+model.fit(
+    tf_train_set,
+    validation_data=tf_validation_set,
+    epochs=num_train_epochs,
+)
+```
+
+<a id='qa_squad'></a>
+
+## Question Answering with SQuAD
+
+There are many types of question answering (QA) tasks. Extractive QA focuses on identifying the answer from the text
+given a question. In this example, learn how to fine-tune a model on the [SQuAD](https://huggingface.co/datasets/squad) dataset.
+
+<Tip>
+
+For a more in-depth example of how to fine-tune a model for question answering, take a look at the corresponding
+[PyTorch notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/question_answering.ipynb)
+or [TensorFlow notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/question_answering-tf.ipynb).
+
+</Tip>
+
+### Load SQuAD dataset
+
+Load the SQuAD dataset from the 🤗 Datasets library:
+
+```python
+from datasets import load_dataset
+squad = load_dataset("squad")
+```
+
+Take a look at an example from the dataset:
+
+```python
+squad["train"][0]
+{'answers': {'answer_start': [515], 'text': ['Saint Bernadette Soubirous']},
+ 'context': 'Architecturally, the school has a Catholic character. Atop the Main Building\'s gold dome is a golden statue of the Virgin Mary. Immediately in front of the Main Building and facing it, is a copper statue of Christ with arms upraised with the legend "Venite Ad Me Omnes". Next to the Main Building is the Basilica of the Sacred Heart. Immediately behind the basilica is the Grotto, a Marian place of prayer and reflection. It is a replica of the grotto at Lourdes, France where the Virgin Mary reputedly appeared to Saint Bernadette Soubirous in 1858. At the end of the main drive (and in a direct line that connects through 3 statues and the Gold Dome), is a simple, modern stone statue of Mary.',
+ 'id': '5733be284776f41900661182',
+ 'question': 'To whom did the Virgin Mary allegedly appear in 1858 in Lourdes France?',
+ 'title': 'University_of_Notre_Dame'
+}
+```
+
+### Preprocess
+
+Load the DistilBERT tokenizer with an [`AutoTokenizer`]:
+
+```python
+from transformers import AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
+```
+
+There are a few things to be aware of when preprocessing text for question answering:
+
+1. Some examples in a dataset may have a very long `context` that exceeds the maximum input length of the model. You
+   can deal with this by truncating the `context` and set `truncation="only_second"`.
+2. Next, you need to map the start and end positions of the answer to the original context. Set
+   `return_offset_mapping=True` to handle this.
+3. With the mapping in hand, you can find the start and end tokens of the answer. Use the `sequence_ids` method to
+   find which part of the offset corresponds to the question, and which part of the offset corresponds to the context.
+
+Assemble everything in a preprocessing function as shown below:
+
+```python
+def preprocess_function(examples):
+    questions = [q.strip() for q in examples["question"]]
+    inputs = tokenizer(
+        questions,
+        examples["context"],
+        max_length=384,
+        truncation="only_second",
+        return_offsets_mapping=True,
+        padding="max_length",
+    )
+
+    offset_mapping = inputs.pop("offset_mapping")
+    answers = examples["answers"]
+    start_positions = []
+    end_positions = []
+
+    for i, offset in enumerate(offset_mapping):
+        answer = answers[i]
+        start_char = answer["answer_start"][0]
+        end_char = answer["answer_start"][0] + len(answer["text"][0])
+        sequence_ids = inputs.sequence_ids(i)
+
+        # Find the start and end of the context
+        idx = 0
+        while sequence_ids[idx] != 1:
+            idx += 1
+        context_start = idx
+        while sequence_ids[idx] == 1:
+            idx += 1
+        context_end = idx - 1
+
+        # If the answer is not fully inside the context, label it (0, 0)
+        if offset[context_start][0] > end_char or offset[context_end][1] < start_char:
+            start_positions.append(0)
+            end_positions.append(0)
+        else:
+            # Otherwise it's the start and end token positions
+            idx = context_start
+            while idx <= context_end and offset[idx][0] <= start_char:
+                idx += 1
+            start_positions.append(idx - 1)
+
+            idx = context_end
+            while idx >= context_start and offset[idx][1] >= end_char:
+                idx -= 1
+            end_positions.append(idx + 1)
+
+    inputs["start_positions"] = start_positions
+    inputs["end_positions"] = end_positions
+    return inputs
+```
+
+Apply the preprocessing function over the entire dataset with 🤗 Datasets `map` function:
+
+```python
+tokenized_squad = squad.map(preprocess_function, batched=True, remove_columns=squad["train"].column_names)
+```
+
+Batch the processed examples together:
+
+```python
+from transformers import default_data_collator
+data_collator = default_data_collator
+```
+
+### Fine-tune with the Trainer API
+
+Load your model with the [`AutoModelForQuestionAnswering`] class:
+
+```python
+from transformers import AutoModelForQuestionAnswering, TrainingArguments, Trainer
+model = AutoModelForQuestionAnswering.from_pretrained("distilbert-base-uncased")
+```
+
+Gather your training arguments in [`TrainingArguments`]:
+
+```python
+training_args = TrainingArguments(
+    output_dir='./results',
+    evaluation_strategy="epoch",
+    learning_rate=2e-5,
+    per_device_train_batch_size=16,
+    per_device_eval_batch_size=16,
+    num_train_epochs=3,
+    weight_decay=0.01,
+)
+```
+
+Collect your model, training arguments, dataset, data collator, and tokenizer in [`Trainer`]:
+
+```python
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_squad["train"],
+    eval_dataset=tokenized_squad["validation"],
+    data_collator=data_collator,
+    tokenizer=tokenizer,
+)
+```
+
+Fine-tune your model:
+
+```python
+trainer.train()
+```
+
+### Fine-tune with TensorFlow
+
+Batch the processed examples together with a TensorFlow default data collator:
+
+```python
+from transformers.data.data_collator import tf_default_collator
+data_collator = tf_default_collator
+```
+
+Convert your datasets to the `tf.data.Dataset` format with the `to_tf_dataset` function:
+
+```python
+tf_train_set = tokenized_squad["train"].to_tf_dataset(
+    columns=["attention_mask", "input_ids", "start_positions", "end_positions"],
+    dummy_labels=True,
+    shuffle=True,
+    batch_size=16,
+    collate_fn=data_collator,
+)
+
+tf_validation_set = tokenized_squad["validation"].to_tf_dataset(
+    columns=["attention_mask", "input_ids", "start_positions", "end_positions"],
+    dummy_labels=True,
+    shuffle=False,
+    batch_size=16,
+    collate_fn=data_collator,
+)
+```
+
+Set up an optimizer function, learning rate schedule, and some training hyperparameters:
+
+```python
+from transformers import create_optimizer
+
+batch_size = 16
+num_epochs = 2
+total_train_steps = (len(tokenized_squad["train"]) // batch_size) * num_epochs
+optimizer, schedule = create_optimizer(
+    init_lr=2e-5, 
+    num_warmup_steps=0, 
+    num_train_steps=total_train_steps,
+)
+```
+
+Load your model with the [`TFAutoModelForQuestionAnswering`] class:
+
+```python
+from transformers import TFAutoModelForQuestionAnswering
+model = TFAutoModelForQuestionAnswering("distilbert-base-uncased")
+```
+
+Compile the model:
+
+```python
+import tensorflow as tf
+model.compile(optimizer=optimizer)
+```
+
+Call `model.fit` to fine-tune the model:
+
+```python
+model.fit(
+    tf_train_set,
+    validation_data=tf_validation_set,
+    epochs=num_train_epochs,
+)
+```

docs/source/custom_datasets.rst

@@ -1,729 +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.
-
-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/debugging.mdx

@@ -0,0 +1,299 @@
+<!--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
+
+<Tip>
+
+This feature is currently available for PyTorch-only.
+
+</Tip>
+
+<Tip>
+
+For multi-GPU training it requires DDP (`torch.distributed.launch`).
+
+</Tip>
+
+<Tip>
+
+This feature can be used with any `nn.Module`-based model.
+
+</Tip>
+
+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 [`Trainer`], you just need to add:
+
+```bash
+--debug underflow_overflow
+```
+
+to the normal command line arguments, or pass `debug="underflow_overflow"` when creating the
+[`TrainingArguments`] object.
+
+If you're using your own training loop or another Trainer you can accomplish the same with:
+
+```python
+from .debug_utils import DebugUnderflowOverflow
+debug_overflow = DebugUnderflowOverflow(model)
+```
+
+[`~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):
+
+```
+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:
+
+```
+                  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:
+
+```
+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
+overflow (`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`:
+
+```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:
+
+```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:
+
+```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.:
+
+```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:
+
+```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:
+
+```
+                  *** 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:
+
+```python
+debug_overflow = DebugUnderflowOverflow(model, trace_batch_nums=[1,3], abort_after_batch_num=3)
+```

docs/source/debugging.rst

@@ -1,295 +0,0 @@
-..
-    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/index.mdx

@@ -0,0 +1,275 @@
+<!--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
+
+State-of-the-art Machine Learning for Jax, Pytorch and TensorFlow
+
+🤗 Transformers (formerly known as _pytorch-transformers_ and _pytorch-pretrained-bert_) provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio.
+
+These models can 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 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.
+
+🤗 Transformers is backed by the three most popular deep learning libraries — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — with a seamless integration between them. It's straightforward to train your models with one before loading them for inference with the other.
+
+This is the documentation of our repository [transformers](https://github.com/huggingface/transformers). You can
+also follow our [online course](https://huggingface.co/course) that teaches how to use this library, as well as the
+other libraries developed by Hugging Face and the Hub.
+
+## 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://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>
+
+## Features
+
+1. Easy-to-use state-of-the-art models:
+    - High performance on natural language understanding & generation, computer vision, and audio 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.
+
+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 20,000 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.
+    - Seamlessly pick the right framework for training, evaluation and production.
+
+1. Easily customize a model or an example to your needs:
+    - We provide examples for each architecture to reproduce the results published by its original authors.
+    - Model internals are exposed as consistently as possible.
+    - Model files can be used independently of the library for quick experiments.
+
+[All the model checkpoints](https://huggingface.co/models) 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: <img src="https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen">
+
+## Contents
+
+The documentation is organized in five parts:
+
+- **GET STARTED** contains a quick tour, the installation instructions and some useful information about our philosophy
+  and a glossary.
+- **USING 🤗 TRANSFORMERS** contains general tutorials on how to use the library.
+- **ADVANCED GUIDES** contains more advanced guides that are more specific to a given script or part of the library.
+- **RESEARCH** focuses on tutorials that have less to do with how to use the library but more about general research in
+  transformers model
+- **API** contains the documentation of each public class and function, grouped in:
+
+  - **MAIN CLASSES** for the main classes exposing the important APIs of the library.
+  - **MODELS** for the classes and functions related to each model implemented in the library.
+  - **INTERNAL HELPERS** for the classes and functions we use internally.
+
+The library currently contains Jax, PyTorch and Tensorflow implementations, pretrained model weights, usage scripts and
+conversion utilities for the following models.
+
+### Supported models
+
+<!--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/pdf/1910.13461.pdf) 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. **[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. **[BERT For Sequence Generation](model_doc/bertgeneration)** (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](model_doc/bigbird)** (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](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. **[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. **[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. **[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. **[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. **[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. **[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/master/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/research_projects/distillation) and a German version of DistilBERT.
+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. **[EncoderDecoder](model_doc/encoderdecoder)** (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. **[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. **[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. **[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. **[GPT](model_doc/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-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. **[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. **[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. **[LayoutXLM](model_doc/layoutlmv2)** (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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[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. **[RemBERT](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. **[RoBERTa](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](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](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. **[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. **[Transformer-XL](model_doc/transformerxl)** (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. **[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. **[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. **[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. **[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. **[Wav2Vec2Phoneme](https://huggingface.co/docs/master/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. **[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/xlmprophetnet)** (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/xlmroberta)** (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](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. **[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. **[XLS-R](https://huggingface.co/docs/master/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.
+
+
+### Supported frameworks
+
+The table below represents the current support in the library for each of those models, whether they have a Python
+tokenizer (called "slow"). A "fast" tokenizer backed by the 🤗 Tokenizers library, whether they have support in Jax (via
+Flax), PyTorch, and/or TensorFlow.
+
+<!--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           |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
+|       BigBirdPegasus        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|         Blenderbot          |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|       BlenderbotSmall       |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|          CamemBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|           Canine            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|            CLIP             |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
+|          ConvBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|           DeBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|         DeBERTa-v2          |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|            DeiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|            DETR             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|         DistilBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|             DPR             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
+| FairSeq Machine-Translation |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|          FlauBERT           |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|            FNet             |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|     Funnel Transformer      |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|           GPT Neo           |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
+|            GPT-J            |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
+|           Hubert            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
+|           I-BERT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|          ImageGPT           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|         LayoutLMv2          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|             LED             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|         Longformer          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|            LUKE             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           LXMERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|           M2M100            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           Marian            |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
+|            mBART            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|        MegatronBert         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|         MobileBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|            MPNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|             mT5             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|         OpenAI GPT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|           Pegasus           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|          Perceiver          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|         ProphetNet          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           QDQBert           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|             RAG             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|          Reformer           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|           RemBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|          RetriBERT          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|          RoFormer           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|          SegFormer          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|             SEW             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|            SEW-D            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|   Speech Encoder decoder    |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|         Speech2Text         |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|        Speech2Text2         |       ✅       |       ❌       |       ❌        |         ❌         |      ❌      |
+|          Splinter           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|         SqueezeBERT         |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|            TAPAS            |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|            TrOCR            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|          UniSpeech          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|        UniSpeechSat         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|   Vision Encoder decoder    |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
+|    VisionTextDualEncoder    |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
+|         VisualBert          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|             ViT             |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
+|          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
+|            WavLM            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+|         XLM-RoBERTa         |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|        XLMProphetNet        |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|            XLNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+
+<!-- End table-->

docs/source/index.rst

@@ -1,577 +0,0 @@
-Transformers
-=======================================================================================================================
-
-State-of-the-art Natural Language Processing for Jax, Pytorch and TensorFlow
-
-🤗 Transformers (formerly known as `pytorch-transformers` and `pytorch-pretrained-bert`) provides general-purpose
-architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet...) for Natural Language Understanding (NLU) and Natural
-Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between Jax,
-PyTorch and TensorFlow.
-
-This is the documentation of our repository `transformers <https://github.com/huggingface/transformers>`__. You can
-also follow our `online course <https://huggingface.co/course>`__ that teaches how to use this library, as well as the
-other libraries developed by Hugging Face and the Hub.
-
-If you are looking for custom support from the Hugging Face team
------------------------------------------------------------------------------------------------------------------------
-
-.. raw:: html
-
-    <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>
-
-Features
------------------------------------------------------------------------------------------------------------------------
-
-- High performance on NLU and NLG tasks
-- Low barrier to entry for educators and practitioners
-
-State-of-the-art NLP for everyone:
-
-- Deep learning researchers
-- Hands-on practitioners
-- AI/ML/NLP teachers and educators
-
-..
-    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.
-
-Lower compute costs, smaller carbon footprint:
-
-- Researchers can share trained models instead of always retraining
-- Practitioners can reduce compute time and production costs
-- 8 architectures with over 30 pretrained models, some in more than 100 languages
-
-Choose the right framework for every part of a model's lifetime:
-
-- Train state-of-the-art models in 3 lines of code
-- Deep interoperability between Jax, Pytorch and TensorFlow models
-- Move a single model between Jax/PyTorch/TensorFlow frameworks at will
-- Seamlessly pick the right framework for training, evaluation, production
-
-The support for Jax is still experimental (with a few models right now), expect to see it grow in the coming months!
-
-`All the model checkpoints <https://huggingface.co/models>`__ 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: |checkpoints|
-
-.. |checkpoints| image:: https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen
-
-Contents
------------------------------------------------------------------------------------------------------------------------
-
-The documentation is organized in five parts:
-
-- **GET STARTED** contains a quick tour, the installation instructions and some useful information about our philosophy
-  and a glossary.
-- **USING 🤗 TRANSFORMERS** contains general tutorials on how to use the library.
-- **ADVANCED GUIDES** contains more advanced guides that are more specific to a given script or part of the library.
-- **RESEARCH** focuses on tutorials that have less to do with how to use the library but more about general research in
-  transformers model
-- The three last section contain the documentation of each public class and function, grouped in:
-
-    - **MAIN CLASSES** for the main classes exposing the important APIs of the library.
-    - **MODELS** for the classes and functions related to each model implemented in the library.
-    - **INTERNAL HELPERS** for the classes and functions we use internally.
-
-The library currently contains Jax, PyTorch and Tensorflow implementations, pretrained model weights, usage scripts and
-conversion utilities for the following models.
-
-Supported models
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-..
-    This list is updated automatically from the README with `make fix-copies`. Do not update manually!
-
-1. :doc:`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.
-2. :doc:`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/pdf/1910.13461.pdf>`__ by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman
-   Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-3. :doc:`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.
-4. :doc:`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.
-5. :doc:`BERT For Sequence Generation <model_doc/bertgeneration>` (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.
-6. :doc:`BigBird-RoBERTa <model_doc/bigbird>` (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.
-7. :doc:`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.
-8. :doc:`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.
-9. :doc:`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.
-10. :doc:`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.
-11. :doc:`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.
-12. :doc:`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.
-13. :doc:`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.
-14. :doc:`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.
-15. :doc:`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.
-16. :doc:`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.
-17. :doc:`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.
-18. :doc:`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.
-19. :doc:`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.
-20. :doc:`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.
-21. :doc:`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.
-22. :doc:`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/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.
-23. :doc:`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.
-24. :doc:`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.
-25. :doc:`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.
-26. :doc:`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.
-27. :doc:`GPT <model_doc/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.
-28. :doc:`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**.
-29. :doc:`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.
-30. :doc:`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.
-31. :doc:`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
-32. :doc:`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.
-33. :doc:`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.
-34. :doc:`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.
-35. :doc:`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.
-36. :doc:`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.
-37. :doc:`M2M100 <model_doc/m2m_100>` (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.
-38. :doc:`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.
-39. :doc:`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.
-40. :doc:`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.
-41. :doc:`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.
-42. :doc:`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.
-43. :doc:`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.
-44. :doc:`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.
-45. :doc:`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.
-46. :doc:`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.
-47. :doc:`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.
-48. :doc:`RoBERTa <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.
-49. :doc:`RoFormer <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.
-50. :doc:`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.
-51. :doc:`SqueezeBert <model_doc/squeezebert>` 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.
-52. :doc:`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.
-53. :doc:`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.
-54. :doc:`Transformer-XL <model_doc/transformerxl>` (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.
-55. :doc:`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.
-56. :doc:`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.
-57. :doc:`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.
-58. :doc:`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.
-59. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (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.
-60. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (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.
-61. :doc:`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.
-62. :doc:`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.
-
-
-Supported frameworks
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The table below represents the current support in the library for each of those models, whether they have a Python
-tokenizer (called "slow"). A "fast" tokenizer backed by the 🤗 Tokenizers library, whether they have support in Jax (via
-Flax), PyTorch, and/or TensorFlow.
-
-..
-    This table is updated automatically from the auto modules with `make fix-copies`. Do not update manually!
-
-.. rst-class:: center-aligned-table
-
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            Model            | Tokenizer slow | Tokenizer fast | PyTorch support | TensorFlow support | Flax Support |
-+=============================+================+================+=================+====================+==============+
-|           ALBERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            BART             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            BERT             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|       Bert Generation       |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           BigBird           |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|       BigBirdPegasus        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         Blenderbot          |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|       BlenderbotSmall       |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            CLIP             |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          CamemBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          ConvBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            DETR             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             DPR             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           DeBERTa           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         DeBERTa-v2          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            DeiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         DistilBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-| FairSeq Machine-Translation |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          FlauBERT           |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|     Funnel Transformer      |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           GPT Neo           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           Hubert            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           I-BERT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             LED             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            LUKE             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           LXMERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         Longformer          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           M2M100            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            MPNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           Marian            |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|        MegatronBert         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         MobileBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         OpenAI GPT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           Pegasus           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         ProphetNet          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             RAG             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          Reformer           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          RetriBERT          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          RoFormer           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         Speech2Text         |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         SqueezeBERT         |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            TAPAS            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             ViT             |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         VisualBert          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         XLM-RoBERTa         |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|        XLMProphetNet        |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            XLNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            mBART            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             mT5             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Get started
-
-    quicktour
-    installation
-    philosophy
-    glossary
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Using 🤗 Transformers
-
-    task_summary
-    model_summary
-    preprocessing
-    training
-    model_sharing
-    tokenizer_summary
-    multilingual
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Advanced guides
-
-    pretrained_models
-    examples
-    troubleshooting
-    custom_datasets
-    notebooks
-    sagemaker
-    community
-    converting_tensorflow_models
-    migration
-    contributing
-    add_new_model
-    fast_tokenizers
-    performance
-    testing
-    debugging
-    serialization
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Research
-
-    bertology
-    perplexity
-    benchmarks
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Main Classes
-
-    main_classes/callback
-    main_classes/configuration
-    main_classes/data_collator
-    main_classes/logging
-    main_classes/model
-    main_classes/optimizer_schedules
-    main_classes/output
-    main_classes/pipelines
-    main_classes/processors
-    main_classes/tokenizer
-    main_classes/trainer
-    main_classes/deepspeed
-    main_classes/feature_extractor
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Models
-
-    model_doc/albert
-    model_doc/auto
-    model_doc/bart
-    model_doc/barthez
-    model_doc/bert
-    model_doc/bertweet
-    model_doc/bertgeneration
-    model_doc/bert_japanese
-    model_doc/bigbird
-    model_doc/bigbird_pegasus
-    model_doc/blenderbot
-    model_doc/blenderbot_small
-    model_doc/bort
-    model_doc/byt5
-    model_doc/camembert
-    model_doc/clip
-    model_doc/convbert
-    model_doc/cpm
-    model_doc/ctrl
-    model_doc/deberta
-    model_doc/deberta_v2
-    model_doc/deit
-    model_doc/detr
-    model_doc/dialogpt
-    model_doc/distilbert
-    model_doc/dpr
-    model_doc/electra
-    model_doc/encoderdecoder
-    model_doc/flaubert
-    model_doc/fsmt
-    model_doc/funnel
-    model_doc/herbert
-    model_doc/ibert
-    model_doc/layoutlm
-    model_doc/led
-    model_doc/longformer
-    model_doc/luke
-    model_doc/lxmert
-    model_doc/marian
-    model_doc/m2m_100
-    model_doc/mbart
-    model_doc/megatron_bert
-    model_doc/megatron_gpt2
-    model_doc/mobilebert
-    model_doc/mpnet
-    model_doc/mt5
-    model_doc/gpt
-    model_doc/gpt2
-    model_doc/gpt_neo
-    model_doc/hubert
-    model_doc/pegasus
-    model_doc/phobert
-    model_doc/prophetnet
-    model_doc/rag
-    model_doc/reformer
-    model_doc/retribert
-    model_doc/roberta
-    model_doc/roformer
-    model_doc/speech_to_text
-    model_doc/squeezebert
-    model_doc/t5
-    model_doc/tapas
-    model_doc/transformerxl
-    model_doc/vit
-    model_doc/visual_bert
-    model_doc/wav2vec2
-    model_doc/xlm
-    model_doc/xlmprophetnet
-    model_doc/xlmroberta
-    model_doc/xlnet
-    model_doc/xlsr_wav2vec2
-
-.. toctree::
-    :maxdepth: 2
-    :caption: Internal Helpers
-
-    internal/modeling_utils
-    internal/pipelines_utils
-    internal/tokenization_utils
-    internal/trainer_utils
-    internal/generation_utils
-    internal/file_utils

docs/source/installation.md

@@ -79,9 +79,9 @@ Here is how to quickly install `transformers` from source:
 pip install git+https://github.com/huggingface/transformers
 ```
 
-Note that this will install not the latest released version, but the bleeding edge `master` version, which you may want to use in case a bug has been fixed since the last official release and a new release hasn't  been yet rolled out.
+Note that this will install not the latest released version, but the bleeding edge `master` version, which you may want to use in case a bug has been fixed since the last official release and a new release hasn't been yet rolled out.
 
-While we strive to keep `master` operational at all times, if you notice some issues, they usually get fixed within a few hours or a day and and you're more than welcome to help us detect any problems by opening an [Issue](https://github.com/huggingface/transformers/issues) and this way, things will get fixed even sooner.
+While we strive to keep `master` operational at all times, if you notice some issues, they usually get fixed within a few hours or a day and you're more than welcome to help us detect any problems by opening an [Issue](https://github.com/huggingface/transformers/issues) and this way, things will get fixed even sooner.
 
 Again, you can run:
 

docs/source/internal/file_utils.rst

@@ -51,4 +51,4 @@ Special Properties
 Other Utilities
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-.. autoclass:: transformers.file_utils._BaseLazyModule
+.. autoclass:: transformers.file_utils._LazyModule

docs/source/internal/modeling_utils.rst

@@ -63,7 +63,6 @@ TensorFlow custom layers
     :members: call
 
 .. autoclass:: transformers.modeling_tf_utils.TFSequenceSummary
-    :members: call
 
 
 TensorFlow loss functions

docs/source/main_classes/configuration.rst

@@ -17,9 +17,15 @@ The base class :class:`~transformers.PretrainedConfig` implements the common met
 either from a local file or directory, or from a pretrained model configuration provided by the library (downloaded
 from HuggingFace's AWS S3 repository).
 
+Each derived config class implements model specific attributes. Common attributes present in all config classes are:
+:obj:`hidden_size`, :obj:`num_attention_heads`, and :obj:`num_hidden_layers`. Text models further implement:
+:obj:`vocab_size`.
+
+
 
 PretrainedConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.PretrainedConfig
+    :special-members: push_to_hub
     :members:

docs/source/main_classes/data_collator.rst

@@ -18,7 +18,7 @@ the same type as the elements of :obj:`train_dataset` or :obj:`eval_dataset`.
 
 To be able to build batches, data collators may apply some processing (like padding). Some of them (like
 :class:`~transformers.DataCollatorForLanguageModeling`) also apply some random data augmentation (like random masking)
-oin the formed batch.
+on the formed batch.
 
 Examples of use can be found in the :doc:`example scripts <../examples>` or :doc:`example notebooks <../notebooks>`.
 
@@ -29,6 +29,13 @@ Default data collator
 .. autofunction:: transformers.data.data_collator.default_data_collator
 
 
+DefaultDataCollator
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DefaultDataCollator
+    :members:
+
+
 DataCollatorWithPadding
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -54,18 +61,18 @@ DataCollatorForLanguageModeling
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.data.data_collator.DataCollatorForLanguageModeling
-    :members: mask_tokens
+    :members: numpy_mask_tokens, tf_mask_tokens, torch_mask_tokens
 
 
 DataCollatorForWholeWordMask
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.data.data_collator.DataCollatorForWholeWordMask
-    :members: mask_tokens
+    :members: numpy_mask_tokens, tf_mask_tokens, torch_mask_tokens
 
 
 DataCollatorForPermutationLanguageModeling
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.data.data_collator.DataCollatorForPermutationLanguageModeling
-    :members: mask_tokens
+    :members: numpy_mask_tokens, tf_mask_tokens, torch_mask_tokens

docs/source/main_classes/keras_callbacks.rst

@@ -0,0 +1,22 @@
+..
+    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.
+
+Keras callbacks
+=======================================================================================================================
+
+When training a Transformers model with Keras, there are some library-specific callbacks available to automate common
+tasks:
+
+PushToHubCallback
+-----------------------------------------------------------------------------------------------------------------------
+
+.. autoclass:: transformers.keras_callbacks.PushToHubCallback

docs/source/main_classes/logging.rst

@@ -1,4 +1,4 @@
-.. 
+..
     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
@@ -32,6 +32,15 @@ to one of the following: ``debug``, ``info``, ``warning``, ``error``, ``critical
 
     TRANSFORMERS_VERBOSITY=error ./myprogram.py
 
+Additionally, some ``warnings`` can be disabled by setting the environment variable
+``TRANSFORMERS_NO_ADVISORY_WARNINGS`` to a true value, like `1`. This will disable any warning that is logged using
+:meth:`logger.warning_advice`. For example:
+
+
+.. code-block:: bash
+
+    TRANSFORMERS_NO_ADVISORY_WARNINGS=1 ./myprogram.py
+
 All the methods of this logging module are documented below, the main ones are
 :func:`transformers.logging.get_verbosity` to get the current level of verbosity in the logger and
 :func:`transformers.logging.set_verbosity` to set the verbosity to the level of your choice. In order (from the least

docs/source/main_classes/model.rst

@@ -1,4 +1,4 @@
-.. 
+..
     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
@@ -35,9 +35,41 @@ PreTrainedModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.PreTrainedModel
+    :special-members: push_to_hub
     :members:
 
 
+.. _from_pretrained-torch-dtype:
+
+Model Instantiation dtype
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Under Pytorch a model normally gets instantiated with ``torch.float32`` format. This can be an issue if one tries to
+load a model whose weights are in fp16, since it'd require twice as much memory. To overcome this limitation, you can
+either explicitly pass the desired ``dtype`` using ``torch_dtype`` argument:
+
+.. code-block:: python
+
+    model = T5ForConditionalGeneration.from_pretrained("t5", torch_dtype=torch.float16)
+
+or, if you want the model to always load in the most optimal memory pattern, you can use the special value ``"auto"``,
+and then ``dtype`` will be automatically derived from the model's weights:
+
+.. code-block:: python
+
+    model = T5ForConditionalGeneration.from_pretrained("t5", torch_dtype="auto")
+
+Models instantiated from scratch can also be told which ``dtype`` to use with:
+
+.. code-block:: python
+
+    config = T5Config.from_pretrained("t5")
+    model = AutoModel.from_config(config)
+
+Due to Pytorch design, this functionality is only available for floating dtypes.
+
+
+
 ModuleUtilsMixin
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -49,6 +81,7 @@ TFPreTrainedModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.TFPreTrainedModel
+    :special-members: push_to_hub
     :members:
 
 
@@ -63,6 +96,7 @@ FlaxPreTrainedModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.FlaxPreTrainedModel
+    :special-members: push_to_hub
     :members:
 
 

docs/source/main_classes/optimizer_schedules.rst

@@ -52,30 +52,30 @@ Learning Rate Schedules (Pytorch)
 
 .. autofunction:: transformers.get_constant_schedule_with_warmup
 
-.. image:: /imgs/warmup_constant_schedule.png
-    :target: /imgs/warmup_constant_schedule.png
+.. image:: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_constant_schedule.png
+    :target: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_constant_schedule.png
     :alt:
 
 
 .. autofunction:: transformers.get_cosine_schedule_with_warmup
 
-.. image:: /imgs/warmup_cosine_schedule.png
-    :target: /imgs/warmup_cosine_schedule.png
+.. image:: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_cosine_schedule.png
+    :target: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_cosine_schedule.png
     :alt:
 
 
 .. autofunction:: transformers.get_cosine_with_hard_restarts_schedule_with_warmup
 
-.. image:: /imgs/warmup_cosine_hard_restarts_schedule.png
-    :target: /imgs/warmup_cosine_hard_restarts_schedule.png
+.. image:: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_cosine_hard_restarts_schedule.png
+    :target: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_cosine_hard_restarts_schedule.png
     :alt:
 
 
 
 .. autofunction:: transformers.get_linear_schedule_with_warmup
 
-.. image:: /imgs/warmup_linear_schedule.png
-    :target: /imgs/warmup_linear_schedule.png
+.. image:: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_linear_schedule.png
+    :target: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/warmup_linear_schedule.png
     :alt:
 
 

docs/source/main_classes/output.rst

@@ -210,6 +210,13 @@ TFBaseModelOutputWithPooling
     :members:
 
 
+TFBaseModelOutputWithPoolingAndCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_tf_outputs.TFBaseModelOutputWithPoolingAndCrossAttentions
+    :members:
+
+
 TFBaseModelOutputWithPast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -217,6 +224,13 @@ TFBaseModelOutputWithPast
     :members:
 
 
+TFBaseModelOutputWithPastAndCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_tf_outputs.TFBaseModelOutputWithPastAndCrossAttentions
+    :members:
+
+
 TFSeq2SeqModelOutput
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -231,6 +245,13 @@ TFCausalLMOutput
     :members:
 
 
+TFCausalLMOutputWithCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_tf_outputs.TFCausalLMOutputWithCrossAttentions
+    :members:
+
+
 TFCausalLMOutputWithPast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -299,3 +320,93 @@ TFSeq2SeqQuestionAnsweringModelOutput
 
 .. autoclass:: transformers.modeling_tf_outputs.TFSeq2SeqQuestionAnsweringModelOutput
     :members:
+
+
+FlaxBaseModelOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutput
+
+
+FlaxBaseModelOutputWithPast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPast
+
+
+FlaxBaseModelOutputWithPooling
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling
+
+
+FlaxBaseModelOutputWithPastAndCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPastAndCrossAttentions
+
+
+FlaxSeq2SeqModelOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqModelOutput
+
+
+FlaxCausalLMOutputWithCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxCausalLMOutputWithCrossAttentions
+
+
+FlaxMaskedLMOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxMaskedLMOutput
+
+
+FlaxSeq2SeqLMOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqLMOutput
+
+
+FlaxNextSentencePredictorOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxNextSentencePredictorOutput
+
+
+FlaxSequenceClassifierOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxSequenceClassifierOutput
+
+
+FlaxSeq2SeqSequenceClassifierOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqSequenceClassifierOutput
+
+
+FlaxMultipleChoiceModelOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxMultipleChoiceModelOutput
+
+
+FlaxTokenClassifierOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxTokenClassifierOutput
+
+
+FlaxQuestionAnsweringModelOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxQuestionAnsweringModelOutput
+
+
+FlaxSeq2SeqQuestionAnsweringModelOutput
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqQuestionAnsweringModelOutput

docs/source/main_classes/pipelines.rst

@@ -23,33 +23,262 @@ There are two categories of pipeline abstractions to be aware about:
 - The :func:`~transformers.pipeline` which is the most powerful object encapsulating all other pipelines.
 - The other task-specific pipelines:
 
+    - :class:`~transformers.AudioClassificationPipeline`
     - :class:`~transformers.AutomaticSpeechRecognitionPipeline`
     - :class:`~transformers.ConversationalPipeline`
     - :class:`~transformers.FeatureExtractionPipeline`
     - :class:`~transformers.FillMaskPipeline`
     - :class:`~transformers.ImageClassificationPipeline`
+    - :class:`~transformers.ImageSegmentationPipeline`
+    - :class:`~transformers.ObjectDetectionPipeline`
     - :class:`~transformers.QuestionAnsweringPipeline`
     - :class:`~transformers.SummarizationPipeline`
+    - :class:`~transformers.TableQuestionAnsweringPipeline`
     - :class:`~transformers.TextClassificationPipeline`
     - :class:`~transformers.TextGenerationPipeline`
+    - :class:`~transformers.Text2TextGenerationPipeline`
     - :class:`~transformers.TokenClassificationPipeline`
     - :class:`~transformers.TranslationPipeline`
     - :class:`~transformers.ZeroShotClassificationPipeline`
-    - :class:`~transformers.Text2TextGenerationPipeline`
-    - :class:`~transformers.TableQuestionAnsweringPipeline`
 
 The pipeline abstraction
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 The `pipeline` abstraction is a wrapper around all the other available pipelines. It is instantiated as any other
-pipeline but requires an additional argument which is the `task`.
+pipeline but can provide additional quality of life.
+
+Simple call on one item:
+
+.. code-block::
+
+    >>> pipe = pipeline("text-classification")
+    >>> pipe("This restaurant is awesome")
+    [{'label': 'POSITIVE', 'score': 0.9998743534088135}]
+
+If you want to use a specific model from the `hub <https://huggingface.co>`__ you can ignore the task if the model on
+the hub already defines it:
+
+.. code-block::
+
+    >>> pipe = pipeline(model="roberta-large-mnli")
+    >>> pipe("This restaurant is awesome")
+    [{'label': 'POSITIVE', 'score': 0.9998743534088135}]
+
+To call a pipeline on many items, you can either call with a `list`.
+
+.. code-block::
+
+    >>> pipe = pipeline("text-classification")
+    >>> pipe(["This restaurant is awesome", "This restaurant is aweful"])
+    [{'label': 'POSITIVE', 'score': 0.9998743534088135},
+     {'label': 'NEGATIVE', 'score': 0.9996669292449951}]
+
+
+To iterate of full datasets it is recommended to use a :obj:`dataset` directly. This means you don't need to allocate
+the whole dataset at once, nor do you need to do batching yourself. This should work just as fast as custom loops on
+GPU. If it doesn't don't hesitate to create an issue.
+
+.. code-block::
+
+    import datasets
+    from transformers import pipeline
+    from transformers.pipelines.base import KeyDataset
+    import tqdm
+
+    pipe = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h", device=0)
+    dataset = datasets.load_dataset("superb", name="asr", split="test")
+
+    # KeyDataset (only `pt`) will simply return the item in the dict returned by the dataset item
+    # as we're not interested in the `target` part of the dataset.
+    for out in tqdm.tqdm(pipe(KeyDataset(dataset, "file"))):
+        print(out)
+        # {"text": "NUMBER TEN FRESH NELLY IS WAITING ON YOU GOOD NIGHT HUSBAND"}
+        # {"text": ....}
+        # ....
+
 
 .. autofunction:: transformers.pipeline
 
+Pipeline batching
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+All pipelines (except `zero-shot-classification` and `question-answering` currently) can use batching. This will work
+whenever the pipeline uses its streaming ability (so when passing lists or :obj:`Dataset`).
+
+.. code-block::
+
+    from transformers import pipeline                                                   
+    from transformers.pipelines.base import KeyDataset
+    import datasets
+    import tqdm                                                                         
+
+    dataset = datasets.load_dataset("imdb", name="plain_text", split="unsupervised")
+    pipe = pipeline("text-classification", device=0)
+    for out in pipe(KeyDataset(dataset, "text"), batch_size=8, truncation="only_first"):
+        print(out)
+        # [{'label': 'POSITIVE', 'score': 0.9998743534088135}]
+        # Exactly the same output as before, but the content are passed
+        # as batches to the model
+
+
+.. warning::
+
+    However, this is not automatically a win for performance. It can be either a 10x speedup or 5x slowdown depending
+    on hardware, data and the actual model being used.
+
+    Example where it's most a speedup:
+
+
+.. code-block::
+
+    from transformers import pipeline                                                   
+    from torch.utils.data import Dataset                                                
+    import tqdm                                                                         
+
+
+    pipe = pipeline("text-classification", device=0)                                    
+
+
+    class MyDataset(Dataset):                                                           
+        def __len__(self):                                                              
+            return 5000                                                                 
+
+        def __getitem__(self, i):                                                       
+            return "This is a test"                                                     
+
+
+    dataset = MyDataset()   
+
+    for batch_size in [1, 8, 64, 256]:
+        print("-" * 30)                                                                     
+        print(f"Streaming batch_size={batch_size}")    
+        for out in tqdm.tqdm(pipe(dataset, batch_size=batch_size), total=len(dataset)):              
+            pass
+
+
+.. code-block::
+
+    # On GTX 970
+    ------------------------------
+    Streaming no batching
+    100%|██████████████████████████████████████████████████████████████████████| 5000/5000 [00:26<00:00, 187.52it/s]
+    ------------------------------
+    Streaming batch_size=8
+    100%|█████████████████████████████████████████████████████████████████████| 5000/5000 [00:04<00:00, 1205.95it/s]
+    ------------------------------
+    Streaming batch_size=64
+    100%|█████████████████████████████████████████████████████████████████████| 5000/5000 [00:02<00:00, 2478.24it/s]
+    ------------------------------
+    Streaming batch_size=256
+    100%|█████████████████████████████████████████████████████████████████████| 5000/5000 [00:01<00:00, 2554.43it/s]
+    (diminishing returns, saturated the GPU)
+
+
+Example where it's most a slowdown:
+
+.. code-block::
+
+    class MyDataset(Dataset):                                                           
+        def __len__(self):                                                              
+            return 5000                                                                 
+
+        def __getitem__(self, i):                                                       
+            if i % 64 == 0:                                                          
+                n = 100                                                              
+            else:                                                                    
+                n = 1                                                                
+            return "This is a test" * n
+
+This is a occasional very long sentence compared to the other. In that case, the **whole** batch will need to be 400
+tokens long, so the whole batch will be [64, 400] instead of [64, 4], leading to the high slowdown. Even worse, on
+bigger batches, the program simply crashes.
+
+
+.. code-block::
+
+    ------------------------------
+    Streaming no batching
+    100%|█████████████████████████████████████████████████████████████████████| 1000/1000 [00:05<00:00, 183.69it/s]
+    ------------------------------
+    Streaming batch_size=8
+    100%|█████████████████████████████████████████████████████████████████████| 1000/1000 [00:03<00:00, 265.74it/s]
+    ------------------------------
+    Streaming batch_size=64
+    100%|██████████████████████████████████████████████████████████████████████| 1000/1000 [00:26<00:00, 37.80it/s]
+    ------------------------------
+    Streaming batch_size=256
+      0%|                                                                                 | 0/1000 [00:00<?, ?it/s]
+    Traceback (most recent call last):
+      File "/home/nicolas/src/transformers/test.py", line 42, in <module>
+        for out in tqdm.tqdm(pipe(dataset, batch_size=256), total=len(dataset)):
+    ....
+        q = q / math.sqrt(dim_per_head)  # (bs, n_heads, q_length, dim_per_head)
+    RuntimeError: CUDA out of memory. Tried to allocate 376.00 MiB (GPU 0; 3.95 GiB total capacity; 1.72 GiB already allocated; 354.88 MiB free; 2.46 GiB reserved in total by PyTorch)
+
+
+There are no good (general) solutions for this problem, and your mileage may vary depending on your use cases. Rule of
+thumb:
+
+For users, a rule of thumb is:
+
+- **Measure performance on your load, with your hardware. Measure, measure, and keep measuring. Real numbers are the
+  only way to go.**
+- If you are latency constrained (live product doing inference), don't batch
+- If you are using CPU, don't batch.
+- If you are using throughput (you want to run your model on a bunch of static data), on GPU, then:
+
+      - If you have no clue about the size of the sequence_length ("natural" data), by default don't batch, measure and
+        try tentatively to add it, add OOM checks to recover when it will fail (and it will at some point if you don't
+        control the sequence_length.)
+      - If your sequence_length is super regular, then batching is more likely to be VERY interesting, measure and push
+        it until you get OOMs.
+      - The larger the GPU the more likely batching is going to be more interesting
+- As soon as you enable batching, make sure you can handle OOMs nicely.
+
+Pipeline custom code
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+If you want to override a specific pipeline.
+
+Don't hesitate to create an issue for your task at hand, the goal of the pipeline is to be easy to use and support most
+cases, so :obj:`transformers` could maybe support your use case.
+
+
+If you want to try simply you can:
+
+- Subclass your pipeline of choice
+
+.. code-block::
+
+    class MyPipeline(TextClassificationPipeline):
+        def postprocess(...):
+            ...
+            scores = scores * 100
+            ...
+
+    my_pipeline = MyPipeline(model=model, tokenizer=tokenizer, ...)
+    # or if you use `pipeline` function, then:
+    my_pipeline = pipeline(model="xxxx", pipeline_class=MyPipeline)
+
+That should enable you to do all the custom code you want.
+
+
+Implementing a pipeline
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+:doc:`Implementing a new pipeline <../add_new_pipeline>`
 
 The task specific pipelines
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+
+AudioClassificationPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.AudioClassificationPipeline
+    :special-members: __call__
+    :members:
+
 AutomaticSpeechRecognitionPipeline
 =======================================================================================================================
 
@@ -87,6 +316,13 @@ ImageClassificationPipeline
     :special-members: __call__
     :members:
 
+ImageSegmentationPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.ImageSegmentationPipeline
+    :special-members: __call__
+    :members:
+
 NerPipeline
 =======================================================================================================================
 
@@ -94,6 +330,13 @@ NerPipeline
 
 See :class:`~transformers.TokenClassificationPipeline` for all details.
 
+ObjectDetectionPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.ObjectDetectionPipeline
+    :special-members: __call__
+    :members:
+
 QuestionAnsweringPipeline
 =======================================================================================================================
 

docs/source/main_classes/tokenizer.rst

@@ -20,7 +20,7 @@ Rust library `tokenizers <https://github.com/huggingface/tokenizers>`__. The "Fa
 1. a significant speed-up in particular when doing batched tokenization and
 2. additional methods to map between the original string (character and words) and the token space (e.g. getting the
    index of the token comprising a given character or the span of characters corresponding to a given token). Currently
-   no "Fast" implementation is available for the SentencePiece-based tokenizers (for T5, ALBERT, CamemBERT, XLMRoBERTa
+   no "Fast" implementation is available for the SentencePiece-based tokenizers (for T5, ALBERT, CamemBERT, XLM-RoBERTa
    and XLNet models).
 
 The base classes :class:`~transformers.PreTrainedTokenizer` and :class:`~transformers.PreTrainedTokenizerFast`
@@ -39,7 +39,8 @@ methods for using all the tokenizers:
 - Managing special tokens (like mask, beginning-of-sentence, etc.): adding them, assigning them to attributes in the
   tokenizer for easy access and making sure they are not split during tokenization.
 
-:class:`~transformers.BatchEncoding` holds the output of the tokenizer's encoding methods (``__call__``,
+:class:`~transformers.BatchEncoding` holds the output of the
+:class:`~transformers.tokenization_utils_base.PreTrainedTokenizerBase`'s encoding methods (``__call__``,
 ``encode_plus`` and ``batch_encode_plus``) and is derived from a Python dictionary. When the tokenizer is a pure python
 tokenizer, this class behaves just like a standard python dictionary and holds the various model inputs computed by
 these methods (``input_ids``, ``attention_mask``...). When the tokenizer is a "Fast" tokenizer (i.e., backed by
@@ -53,10 +54,8 @@ PreTrainedTokenizer
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.PreTrainedTokenizer
-    :special-members: __call__
-    :members: batch_decode, convert_ids_to_tokens, convert_tokens_to_ids, convert_tokens_to_string, decode, encode, 
-        get_added_vocab, get_special_tokens_mask, num_special_tokens_to_add, prepare_for_tokenization, tokenize,
-        vocab_size
+    :special-members: __call__, batch_decode, decode, encode, push_to_hub
+    :members: 
 
 
 PreTrainedTokenizerFast
@@ -68,10 +67,8 @@ loaded very simply into 🤗 transformers. Take a look at the :doc:`Using tokeni
 <../fast_tokenizers>` page to understand how this is done.
 
 .. autoclass:: transformers.PreTrainedTokenizerFast
-    :special-members: __call__
-    :members: batch_decode, convert_ids_to_tokens, convert_tokens_to_ids, convert_tokens_to_string, decode, encode, 
-        get_added_vocab, get_special_tokens_mask, num_special_tokens_to_add,
-        set_truncation_and_padding,tokenize, vocab_size
+    :special-members: __call__, batch_decode, decode, encode, push_to_hub
+    :members:
 
 
 BatchEncoding

docs/source/main_classes/trainer.mdx

@@ -0,0 +1,470 @@
+<!--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.
+-->
+
+# Trainer
+
+The [`Trainer`] class provides an API for feature-complete training in PyTorch for most standard use cases. It's used in most of the [example scripts](../examples).
+
+Before instantiating your [`Trainer`], create a [`TrainingArguments`] to access all the points of customization during training.
+
+The API supports distributed training on multiple GPUs/TPUs, mixed precision through [NVIDIA Apex](https://github.com/NVIDIA/apex) and Native AMP for PyTorch.
+
+The [`Trainer`] contains the basic training loop which supports the above features. To inject custom behavior you can subclass them and override the following methods:
+
+- **get_train_dataloader** -- Creates the training DataLoader.
+- **get_eval_dataloader** -- Creates the evaluation DataLoader.
+- **get_test_dataloader** -- Creates the test DataLoader.
+- **log** -- Logs information on the various objects watching training.
+- **create_optimizer_and_scheduler** -- Sets up the optimizer and learning rate scheduler if they were not passed at
+  init. Note, that you can also subclass or override the `create_optimizer` and `create_scheduler` methods
+  separately.
+- **create_optimizer** -- Sets up the optimizer if it wasn't passed at init.
+- **create_scheduler** -- Sets up the learning rate scheduler if it wasn't passed at init.
+- **compute_loss** - Computes the loss on a batch of training inputs.
+- **training_step** -- Performs a training step.
+- **prediction_step** -- Performs an evaluation/test step.
+- **evaluate** -- Runs an evaluation loop and returns metrics.
+- **predict** -- Returns predictions (with metrics if labels are available) on a test set.
+
+<Tip warning={true}>
+
+The [`Trainer`] class is optimized for 🤗 Transformers models and can have surprising behaviors
+when you use it on other models. When using it on your own model, make sure:
+
+- your model always return tuples or subclasses of [`~file_utils.ModelOutput`].
+- your model can compute the loss if a `labels` argument is provided and that loss is returned as the first
+  element of the tuple (if your model returns tuples)
+- your model can accept multiple label arguments (use the `label_names` in your [`TrainingArguments`] to indicate their name to the [`Trainer`]) but none of them should be named `"label"`.
+
+</Tip>
+
+Here is an example of how to customize [`Trainer`] using a custom loss function for multi-label classification:
+
+```python
+from torch import nn
+from transformers import Trainer
+
+class MultilabelTrainer(Trainer):
+    def compute_loss(self, model, inputs, return_outputs=False):
+        labels = inputs.get("labels")
+        outputs = model(**inputs)
+        logits = outputs.get('logits')
+        loss_fct = nn.BCEWithLogitsLoss()
+        loss = loss_fct(logits.view(-1, self.model.config.num_labels),
+                        labels.float().view(-1, self.model.config.num_labels))
+        return (loss, outputs) if return_outputs else loss
+```
+
+Another way to customize the training loop behavior for the PyTorch [`Trainer`] is to use [callbacks](callback) that can inspect the training loop state (for progress reporting, logging on TensorBoard or other ML platforms...) and take decisions (like early stopping).
+
+
+## Trainer
+
+[[autodoc]] Trainer
+    - all
+
+## Seq2SeqTrainer
+
+[[autodoc]] Seq2SeqTrainer
+    - evaluate
+    - predict
+
+## TrainingArguments
+
+[[autodoc]] TrainingArguments
+    - all
+
+## Seq2SeqTrainingArguments
+
+[[autodoc]] Seq2SeqTrainingArguments
+    - all
+
+## Checkpoints
+
+By default, [`Trainer`] will save all checkpoints in the `output_dir` you set in the
+[`TrainingArguments`] you are using. Those will go in subfolder named `checkpoint-xxx` with xxx
+being the step at which the training was at.
+
+Resuming training from a checkpoint can be done when calling [`Trainer.train`] with either:
+
+- `resume_from_checkpoint=True` which will resume training from the latest checkpoint
+- `resume_from_checkpoint=checkpoint_dir` which will resume training from the specific checkpoint in the directory
+  passed.
+
+In addition, you can easily save your checkpoints on the Model Hub when using `push_to_hub=True`. By default, all
+the models saved in intermediate checkpoints are saved in different commits, but not the optimizer state. You can adapt
+the `hub-strategy` value of your [`TrainingArguments`] to either:
+
+- `"checkpoint"`: the latest checkpoint is also pushed in a subfolder named last-checkpoint, allowing you to
+  resume training easily with `trainer.train(resume_from_checkpoint="output_dir/last-checkpoint")`.
+- `"all_checkpoints"`: all checkpoints are pushed like they appear in the output folder (so you will get one
+  checkpoint folder per folder in your final repository)
+
+
+## Logging
+
+By default [`Trainer`] will use `logging.INFO` for the main process and `logging.WARNING` for the replicas if any.
+
+These defaults can be overridden to use any of the 5 `logging` levels with [`TrainingArguments`]'s
+arguments:
+
+- `log_level` - for the main process
+- `log_level_replica` - for the replicas
+
+Further, if [`TrainingArguments`]'s `log_on_each_node` is set to `False` only the main node will
+use the log level settings for its main process, all other nodes will use the log level settings for replicas.
+
+Note that [`Trainer`] is going to set `transformers`'s log level separately for each node in its
+[`Trainer.__init__`]. So you may want to set this sooner (see the next example) if you tap into other
+`transformers` functionality before creating the [`Trainer`] object.
+
+Here is an example of how this can be used in an application:
+
+```python
+[...]
+logger = logging.getLogger(__name__)
+
+# Setup logging
+logging.basicConfig(
+    format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+    datefmt="%m/%d/%Y %H:%M:%S",
+    handlers=[logging.StreamHandler(sys.stdout)],
+)
+
+# set the main code and the modules it uses to the same log-level according to the node
+log_level = training_args.get_process_log_level()
+logger.setLevel(log_level)
+datasets.utils.logging.set_verbosity(log_level)
+transformers.utils.logging.set_verbosity(log_level)
+
+trainer = Trainer(...)
+```
+
+And then if you only want to see warnings on the main node and all other nodes to not print any most likely duplicated
+warnings you could run it as:
+
+```bash
+my_app.py ... --log_level warning --log_level_replica error
+```
+
+In the multi-node environment if you also don't want the logs to repeat for each node's main process, you will want to
+change the above to:
+
+```bash
+my_app.py ... --log_level warning --log_level_replica error --log_on_each_node 0
+```
+
+and then only the main process of the first node will log at the "warning" level, and all other processes on the main
+node and all processes on other nodes will log at the "error" level.
+
+If you need your application to be as quiet as possible you could do:
+
+```bash
+my_app.py ... --log_level error --log_level_replica error --log_on_each_node 0
+```
+
+(add `--log_on_each_node 0` if on multi-node environment)
+
+
+## Randomness
+
+When resuming from a checkpoint generated by [`Trainer`] all efforts are made to restore the
+_python_, _numpy_ and _pytorch_ RNG states to the same states as they were at the moment of saving that checkpoint,
+which should make the "stop and resume" style of training as close as possible to non-stop training.
+
+However, due to various default non-deterministic pytorch settings this might not fully work. If you want full
+determinism please refer to [Controlling sources of randomness](https://pytorch.org/docs/stable/notes/randomness). As explained in the document, that some of those settings
+that make things deterministic (.e.g., `torch.backends.cudnn.deterministic`) may slow things down, therefore this
+can't be done by default, but you can enable those yourself if needed.
+
+
+## Trainer Integrations
+
+The [`Trainer`] has been extended to support libraries that may dramatically improve your training
+time and fit much bigger models.
+
+Currently it supports third party solutions, [DeepSpeed](https://github.com/microsoft/DeepSpeed) and [FairScale](https://github.com/facebookresearch/fairscale/), which implement parts of the paper [ZeRO: Memory Optimizations
+Toward Training Trillion Parameter Models, by Samyam Rajbhandari, Jeff Rasley, Olatunji Ruwase, Yuxiong He](https://arxiv.org/abs/1910.02054).
+
+This provided support is new and experimental as of this writing.
+
+<a id='zero-install-notes'></a>
+
+### CUDA Extension Installation Notes
+
+As of this writing, both FairScale and Deepspeed require compilation of CUDA C++ code, before they can be used.
+
+While all installation issues should be dealt with through the corresponding GitHub Issues of [FairScale](https://github.com/facebookresearch/fairscale/issues) and [Deepspeed](https://github.com/microsoft/DeepSpeed/issues), there are a few common issues that one may encounter while building
+any PyTorch extension that needs to build CUDA extensions.
+
+Therefore, if you encounter a CUDA-related build issue while doing one of the following or both:
+
+```bash
+pip install fairscale
+pip install deepspeed
+```
+
+please, read the following notes first.
+
+In these notes we give examples for what to do when `pytorch` has been built with CUDA `10.2`. If your situation is
+different remember to adjust the version number to the one you are after.
+
+#### Possible problem #1
+
+While, Pytorch comes with its own CUDA toolkit, to build these two projects you must have an identical version of CUDA
+installed system-wide.
+
+For example, if you installed `pytorch` with `cudatoolkit==10.2` in the Python environment, you also need to have
+CUDA `10.2` installed system-wide.
+
+The exact location may vary from system to system, but `/usr/local/cuda-10.2` is the most common location on many
+Unix systems. When CUDA is correctly set up and added to the `PATH` environment variable, one can find the
+installation location by doing:
+
+```bash
+which nvcc
+```
+
+If you don't have CUDA installed system-wide, install it first. You will find the instructions by using your favorite
+search engine. For example, if you're on Ubuntu you may want to search for: [ubuntu cuda 10.2 install](https://www.google.com/search?q=ubuntu+cuda+10.2+install).
+
+#### Possible problem #2
+
+Another possible common problem is that you may have more than one CUDA toolkit installed system-wide. For example you
+may have:
+
+```bash
+/usr/local/cuda-10.2
+/usr/local/cuda-11.0
+```
+
+Now, in this situation you need to make sure that your `PATH` and `LD_LIBRARY_PATH` environment variables contain
+the correct paths to the desired CUDA version. Typically, package installers will set these to contain whatever the
+last version was installed. If you encounter the problem, where the package build fails because it can't find the right
+CUDA version despite you having it installed system-wide, it means that you need to adjust the 2 aforementioned
+environment variables.
+
+First, you may look at their contents:
+
+```bash
+echo $PATH
+echo $LD_LIBRARY_PATH
+```
+
+so you get an idea of what is inside.
+
+It's possible that `LD_LIBRARY_PATH` is empty.
+
+`PATH` lists the locations of where executables can be found and `LD_LIBRARY_PATH` is for where shared libraries
+are to looked for. In both cases, earlier entries have priority over the later ones. `:` is used to separate multiple
+entries.
+
+Now, to tell the build program where to find the specific CUDA toolkit, insert the desired paths to be listed first by
+doing:
+
+```bash
+export PATH=/usr/local/cuda-10.2/bin:$PATH
+export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH
+```
+
+Note that we aren't overwriting the existing values, but prepending instead.
+
+Of course, adjust the version number, the full path if need be. Check that the directories you assign actually do
+exist. `lib64` sub-directory is where the various CUDA `.so` objects, like `libcudart.so` reside, it's unlikely
+that your system will have it named differently, but if it is adjust it to reflect your reality.
+
+
+#### Possible problem #3
+
+Some older CUDA versions may refuse to build with newer compilers. For example, you my have `gcc-9` but it wants
+`gcc-7`.
+
+There are various ways to go about it.
+
+If you can install the latest CUDA toolkit it typically should support the newer compiler.
+
+Alternatively, you could install the lower version of the compiler in addition to the one you already have, or you may
+already have it but it's not the default one, so the build system can't see it. If you have `gcc-7` installed but the
+build system complains it can't find it, the following might do the trick:
+
+```bash
+sudo ln -s /usr/bin/gcc-7  /usr/local/cuda-10.2/bin/gcc
+sudo ln -s /usr/bin/g++-7  /usr/local/cuda-10.2/bin/g++
+```
+
+Here, we are making a symlink to `gcc-7` from `/usr/local/cuda-10.2/bin/gcc` and since
+`/usr/local/cuda-10.2/bin/` should be in the `PATH` environment variable (see the previous problem's solution), it
+should find `gcc-7` (and `g++7`) and then the build will succeed.
+
+As always make sure to edit the paths in the example to match your situation.
+
+### FairScale
+
+By integrating [FairScale](https://github.com/facebookresearch/fairscale/) the [`Trainer`]
+provides support for the following features from [the ZeRO paper](https://arxiv.org/abs/1910.02054):
+
+1. Optimizer State Sharding
+2. Gradient Sharding
+3. Model Parameters Sharding (new and very experimental)
+4. CPU offload (new and very experimental)
+
+You will need at least two GPUs to use this feature.
+
+
+**Installation**:
+
+Install the library via pypi:
+
+```bash
+pip install fairscale
+```
+
+or via `transformers`' `extras`:
+
+```bash
+pip install transformers[fairscale]
+```
+
+(available starting from `transformers==4.6.0`) or find more details on [the FairScale's GitHub page](https://github.com/facebookresearch/fairscale/#installation).
+
+If you're still struggling with the build, first make sure to read [CUDA Extension Installation Notes](#zero-install-notes).
+
+If it's still not resolved the build issue, here are a few more ideas.
+
+`fairscale` seems to have an issue with the recently introduced by pip build isolation feature. If you have a problem
+with it, you may want to try one of:
+
+```bash
+pip install fairscale --no-build-isolation .
+```
+
+or:
+
+```bash
+git clone https://github.com/facebookresearch/fairscale/
+cd fairscale
+rm -r dist build
+python setup.py bdist_wheel
+pip uninstall -y fairscale
+pip install dist/fairscale-*.whl
+```
+
+`fairscale` also has issues with building against pytorch-nightly, so if you use it you may have to try one of:
+
+```bash
+pip uninstall -y fairscale; pip install fairscale --pre \
+-f https://download.pytorch.org/whl/nightly/cu110/torch_nightly \
+--no-cache --no-build-isolation
+```
+
+or:
+
+```bash
+pip install -v --disable-pip-version-check . \
+-f https://download.pytorch.org/whl/nightly/cu110/torch_nightly --pre
+```
+
+Of course, adjust the urls to match the cuda version you use.
+
+If after trying everything suggested you still encounter build issues, please, proceed with the GitHub Issue of
+[FairScale](https://github.com/facebookresearch/fairscale/issues).
+
+
+
+**Usage**:
+
+To use the first version of Sharded data-parallelism, add `--sharded_ddp simple` to the command line arguments, and
+make sure you have added the distributed launcher `-m torch.distributed.launch --nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE` if you haven't been using it already.
+
+For example here is how you could use it for `run_translation.py` with 2 GPUs:
+
+```bash
+python -m torch.distributed.launch --nproc_per_node=2 examples/pytorch/translation/run_translation.py \
+--model_name_or_path t5-small --per_device_train_batch_size 1   \
+--output_dir output_dir --overwrite_output_dir \
+--do_train --max_train_samples 500 --num_train_epochs 1 \
+--dataset_name wmt16 --dataset_config "ro-en" \
+--source_lang en --target_lang ro \
+--fp16 --sharded_ddp simple
+```
+
+Notes:
+
+- This feature requires distributed training (so multiple GPUs).
+- It is not implemented for TPUs.
+- It works with `--fp16` too, to make things even faster.
+- One of the main benefits of enabling `--sharded_ddp simple` is that it uses a lot less GPU memory, so you should be
+  able to use significantly larger batch sizes using the same hardware (e.g. 3x and even bigger) which should lead to
+  significantly shorter training time.
+
+3. To use the second version of Sharded data-parallelism, add `--sharded_ddp zero_dp_2` or `--sharded_ddp zero_dp_3` to the command line arguments, and make sure you have added the distributed launcher `-m torch.distributed.launch --nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE` if you haven't been using it already.
+
+For example here is how you could use it for `run_translation.py` with 2 GPUs:
+
+```bash
+python -m torch.distributed.launch --nproc_per_node=2 examples/pytorch/translation/run_translation.py \
+--model_name_or_path t5-small --per_device_train_batch_size 1   \
+--output_dir output_dir --overwrite_output_dir \
+--do_train --max_train_samples 500 --num_train_epochs 1 \
+--dataset_name wmt16 --dataset_config "ro-en" \
+--source_lang en --target_lang ro \
+--fp16 --sharded_ddp zero_dp_2
+```
+
+`zero_dp_2` is an optimized version of the simple wrapper, while `zero_dp_3` fully shards model weights,
+gradients and optimizer states.
+
+Both are compatible with adding `cpu_offload` to enable ZeRO-offload (activate it like this: `--sharded_ddp "zero_dp_2 cpu_offload"`).
+
+Notes:
+
+- This feature requires distributed training (so multiple GPUs).
+- It is not implemented for TPUs.
+- It works with `--fp16` too, to make things even faster.
+- The `cpu_offload` additional option requires `--fp16`.
+- This is an area of active development, so make sure you have a source install of fairscale to use this feature as
+  some bugs you encounter may have been fixed there already.
+
+Known caveats:
+
+- This feature is incompatible with `--predict_with_generate` in the _run_translation.py_ script.
+- Using `--sharded_ddp zero_dp_3` requires wrapping each layer of the model in the special container
+  `FullyShardedDataParallelism` of fairscale. It should be used with the option `auto_wrap` if you are not
+  doing this yourself: `--sharded_ddp "zero_dp_3 auto_wrap"`.
+
+
+Sections that were moved:
+
+[ <a href="./deepspeed#deepspeed-trainer-integration">DeepSpeed</a><a id="deepspeed"></a>
+| <a href="./deepspeed#deepspeed-installation">Installation</a><a id="installation"></a>
+| <a href="./deepspeed#deepspeed-multi-gpu">Deployment with multiple GPUs</a><a id="deployment-with-multiple-gpus"></a>
+| <a href="./deepspeed#deepspeed-one-gpu">Deployment with one GPU</a><a id="deployment-with-one-gpu"></a>
+| <a href="./deepspeed#deepspeed-notebook">Deployment in Notebooks</a><a id="deployment-in-notebooks"></a>
+| <a href="./deepspeed#deepspeed-config">Configuration</a><a id="configuration"></a>
+| <a href="./deepspeed#deepspeed-config-passing">Passing Configuration</a><a id="passing-configuration"></a>
+| <a href="./deepspeed#deepspeed-config-shared">Shared Configuration</a><a id="shared-configuration"></a>
+| <a href="./deepspeed#deepspeed-zero">ZeRO</a><a id="zero"></a>
+| <a href="./deepspeed#deepspeed-zero2-config">ZeRO-2 Config</a><a id="zero-2-config"></a>
+| <a href="./deepspeed#deepspeed-zero3-config">ZeRO-3 Config</a><a id="zero-3-config"></a>
+| <a href="./deepspeed#deepspeed-nvme">NVMe Support</a><a id="nvme-support"></a>
+| <a href="./deepspeed#deepspeed-zero2-zero3-performance">ZeRO-2 vs ZeRO-3 Performance</a><a id="zero-2-vs-zero-3-performance"></a>
+| <a href="./deepspeed#deepspeed-zero2-example">ZeRO-2 Example</a><a id="zero-2-example"></a>
+| <a href="./deepspeed#deepspeed-zero3-example">ZeRO-3 Example</a><a id="zero-3-example"></a>
+| <a href="./deepspeed#deepspeed-optimizer">Optimizer</a><a id="optimizer"></a>
+| <a href="./deepspeed#deepspeed-scheduler">Scheduler</a><a id="scheduler"></a>
+| <a href="./deepspeed#deepspeed-fp32">fp32 Precision</a><a id="fp32-precision"></a>
+| <a href="./deepspeed#deepspeed-amp">Automatic Mixed Precision</a><a id="automatic-mixed-precision"></a>
+| <a href="./deepspeed#deepspeed-bs">Batch Size</a><a id="batch-size"></a>
+| <a href="./deepspeed#deepspeed-grad-acc">Gradient Accumulation</a><a id="gradient-accumulation"></a>
+| <a href="./deepspeed#deepspeed-grad-clip">Gradient Clipping</a><a id="gradient-clipping"></a>
+| <a href="./deepspeed#deepspeed-weight-extraction">Getting The Model Weights Out</a><a id="getting-the-model-weights-out"></a>
+]

docs/source/main_classes/trainer.rst

@@ -1,609 +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.
-
-Trainer
------------------------------------------------------------------------------------------------------------------------
-
-The :class:`~transformers.Trainer` and :class:`~transformers.TFTrainer` classes provide an API for feature-complete
-training in most standard use cases. It's used in most of the :doc:`example scripts <../examples>`.
-
-Before instantiating your :class:`~transformers.Trainer`/:class:`~transformers.TFTrainer`, create a
-:class:`~transformers.TrainingArguments`/:class:`~transformers.TFTrainingArguments` to access all the points of
-customization during training.
-
-The API supports distributed training on multiple GPUs/TPUs, mixed precision through `NVIDIA Apex
-<https://github.com/NVIDIA/apex>`__ and Native AMP for PyTorch and :obj:`tf.keras.mixed_precision` for TensorFlow.
-
-Both :class:`~transformers.Trainer` and :class:`~transformers.TFTrainer` contain the basic training loop which supports
-the above features. To inject custom behavior you can subclass them and override the following methods:
-
-- **get_train_dataloader**/**get_train_tfdataset** -- Creates the training DataLoader (PyTorch) or TF Dataset.
-- **get_eval_dataloader**/**get_eval_tfdataset** -- Creates the evaluation DataLoader (PyTorch) or TF Dataset.
-- **get_test_dataloader**/**get_test_tfdataset** -- Creates the test DataLoader (PyTorch) or TF Dataset.
-- **log** -- Logs information on the various objects watching training.
-- **create_optimizer_and_scheduler** -- Sets up the optimizer and learning rate scheduler if they were not passed at
-  init. Note, that you can also subclass or override the ``create_optimizer`` and ``create_scheduler`` methods
-  separately.
-- **create_optimizer** -- Sets up the optimizer if it wasn't passed at init.
-- **create_scheduler** -- Sets up the learning rate scheduler if it wasn't passed at init.
-- **compute_loss** - Computes the loss on a batch of training inputs.
-- **training_step** -- Performs a training step.
-- **prediction_step** -- Performs an evaluation/test step.
-- **run_model** (TensorFlow only) -- Basic pass through the model.
-- **evaluate** -- Runs an evaluation loop and returns metrics.
-- **predict** -- Returns predictions (with metrics if labels are available) on a test set.
-
-.. warning::
-
-    The :class:`~transformers.Trainer` class is optimized for 🤗 Transformers models and can have surprising behaviors
-    when you use it on other models. When using it on your own model, make sure:
-
-    - your model always return tuples or subclasses of :class:`~transformers.file_utils.ModelOutput`.
-    - your model can compute the loss if a :obj:`labels` argument is provided and that loss is returned as the first
-      element of the tuple (if your model returns tuples)
-    - your model can accept multiple label arguments (use the :obj:`label_names` in your
-      :class:`~transformers.TrainingArguments` to indicate their name to the :class:`~transformers.Trainer`) but none
-      of them should be named :obj:`"label"`.
-
-Here is an example of how to customize :class:`~transformers.Trainer` using a custom loss function for multi-label
-classification:
-
-.. code-block:: python
-
-    from torch import nn
-    from transformers import Trainer
-
-    class MultilabelTrainer(Trainer):
-        def compute_loss(self, model, inputs, return_outputs=False):
-            labels = inputs.pop("labels")
-            outputs = model(**inputs)
-            logits = outputs.logits
-            loss_fct = nn.BCEWithLogitsLoss()
-            loss = loss_fct(logits.view(-1, self.model.config.num_labels),
-                            labels.float().view(-1, self.model.config.num_labels))
-            return (loss, outputs) if return_outputs else loss
-
-Another way to customize the training loop behavior for the PyTorch :class:`~transformers.Trainer` is to use
-:doc:`callbacks <callback>` that can inspect the training loop state (for progress reporting, logging on TensorBoard or
-other ML platforms...) and take decisions (like early stopping).
-
-
-Trainer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.Trainer
-    :members:
-
-
-Seq2SeqTrainer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.Seq2SeqTrainer
-    :members: evaluate, predict
-
-
-TFTrainer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFTrainer
-    :members:
-
-
-TrainingArguments
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TrainingArguments
-    :members:
-
-
-Seq2SeqTrainingArguments
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.Seq2SeqTrainingArguments
-    :members:
-
-
-TFTrainingArguments
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFTrainingArguments
-    :members:
-
-
-Logging
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-By default :class:`~transformers.Trainer` will use ``logging.INFO`` for the main process and ``logging.WARNING`` for
-the replicas if any.
-
-These defaults can be overridden to use any of the 5 ``logging`` levels with :class:`~transformers.TrainingArguments`'s
-arguments:
-
-- ``log_level`` - for the main process
-- ``log_level_replica`` - for the replicas
-
-Further, if :class:`~transformers.TrainingArguments`'s ``log_on_each_node`` is set to ``False`` only the main node will
-use the log level settings for its main process, all other nodes will use the log level settings for replicas.
-
-Note that :class:`~transformers.Trainer` is going to set ``transformers``'s log level separately for each node in its
-:meth:`~transformers.Trainer.__init__`. So you may want to set this sooner (see the next example) if you tap into other
-``transformers`` functionality before creating the :class:`~transformers.Trainer` object.
-
-Here is an example of how this can be used in an application:
-
-.. code-block:: python
-
-    [...]
-    logger = logging.getLogger(__name__)
-
-    # Setup logging
-    logging.basicConfig(
-        format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
-        datefmt="%m/%d/%Y %H:%M:%S",
-        handlers=[logging.StreamHandler(sys.stdout)],
-    )
-
-    # set the main code and the modules it uses to the same log-level according to the node
-    log_level = training_args.get_process_log_level()
-    logger.setLevel(log_level)
-    datasets.utils.logging.set_verbosity(log_level)
-    transformers.utils.logging.set_verbosity(log_level)
-
-    trainer = Trainer(...)
-
-And then if you only want to see warnings on the main node and all other nodes to not print any most likely duplicated
-warnings you could run it as:
-
-.. code-block:: bash
-
-    my_app.py ... --log_level warning --log_level_replica error
-
-In the multi-node environment if you also don't want the logs to repeat for each node's main process, you will want to
-change the above to:
-
-.. code-block:: bash
-
-    my_app.py ... --log_level warning --log_level_replica error --log_on_each_node 0
-
-and then only the main process of the first node will log at the "warning" level, and all other processes on the main
-node and all processes on other nodes will log at the "error" level.
-
-If you need your application to be as quiet as possible you could do:
-
-.. code-block:: bash
-
-    my_app.py ... --log_level error --log_level_replica error --log_on_each_node 0
-
-(add ``--log_on_each_node 0`` if on multi-node environment)
-
-
-
-Randomness
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-When resuming from a checkpoint generated by :class:`~transformers.Trainer` all efforts are made to restore the
-`python`, `numpy` and `pytorch` RNG states to the same states as they were at the moment of saving that checkpoint,
-which should make the "stop and resume" style of training as close as possible to non-stop training.
-
-However, due to various default non-deterministic pytorch settings this might not fully work. If you want full
-determinism please refer to `Controlling sources of randomness
-<https://pytorch.org/docs/stable/notes/randomness.html>`__. As explained in the document, that some of those settings
-that make things determinstic (.e.g., ``torch.backends.cudnn.deterministic``) may slow things down, therefore this
-can't be done by default, but you can enable those yourself if needed.
-
-
-Trainer Integrations
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-
-
-The :class:`~transformers.Trainer` has been extended to support libraries that may dramatically improve your training
-time and fit much bigger models.
-
-Currently it supports third party solutions, `DeepSpeed <https://github.com/microsoft/DeepSpeed>`__ and `FairScale
-<https://github.com/facebookresearch/fairscale/>`__, which implement parts of the paper `ZeRO: Memory Optimizations
-Toward Training Trillion Parameter Models, by Samyam Rajbhandari, Jeff Rasley, Olatunji Ruwase, Yuxiong He
-<https://arxiv.org/abs/1910.02054>`__.
-
-This provided support is new and experimental as of this writing.
-
-.. _zero-install-notes:
-
-CUDA Extension Installation Notes
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-As of this writing, both FairScale and Deepspeed require compilation of CUDA C++ code, before they can be used.
-
-While all installation issues should be dealt with through the corresponding GitHub Issues of `FairScale
-<https://github.com/facebookresearch/fairscale/issues>`__ and `Deepspeed
-<https://github.com/microsoft/DeepSpeed/issues>`__, there are a few common issues that one may encounter while building
-any PyTorch extension that needs to build CUDA extensions.
-
-Therefore, if you encounter a CUDA-related build issue while doing one of the following or both:
-
-.. code-block:: bash
-
-    pip install fairscale
-    pip install deepspeed
-
-please, read the following notes first.
-
-In these notes we give examples for what to do when ``pytorch`` has been built with CUDA ``10.2``. If your situation is
-different remember to adjust the version number to the one you are after.
-
-Possible problem #1
-=======================================================================================================================
-
-While, Pytorch comes with its own CUDA toolkit, to build these two projects you must have an identical version of CUDA
-installed system-wide.
-
-For example, if you installed ``pytorch`` with ``cudatoolkit==10.2`` in the Python environment, you also need to have
-CUDA ``10.2`` installed system-wide.
-
-The exact location may vary from system to system, but ``/usr/local/cuda-10.2`` is the most common location on many
-Unix systems. When CUDA is correctly set up and added to the ``PATH`` environment variable, one can find the
-installation location by doing:
-
-.. code-block:: bash
-
-    which nvcc
-
-If you don't have CUDA installed system-wide, install it first. You will find the instructions by using your favorite
-search engine. For example, if you're on Ubuntu you may want to search for: `ubuntu cuda 10.2 install
-<https://www.google.com/search?q=ubuntu+cuda+10.2+install>`__.
-
-Possible problem #2
-=======================================================================================================================
-
-Another possible common problem is that you may have more than one CUDA toolkit installed system-wide. For example you
-may have:
-
-.. code-block:: bash
-
-    /usr/local/cuda-10.2
-    /usr/local/cuda-11.0
-
-Now, in this situation you need to make sure that your ``PATH`` and ``LD_LIBRARY_PATH`` environment variables contain
-the correct paths to the desired CUDA version. Typically, package installers will set these to contain whatever the
-last version was installed. If you encounter the problem, where the package build fails because it can't find the right
-CUDA version despite you having it installed system-wide, it means that you need to adjust the 2 aforementioned
-environment variables.
-
-First, you may look at their contents:
-
-.. code-block:: bash
-
-    echo $PATH
-    echo $LD_LIBRARY_PATH
-
-so you get an idea of what is inside.
-
-It's possible that ``LD_LIBRARY_PATH`` is empty.
-
-``PATH`` lists the locations of where executables can be found and ``LD_LIBRARY_PATH`` is for where shared libraries
-are to looked for. In both cases, earlier entries have priority over the later ones. ``:`` is used to separate multiple
-entries.
-
-Now, to tell the build program where to find the specific CUDA toolkit, insert the desired paths to be listed first by
-doing:
-
-.. code-block:: bash
-
-    export PATH=/usr/local/cuda-10.2/bin:$PATH
-    export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH
-
-Note that we aren't overwriting the existing values, but prepending instead.
-
-Of course, adjust the version number, the full path if need be. Check that the directories you assign actually do
-exist. ``lib64`` sub-directory is where the various CUDA ``.so`` objects, like ``libcudart.so`` reside, it's unlikely
-that your system will have it named differently, but if it is adjust it to reflect your reality.
-
-
-Possible problem #3
-=======================================================================================================================
-
-Some older CUDA versions may refuse to build with newer compilers. For example, you my have ``gcc-9`` but it wants
-``gcc-7``.
-
-There are various ways to go about it.
-
-If you can install the latest CUDA toolkit it typically should support the newer compiler.
-
-Alternatively, you could install the lower version of the compiler in addition to the one you already have, or you may
-already have it but it's not the default one, so the build system can't see it. If you have ``gcc-7`` installed but the
-build system complains it can't find it, the following might do the trick:
-
-.. code-block:: bash
-
-    sudo ln -s /usr/bin/gcc-7  /usr/local/cuda-10.2/bin/gcc
-    sudo ln -s /usr/bin/g++-7  /usr/local/cuda-10.2/bin/g++
-
-
-Here, we are making a symlink to ``gcc-7`` from ``/usr/local/cuda-10.2/bin/gcc`` and since
-``/usr/local/cuda-10.2/bin/`` should be in the ``PATH`` environment variable (see the previous problem's solution), it
-should find ``gcc-7`` (and ``g++7``) and then the build will succeed.
-
-As always make sure to edit the paths in the example to match your situation.
-
-FairScale
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-By integrating `FairScale <https://github.com/facebookresearch/fairscale/>`__ the :class:`~transformers.Trainer`
-provides support for the following features from `the ZeRO paper <https://arxiv.org/abs/1910.02054>`__:
-
-1. Optimizer State Sharding
-2. Gradient Sharding
-3. Model Parameters Sharding (new and very experimental)
-4. CPU offload (new and very experimental)
-
-You will need at least two GPUs to use this feature.
-
-
-**Installation**:
-
-Install the library via pypi:
-
-.. code-block:: bash
-
-    pip install fairscale
-
-or via ``transformers``' ``extras``:
-
-.. code-block:: bash
-
-    pip install transformers[fairscale]
-
-(will become available starting from ``transformers==4.6.0``)
-
-or find more details on `the FairScale's GitHub page <https://github.com/facebookresearch/fairscale/#installation>`__.
-
-If you're still struggling with the build, first make sure to read :ref:`zero-install-notes`.
-
-If it's still not resolved the build issue, here are a few more ideas.
-
-``fairscale`` seems to have an issue with the recently introduced by pip build isolation feature. If you have a problem
-with it, you may want to try one of:
-
-.. code-block:: bash
-
-    pip install fairscale --no-build-isolation .
-
-or:
-
-.. code-block:: bash
-
-    git clone https://github.com/facebookresearch/fairscale/
-    cd fairscale
-    rm -r dist build
-    python setup.py bdist_wheel
-    pip uninstall -y fairscale
-    pip install dist/fairscale-*.whl
-
-``fairscale`` also has issues with building against pytorch-nightly, so if you use it you may have to try one of:
-
-.. code-block:: bash
-
-    pip uninstall -y fairscale; pip install fairscale --pre \
-    -f https://download.pytorch.org/whl/nightly/cu110/torch_nightly.html \
-    --no-cache --no-build-isolation
-
-or:
-
-.. code-block:: bash
-
-    pip install -v --disable-pip-version-check . \
-    -f https://download.pytorch.org/whl/nightly/cu110/torch_nightly.html --pre
-
-Of course, adjust the urls to match the cuda version you use.
-
-If after trying everything suggested you still encounter build issues, please, proceed with the GitHub Issue of
-`FairScale <https://github.com/facebookresearch/fairscale/issues>`__.
-
-
-
-**Usage**:
-
-To use the first version of Sharded data-parallelism, add ``--sharded_ddp simple`` to the command line arguments, and
-make sure you have added the distributed launcher ``-m torch.distributed.launch
---nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE`` if you haven't been using it already.
-
-For example here is how you could use it for ``run_translation.py`` with 2 GPUs:
-
-.. code-block:: bash
-
-    python -m torch.distributed.launch --nproc_per_node=2 examples/pytorch/translation/run_translation.py \
-    --model_name_or_path t5-small --per_device_train_batch_size 1   \
-    --output_dir output_dir --overwrite_output_dir \
-    --do_train --max_train_samples 500 --num_train_epochs 1 \
-    --dataset_name wmt16 --dataset_config "ro-en" \
-    --source_lang en --target_lang ro \
-    --fp16 --sharded_ddp simple
-
-Notes:
-
-- This feature requires distributed training (so multiple GPUs).
-- It is not implemented for TPUs.
-- It works with ``--fp16`` too, to make things even faster.
-- One of the main benefits of enabling ``--sharded_ddp simple`` is that it uses a lot less GPU memory, so you should be
-  able to use significantly larger batch sizes using the same hardware (e.g. 3x and even bigger) which should lead to
-  significantly shorter training time.
-
-3. To use the second version of Sharded data-parallelism, add ``--sharded_ddp zero_dp_2`` or ``--sharded_ddp
-   zero_dp_3`` to the command line arguments, and make sure you have added the distributed launcher ``-m
-   torch.distributed.launch --nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE`` if you haven't been using it already.
-
-For example here is how you could use it for ``run_translation.py`` with 2 GPUs:
-
-.. code-block:: bash
-
-    python -m torch.distributed.launch --nproc_per_node=2 examples/pytorch/translation/run_translation.py \
-    --model_name_or_path t5-small --per_device_train_batch_size 1   \
-    --output_dir output_dir --overwrite_output_dir \
-    --do_train --max_train_samples 500 --num_train_epochs 1 \
-    --dataset_name wmt16 --dataset_config "ro-en" \
-    --source_lang en --target_lang ro \
-    --fp16 --sharded_ddp zero_dp_2
-
-:obj:`zero_dp_2` is an optimized version of the simple wrapper, while :obj:`zero_dp_3` fully shards model weights,
-gradients and optimizer states.
-
-Both are compatible with adding :obj:`cpu_offload` to enable ZeRO-offload (activate it like this: :obj:`--sharded_ddp
-"zero_dp_2 cpu_offload"`).
-
-Notes:
-
-- This feature requires distributed training (so multiple GPUs).
-- It is not implemented for TPUs.
-- It works with ``--fp16`` too, to make things even faster.
-- The ``cpu_offload`` additional option requires ``--fp16``.
-- This is an area of active development, so make sure you have a source install of fairscale to use this feature as
-  some bugs you encounter may have been fixed there already.
-
-Known caveats:
-
-- This feature is incompatible with :obj:`--predict_with_generate` in the `run_translation.py` script.
-- Using :obj:`--sharded_ddp zero_dp_3` requires wrapping each layer of the model in the special container
-  :obj:`FullyShardedDataParallelism` of fairscale. It should be used with the option :obj:`auto_wrap` if you are not
-  doing this yourself: :obj:`--sharded_ddp "zero_dp_3 auto_wrap"`.
-
-
-DeepSpeed
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-
-Moved to :ref:`deepspeed-trainer-integration`.
-
-
-Installation
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-installation`.
-
-
-Deployment with multiple GPUs
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-multi-gpu`.
-
-
-Deployment with one GPU
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-one-gpu`.
-
-
-Deployment in Notebooks
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-notebook`.
-
-
-Configuration
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-config`.
-
-
-Passing Configuration
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-config-passing`.
-
-
-Shared Configuration
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-config-shared`.
-
-ZeRO
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-zero`.
-
-ZeRO-2 Config
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-zero2-config`.
-
-ZeRO-3 Config
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-zero3-config`.
-
-
-NVMe Support
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-nvme`.
-
-ZeRO-2 vs ZeRO-3 Performance
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-zero2-zero3-performance`.
-
-ZeRO-2 Example
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-zero2-example`.
-
-ZeRO-3 Example
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-zero3-example`.
-
-Optimizer and Scheduler
-=======================================================================================================================
-
-
-
-Optimizer
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-optimizer`.
-
-
-Scheduler
-+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-
-Moved to :ref:`deepspeed-scheduler`.
-
-fp32 Precision
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-fp32`.
-
-Automatic Mixed Precision
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-amp`.
-
-Batch Size
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-bs`.
-
-Gradient Accumulation
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-grad-acc`.
-
-
-Gradient Clipping
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-grad-clip`.
-
-
-Getting The Model Weights Out
-=======================================================================================================================
-
-Moved to :ref:`deepspeed-weight-extraction`.

docs/source/migration.md

@@ -31,7 +31,7 @@ This introduces two breaking changes:
 
 ##### How to obtain the same behavior as v3.x in v4.x
 
-- The pipelines now contain additional features out of the box. See the [token-classification pipeline with the `grouped_entities` flag](https://huggingface.co/transformers/main_classes/pipelines.html?highlight=textclassification#tokenclassificationpipeline).
+- The pipelines now contain additional features out of the box. See the [token-classification pipeline with the `grouped_entities` flag](main_classes/pipelines#transformers.TokenClassificationPipeline).
 - The auto-tokenizers now return rust tokenizers. In order to obtain the python tokenizers instead, the user may use the `use_fast` flag by setting it to `False`:
 
 In version `v3.x`:
@@ -98,7 +98,7 @@ from transformers.models.bert.modeling_bert import BertLayer
 
 #### 4. Switching the `return_dict` argument to `True` by default
 
-The [`return_dict` argument](https://huggingface.co/transformers/main_classes/output.html) enables the return of dict-like python objects containing the model outputs, instead of the standard tuples. This object is self-documented as keys can be used to retrieve values, while also behaving as a tuple as users may retrieve objects by index or by slice.
+The [`return_dict` argument](main_classes/output) enables the return of dict-like python objects containing the model outputs, instead of the standard tuples. This object is self-documented as keys can be used to retrieve values, while also behaving as a tuple as users may retrieve objects by index or by slice.
 
 This is a breaking change as the limitation of that tuple is that it cannot be unpacked: `value0, value1 = outputs` will not work.
 

docs/source/model_doc/albert.mdx

@@ -0,0 +1,170 @@
+<!--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.
+-->
+
+# ALBERT
+
+## Overview
+
+The ALBERT model was proposed in [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. It presents two parameter-reduction techniques to lower memory consumption and increase the training
+speed of BERT:
+
+- Splitting the embedding matrix into two smaller matrices.
+- Using repeating layers split among groups.
+
+The abstract from the paper is the following:
+
+*Increasing model size when pretraining natural language representations often results in improved performance on
+downstream tasks. However, at some point further model increases become harder due to GPU/TPU memory limitations,
+longer training times, and unexpected model degradation. To address these problems, we present two parameter-reduction
+techniques to lower memory consumption and increase the training speed of BERT. Comprehensive empirical evidence shows
+that our proposed methods lead to models that scale much better compared to the original BERT. We also use a
+self-supervised loss that focuses on modeling inter-sentence coherence, and show it consistently helps downstream tasks
+with multi-sentence inputs. As a result, our best model establishes new state-of-the-art results on the GLUE, RACE, and
+SQuAD benchmarks while having fewer parameters compared to BERT-large.*
+
+Tips:
+
+- ALBERT is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather
+  than the left.
+- ALBERT uses repeating layers which results in a small memory footprint, however the computational cost remains
+  similar to a BERT-like architecture with the same number of hidden layers as it has to iterate through the same
+  number of (repeating) layers.
+
+This model was contributed by [lysandre](https://huggingface.co/lysandre). This model jax version was contributed by
+[kamalkraj](https://huggingface.co/kamalkraj). The original code can be found [here](https://github.com/google-research/ALBERT).
+
+## AlbertConfig
+
+[[autodoc]] AlbertConfig
+
+## AlbertTokenizer
+
+[[autodoc]] AlbertTokenizer
+    - build_inputs_with_special_tokens
+    - get_special_tokens_mask
+    - create_token_type_ids_from_sequences
+    - save_vocabulary
+
+## AlbertTokenizerFast
+
+[[autodoc]] AlbertTokenizerFast
+
+## Albert specific outputs
+
+[[autodoc]] models.albert.modeling_albert.AlbertForPreTrainingOutput
+
+[[autodoc]] models.albert.modeling_tf_albert.TFAlbertForPreTrainingOutput
+
+## AlbertModel
+
+[[autodoc]] AlbertModel
+    - forward
+
+## AlbertForPreTraining
+
+[[autodoc]] AlbertForPreTraining
+    - forward
+
+## AlbertForMaskedLM
+
+[[autodoc]] AlbertForMaskedLM
+    - forward
+
+## AlbertForSequenceClassification
+
+[[autodoc]] AlbertForSequenceClassification
+    - forward
+
+## AlbertForMultipleChoice
+
+[[autodoc]] AlbertForMultipleChoice
+
+## AlbertForTokenClassification
+
+[[autodoc]] AlbertForTokenClassification
+    - forward
+
+## AlbertForQuestionAnswering
+
+[[autodoc]] AlbertForQuestionAnswering
+    - forward
+
+## TFAlbertModel
+
+[[autodoc]] TFAlbertModel
+    - call
+
+## TFAlbertForPreTraining
+
+[[autodoc]] TFAlbertForPreTraining
+    - call
+
+## TFAlbertForMaskedLM
+
+[[autodoc]] TFAlbertForMaskedLM
+    - call
+
+## TFAlbertForSequenceClassification
+
+[[autodoc]] TFAlbertForSequenceClassification
+    - call
+
+## TFAlbertForMultipleChoice
+
+[[autodoc]] TFAlbertForMultipleChoice
+    - call
+
+## TFAlbertForTokenClassification
+
+[[autodoc]] TFAlbertForTokenClassification
+    - call
+
+## TFAlbertForQuestionAnswering
+
+[[autodoc]] TFAlbertForQuestionAnswering
+    - call
+
+## FlaxAlbertModel
+
+[[autodoc]] FlaxAlbertModel
+    - __call__
+
+## FlaxAlbertForPreTraining
+
+[[autodoc]] FlaxAlbertForPreTraining
+    - __call__
+
+## FlaxAlbertForMaskedLM
+
+[[autodoc]] FlaxAlbertForMaskedLM
+    - __call__
+
+## FlaxAlbertForSequenceClassification
+
+[[autodoc]] FlaxAlbertForSequenceClassification
+    - __call__
+
+## FlaxAlbertForMultipleChoice
+
+[[autodoc]] FlaxAlbertForMultipleChoice
+    - __call__
+
+## FlaxAlbertForTokenClassification
+
+[[autodoc]] FlaxAlbertForTokenClassification
+    - __call__
+
+## FlaxAlbertForQuestionAnswering
+
+[[autodoc]] FlaxAlbertForQuestionAnswering
+    - __call__

docs/source/model_doc/albert.rst

@@ -1,176 +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.
-
-ALBERT
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The ALBERT model was proposed in `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. It presents two parameter-reduction techniques to lower memory consumption and increase the training
-speed of BERT:
-
-- Splitting the embedding matrix into two smaller matrices.
-- Using repeating layers split among groups.
-
-The abstract from the paper is the following:
-
-*Increasing model size when pretraining natural language representations often results in improved performance on
-downstream tasks. However, at some point further model increases become harder due to GPU/TPU memory limitations,
-longer training times, and unexpected model degradation. To address these problems, we present two parameter-reduction
-techniques to lower memory consumption and increase the training speed of BERT. Comprehensive empirical evidence shows
-that our proposed methods lead to models that scale much better compared to the original BERT. We also use a
-self-supervised loss that focuses on modeling inter-sentence coherence, and show it consistently helps downstream tasks
-with multi-sentence inputs. As a result, our best model establishes new state-of-the-art results on the GLUE, RACE, and
-SQuAD benchmarks while having fewer parameters compared to BERT-large.*
-
-Tips:
-
-- ALBERT is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather
-  than the left.
-- ALBERT uses repeating layers which results in a small memory footprint, however the computational cost remains
-  similar to a BERT-like architecture with the same number of hidden layers as it has to iterate through the same
-  number of (repeating) layers.
-
-This model was contributed by `lysandre <https://huggingface.co/lysandre>`__. The original code can be found `here
-<https://github.com/google-research/ALBERT>`__.
-
-AlbertConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertConfig
-    :members:
-
-
-AlbertTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertTokenizer
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
-
-
-AlbertTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertTokenizerFast
-    :members:
-
-
-Albert specific outputs
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.models.albert.modeling_albert.AlbertForPreTrainingOutput
-    :members:
-
-.. autoclass:: transformers.models.albert.modeling_tf_albert.TFAlbertForPreTrainingOutput
-    :members:
-
-
-AlbertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertModel
-    :members: forward
-
-
-AlbertForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertForPreTraining
-    :members: forward
-
-
-AlbertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertForMaskedLM
-    :members: forward
-
-
-AlbertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertForSequenceClassification
-    :members: forward
-
-
-AlbertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertForMultipleChoice
-    :members:
-
-
-AlbertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertForTokenClassification
-    :members: forward
-
-
-AlbertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.AlbertForQuestionAnswering
-    :members: forward
-
-
-TFAlbertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertModel
-    :members: call
-
-
-TFAlbertForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertForPreTraining
-    :members: call
-
-
-TFAlbertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertForMaskedLM
-    :members: call
-
-
-TFAlbertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertForSequenceClassification
-    :members: call
-
-
-TFAlbertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertForMultipleChoice
-    :members: call
-
-
-TFAlbertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertForTokenClassification
-    :members: call
-
-
-TFAlbertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFAlbertForQuestionAnswering
-    :members: call

docs/source/model_doc/auto.rst

@@ -27,7 +27,32 @@ Instantiating one of :class:`~transformers.AutoConfig`, :class:`~transformers.Au
 
 will create a model that is an instance of :class:`~transformers.BertModel`.
 
-There is one class of :obj:`AutoModel` for each task, and for each backend (PyTorch or TensorFlow).
+There is one class of :obj:`AutoModel` for each task, and for each backend (PyTorch, TensorFlow, or Flax).
+
+Extending the Auto Classes
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Each of the auto classes has a method to be extended with your custom classes. For instance, if you have defined a
+custom class of model :obj:`NewModel`, make sure you have a :obj:`NewModelConfig` then you can add those to the auto
+classes like this:
+
+.. code-block::
+
+    from transformers import AutoConfig, AutoModel
+
+    AutoConfig.register("new-model", NewModelConfig)
+    AutoModel.register(NewModelConfig, NewModel)
+
+You will then be able to use the auto classes like you would usually do!
+
+.. warning::
+
+    If your :obj:`NewModelConfig` is a subclass of :class:`~transformer.PretrainedConfig`, make sure its
+    :obj:`model_type` attribute is set to the same key you use when registering the config (here :obj:`"new-model"`).
+
+    Likewise, if your :obj:`NewModel` is a subclass of :class:`~transformers.PreTrainedModel`, make sure its
+    :obj:`config_class` attribute is set to the same class you use when registering the model (here
+    :obj:`NewModelConfig`).
 
 
 AutoConfig
@@ -51,6 +76,13 @@ AutoFeatureExtractor
     :members:
 
 
+AutoProcessor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoProcessor
+    :members:
+
+
 AutoModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -135,6 +167,62 @@ AutoModelForImageClassification
     :members:
 
 
+AutoModelForVision2Seq
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForVision2Seq
+    :members:
+
+
+AutoModelForAudioClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForAudioClassification
+    :members:
+
+
+AutoModelForAudioFrameClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForAudioFrameClassification
+    :members:
+
+
+AutoModelForCTC
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForCTC
+    :members:
+
+
+AutoModelForSpeechSeq2Seq
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForSpeechSeq2Seq
+    :members:
+
+
+AutoModelForAudioXVector
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForAudioXVector
+    :members:
+
+
+AutoModelForObjectDetection
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForObjectDetection
+    :members:
+
+
+AutoModelForImageSegmentation
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForImageSegmentation
+    :members:
+
+
 TFAutoModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -156,6 +244,13 @@ TFAutoModelForCausalLM
     :members:
 
 
+TFAutoModelForImageClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFAutoModelForImageClassification
+    :members:
+
+
 TFAutoModelForMaskedLM
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -184,6 +279,13 @@ TFAutoModelForMultipleChoice
     :members:
 
 
+TFAutoModelForTableQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFAutoModelForTableQuestionAnswering
+    :members:
+
+
 TFAutoModelForTokenClassification
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -273,3 +375,10 @@ FlaxAutoModelForImageClassification
 
 .. autoclass:: transformers.FlaxAutoModelForImageClassification
     :members:
+
+
+FlaxAutoModelForVision2Seq
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForVision2Seq
+    :members:

docs/source/model_doc/bart.mdx

@@ -0,0 +1,151 @@
+<!--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.
+-->
+
+# BART
+
+**DISCLAIMER:** If you see something strange, file a [Github Issue](https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title) and assign
+@patrickvonplaten
+
+## Overview
+
+The Bart model was proposed in [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 on 29 Oct, 2019.
+
+According to the abstract,
+
+- Bart uses a standard seq2seq/machine translation architecture with a bidirectional encoder (like BERT) and a
+  left-to-right decoder (like GPT).
+- The pretraining task involves randomly shuffling the order of the original sentences and a novel in-filling scheme,
+  where spans of text are replaced with a single mask token.
+- BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It
+  matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new
+  state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains
+  of up to 6 ROUGE.
+
+This model was contributed by [sshleifer](https://huggingface.co/sshleifer). The Authors' code can be found [here](https://github.com/pytorch/fairseq/tree/master/examples/bart).
+
+
+### Examples
+
+- Examples and scripts for fine-tuning BART and other models for sequence to sequence tasks can be found in
+  [examples/pytorch/summarization/](https://github.com/huggingface/transformers/tree/master/examples/pytorch/summarization/README.md).
+- An example of how to train [`BartForConditionalGeneration`] with a Hugging Face `datasets`
+  object can be found in this [forum discussion](https://discuss.huggingface.co/t/train-bart-for-conditional-generation-e-g-summarization/1904).
+- [Distilled checkpoints](https://huggingface.co/models?search=distilbart) are described in this [paper](https://arxiv.org/abs/2010.13002).
+
+
+## Implementation Notes
+
+- Bart doesn't use `token_type_ids` for sequence classification. Use [`BartTokenizer`] or
+  [`~BartTokenizer.encode`] to get the proper splitting.
+- The forward pass of [`BartModel`] will create the `decoder_input_ids` if they are not passed.
+  This is different than some other modeling APIs. A typical use case of this feature is mask filling.
+- Model predictions are intended to be identical to the original implementation when
+  `force_bos_token_to_be_generated=True`. This only works, however, if the string you pass to
+  [`fairseq.encode`] starts with a space.
+- [`~generation_utils.GenerationMixin.generate`] should be used for conditional generation tasks like
+  summarization, see the example in that docstrings.
+- Models that load the *facebook/bart-large-cnn* weights will not have a `mask_token_id`, or be able to perform
+  mask-filling tasks.
+
+## Mask Filling
+
+The `facebook/bart-base` and `facebook/bart-large` checkpoints can be used to fill multi-token masks.
+
+```python
+from transformers import BartForConditionalGeneration, BartTokenizer
+model = BartForConditionalGeneration.from_pretrained("facebook/bart-large", forced_bos_token_id=0)
+tok = BartTokenizer.from_pretrained("facebook/bart-large")
+example_english_phrase = "UN Chief Says There Is No <mask> in Syria"
+batch = tok(example_english_phrase, return_tensors='pt')
+generated_ids = model.generate(batch['input_ids'])
+assert tok.batch_decode(generated_ids, skip_special_tokens=True) == ['UN Chief Says There Is No Plan to Stop Chemical Weapons in Syria']
+```
+
+## BartConfig
+
+[[autodoc]] BartConfig
+    - all
+
+## BartTokenizer
+
+[[autodoc]] BartTokenizer
+    - all
+
+## BartTokenizerFast
+
+[[autodoc]] BartTokenizerFast
+    - all
+
+## BartModel
+
+[[autodoc]] BartModel
+    - forward
+
+## BartForConditionalGeneration
+
+[[autodoc]] BartForConditionalGeneration
+    - forward
+
+## BartForSequenceClassification
+
+[[autodoc]] BartForSequenceClassification
+    - forward
+
+## BartForQuestionAnswering
+
+[[autodoc]] BartForQuestionAnswering
+    - forward
+
+## BartForCausalLM
+
+[[autodoc]] BartForCausalLM
+    - forward
+
+## TFBartModel
+
+[[autodoc]] TFBartModel
+    - call
+
+## TFBartForConditionalGeneration
+
+[[autodoc]] TFBartForConditionalGeneration
+    - call
+
+## FlaxBartModel
+
+[[autodoc]] FlaxBartModel
+    - __call__
+    - encode
+    - decode
+
+## FlaxBartForConditionalGeneration
+
+[[autodoc]] FlaxBartForConditionalGeneration
+    - __call__
+    - encode
+    - decode
+
+## FlaxBartForSequenceClassification
+
+[[autodoc]] FlaxBartForSequenceClassification
+    - __call__
+    - encode
+    - decode
+
+## FlaxBartForQuestionAnswering
+
+[[autodoc]] FlaxBartForQuestionAnswering
+    - __call__
+    - encode
+    - decode

docs/source/model_doc/bart.rst

@@ -1,182 +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.
-
-BART
------------------------------------------------------------------------------------------------------------------------
-
-**DISCLAIMER:** If you see something strange, file a `Github Issue
-<https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title>`__ and assign
-@patrickvonplaten
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The Bart model was proposed in `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 on 29 Oct, 2019.
-
-According to the abstract,
-
-- Bart uses a standard seq2seq/machine translation architecture with a bidirectional encoder (like BERT) and a
-  left-to-right decoder (like GPT).
-- The pretraining task involves randomly shuffling the order of the original sentences and a novel in-filling scheme,
-  where spans of text are replaced with a single mask token.
-- BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It
-  matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new
-  state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains
-  of up to 6 ROUGE.
-
-This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__. The Authors' code can be found `here
-<https://github.com/pytorch/fairseq/tree/master/examples/bart>`__.
-
-
-Examples
-_______________________________________________________________________________________________________________________
-
-- Examples and scripts for fine-tuning BART and other models for sequence to sequence tasks can be found in
-  :prefix_link:`examples/pytorch/summarization/ <examples/pytorch/summarization/README.md>`.
-- An example of how to train :class:`~transformers.BartForConditionalGeneration` with a Hugging Face :obj:`datasets`
-  object can be found in this `forum discussion
-  <https://discuss.huggingface.co/t/train-bart-for-conditional-generation-e-g-summarization/1904>`__.
-- `Distilled checkpoints <https://huggingface.co/models?search=distilbart>`__ are described in this `paper
-  <https://arxiv.org/abs/2010.13002>`__.
-
-
-Implementation Notes
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-- Bart doesn't use :obj:`token_type_ids` for sequence classification. Use :class:`~transformers.BartTokenizer` or
-  :meth:`~transformers.BartTokenizer.encode` to get the proper splitting.
-- The forward pass of :class:`~transformers.BartModel` will create the ``decoder_input_ids`` if they are not passed.
-  This is different than some other modeling APIs. A typical use case of this feature is mask filling.
-- Model predictions are intended to be identical to the original implementation when
-  :obj:`force_bos_token_to_be_generated=True`. This only works, however, if the string you pass to
-  :func:`fairseq.encode` starts with a space.
-- :meth:`~transformers.generation_utils.GenerationMixin.generate` should be used for conditional generation tasks like
-  summarization, see the example in that docstrings.
-- Models that load the `facebook/bart-large-cnn` weights will not have a :obj:`mask_token_id`, or be able to perform
-  mask-filling tasks.
-
-Mask Filling
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The :obj:`facebook/bart-base` and :obj:`facebook/bart-large` checkpoints can be used to fill multi-token masks.
-
-.. code-block::
-
-    from transformers import BartForConditionalGeneration, BartTokenizer
-    model = BartForConditionalGeneration.from_pretrained("facebook/bart-large", force_bos_token_to_be_generated=True)
-    tok = BartTokenizer.from_pretrained("facebook/bart-large")
-    example_english_phrase = "UN Chief Says There Is No <mask> in Syria"
-    batch = tok(example_english_phrase, return_tensors='pt')
-    generated_ids = model.generate(batch['input_ids'])
-    assert tok.batch_decode(generated_ids, skip_special_tokens=True) == ['UN Chief Says There Is No Plan to Stop Chemical Weapons in Syria']
-
-
-
-BartConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartConfig
-    :members:
-
-
-BartTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartTokenizer
-    :members:
-
-
-BartTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartTokenizerFast
-    :members:
-
-
-BartModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartModel
-    :members: forward
-
-
-BartForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartForConditionalGeneration
-    :members: forward
-
-
-BartForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartForSequenceClassification
-    :members: forward
-
-
-BartForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartForQuestionAnswering
-    :members: forward
-
-
-BartForCausalLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BartForCausalLM
-    :members: forward
-
-
-TFBartModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBartModel
-    :members: call
-
-
-TFBartForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBartForConditionalGeneration
-    :members: call
-
-
-FlaxBartModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBartModel
-    :members: __call__, encode, decode
-
-
-FlaxBartForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBartForConditionalGeneration
-    :members: __call__, encode, decode
-
-
-FlaxBartForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBartForSequenceClassification
-    :members: __call__, encode, decode
-
-
-FlaxBartForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBartForQuestionAnswering
-    :members: __call__, encode, decode
-

docs/source/model_doc/barthez.mdx

@@ -0,0 +1,50 @@
+<!--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.
+-->
+
+# BARThez
+
+## Overview
+
+The BARThez model was proposed in [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 on 23 Oct,
+2020.
+
+The abstract of the paper:
+
+
+*Inductive transfer learning, enabled by self-supervised learning, have taken the entire Natural Language Processing
+(NLP) field by storm, with models such as BERT and BART setting new state of the art on countless natural language
+understanding tasks. While there are some notable exceptions, most of the available models and research have been
+conducted for the English language. In this work, we introduce BARThez, the first BART model for the French language
+(to the best of our knowledge). BARThez was pretrained on a very large monolingual French corpus from past research
+that we adapted to suit BART's perturbation schemes. Unlike already existing BERT-based French language models such as
+CamemBERT and FlauBERT, BARThez is particularly well-suited for generative tasks, since not only its encoder but also
+its decoder is pretrained. In addition to discriminative tasks from the FLUE benchmark, we evaluate BARThez on a novel
+summarization dataset, OrangeSum, that we release with this paper. We also continue the pretraining of an already
+pretrained multilingual BART on BARThez's corpus, and we show that the resulting model, which we call mBARTHez,
+provides a significant boost over vanilla BARThez, and is on par with or outperforms CamemBERT and FlauBERT.*
+
+This model was contributed by [moussakam](https://huggingface.co/moussakam). The Authors' code can be found [here](https://github.com/moussaKam/BARThez).
+
+
+### Examples
+
+- BARThez can be fine-tuned on sequence-to-sequence tasks in a similar way as BART, check:
+  [examples/pytorch/summarization/](https://github.com/huggingface/transformers/tree/master/examples/pytorch/summarization/README.md).
+
+
+## BarthezTokenizer
+
+[[autodoc]] BarthezTokenizer
+
+## BarthezTokenizerFast
+
+[[autodoc]] BarthezTokenizerFast

docs/source/model_doc/barthez.rst

@@ -1,60 +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.
-
-BARThez
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BARThez model was proposed in `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 on 23 Oct,
-2020.
-
-The abstract of the paper:
-
-
-*Inductive transfer learning, enabled by self-supervised learning, have taken the entire Natural Language Processing
-(NLP) field by storm, with models such as BERT and BART setting new state of the art on countless natural language
-understanding tasks. While there are some notable exceptions, most of the available models and research have been
-conducted for the English language. In this work, we introduce BARThez, the first BART model for the French language
-(to the best of our knowledge). BARThez was pretrained on a very large monolingual French corpus from past research
-that we adapted to suit BART's perturbation schemes. Unlike already existing BERT-based French language models such as
-CamemBERT and FlauBERT, BARThez is particularly well-suited for generative tasks, since not only its encoder but also
-its decoder is pretrained. In addition to discriminative tasks from the FLUE benchmark, we evaluate BARThez on a novel
-summarization dataset, OrangeSum, that we release with this paper. We also continue the pretraining of an already
-pretrained multilingual BART on BARThez's corpus, and we show that the resulting model, which we call mBARTHez,
-provides a significant boost over vanilla BARThez, and is on par with or outperforms CamemBERT and FlauBERT.*
-
-This model was contributed by `moussakam <https://huggingface.co/moussakam>`__. The Authors' code can be found `here
-<https://github.com/moussaKam/BARThez>`__.
-
-
-Examples
-_______________________________________________________________________________________________________________________
-
-- BARThez can be fine-tuned on sequence-to-sequence tasks in a similar way as BART, check:
-  :prefix_link:`examples/pytorch/summarization/ <examples/pytorch/summarization/README.md>`.
-
-
-BarthezTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BarthezTokenizer
-    :members:
-
-
-BarthezTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BarthezTokenizerFast
-    :members:

docs/source/model_doc/bartpho.mdx

@@ -0,0 +1,80 @@
+<!--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.
+-->
+
+# BARTpho
+
+## Overview
+
+The BARTpho model was proposed in [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.
+
+The abstract from the paper is the following:
+
+*We present BARTpho with two versions -- BARTpho_word and BARTpho_syllable -- the first public large-scale monolingual
+sequence-to-sequence models pre-trained for Vietnamese. Our BARTpho uses the "large" architecture and pre-training
+scheme of the sequence-to-sequence denoising model BART, thus especially suitable for generative NLP tasks. Experiments
+on a downstream task of Vietnamese text summarization show that in both automatic and human evaluations, our BARTpho
+outperforms the strong baseline mBART and improves the state-of-the-art. We release BARTpho to facilitate future
+research and applications of generative Vietnamese NLP tasks.*
+
+Example of use:
+
+```python
+>>> import torch
+>>> from transformers import AutoModel, AutoTokenizer
+
+>>> bartpho = AutoModel.from_pretrained("vinai/bartpho-syllable")
+
+>>> tokenizer = AutoTokenizer.from_pretrained("vinai/bartpho-syllable")
+
+>>> line = "Chúng tôi là những nghiên cứu viên."
+
+>>> input_ids = tokenizer(line, return_tensors="pt")
+
+>>> with torch.no_grad():
+...     features = bartpho(**input_ids)  # Models outputs are now tuples
+
+>>> # With TensorFlow 2.0+:
+>>> from transformers import TFAutoModel
+>>> bartpho = TFAutoModel.from_pretrained("vinai/bartpho-syllable")
+>>> input_ids = tokenizer(line, return_tensors="tf")
+>>> features = bartpho(**input_ids)
+```
+
+Tips:
+
+- Following mBART, BARTpho uses the "large" architecture of BART with an additional layer-normalization layer on top of
+  both the encoder and decoder. Thus, usage examples in the [documentation of BART](bart), when adapting to use
+  with BARTpho, should be adjusted by replacing the BART-specialized classes with the mBART-specialized counterparts.
+  For example:
+
+```python
+>>> from transformers import MBartForConditionalGeneration
+>>> bartpho = MBartForConditionalGeneration.from_pretrained("vinai/bartpho-syllable")
+>>> TXT = 'Chúng tôi là <mask> nghiên cứu viên.'
+>>> input_ids = tokenizer([TXT], return_tensors='pt')['input_ids']
+>>> logits = bartpho(input_ids).logits
+>>> masked_index = (input_ids[0] == tokenizer.mask_token_id).nonzero().item()
+>>> probs = logits[0, masked_index].softmax(dim=0)
+>>> values, predictions = probs.topk(5)
+>>> print(tokenizer.decode(predictions).split())
+```
+
+- This implementation is only for tokenization: "monolingual_vocab_file" consists of Vietnamese-specialized types
+  extracted from the pre-trained SentencePiece model "vocab_file" that is available from the multilingual XLM-RoBERTa.
+  Other languages, if employing this pre-trained multilingual SentencePiece model "vocab_file" for subword
+  segmentation, can reuse BartphoTokenizer with their own language-specialized "monolingual_vocab_file".
+
+This model was contributed by [dqnguyen](https://huggingface.co/dqnguyen). The original code can be found [here](https://github.com/VinAIResearch/BARTpho).
+
+## BartphoTokenizer
+
+[[autodoc]] BartphoTokenizer

docs/source/model_doc/beit.mdx

@@ -0,0 +1,114 @@
+<!--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.
+-->
+
+# BEiT
+
+## Overview
+
+The BEiT model was proposed in [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by
+Hangbo Bao, Li Dong and Furu Wei. Inspired by BERT, BEiT is the first paper that makes self-supervised pre-training of
+Vision Transformers (ViTs) outperform supervised pre-training. Rather than pre-training the model to predict the class
+of an image (as done in the [original ViT paper](https://arxiv.org/abs/2010.11929)), BEiT models are pre-trained to
+predict visual tokens from the codebook of OpenAI's [DALL-E model](https://arxiv.org/abs/2102.12092) given masked
+patches.
+
+The abstract from the paper is the following:
+
+*We introduce a self-supervised vision representation model BEiT, which stands for Bidirectional Encoder representation
+from Image Transformers. Following BERT developed in the natural language processing area, we propose a masked image
+modeling task to pretrain vision Transformers. Specifically, each image has two views in our pre-training, i.e, image
+patches (such as 16x16 pixels), and visual tokens (i.e., discrete tokens). We first "tokenize" the original image into
+visual tokens. Then we randomly mask some image patches and fed them into the backbone Transformer. The pre-training
+objective is to recover the original visual tokens based on the corrupted image patches. After pre-training BEiT, we
+directly fine-tune the model parameters on downstream tasks by appending task layers upon the pretrained encoder.
+Experimental results on image classification and semantic segmentation show that our model achieves competitive results
+with previous pre-training methods. For example, base-size BEiT achieves 83.2% top-1 accuracy on ImageNet-1K,
+significantly outperforming from-scratch DeiT training (81.8%) with the same setup. Moreover, large-size BEiT obtains
+86.3% only using ImageNet-1K, even outperforming ViT-L with supervised pre-training on ImageNet-22K (85.2%).*
+
+Tips:
+
+- BEiT models are regular Vision Transformers, but pre-trained in a self-supervised way rather than supervised. They
+  outperform both the [original model (ViT)](vit) as well as [Data-efficient Image Transformers (DeiT)](deit) when fine-tuned on ImageNet-1K and CIFAR-100. You can check out demo notebooks regarding inference as well as
+  fine-tuning on custom data [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/VisionTransformer) (you can just replace
+  [`ViTFeatureExtractor`] by [`BeitFeatureExtractor`] and
+  [`ViTForImageClassification`] by [`BeitForImageClassification`]).
+- There's also a demo notebook available which showcases how to combine DALL-E's image tokenizer with BEiT for
+  performing masked image modeling. You can find it [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/BEiT).
+- As the BEiT models expect each image to be of the same size (resolution), one can use
+  [`BeitFeatureExtractor`] to resize (or rescale) and normalize images for the model.
+- Both the patch resolution and image resolution used during pre-training or fine-tuning are reflected in the name of
+  each checkpoint. For example, `microsoft/beit-base-patch16-224` refers to a base-sized architecture with patch
+  resolution of 16x16 and fine-tuning resolution of 224x224. All checkpoints can be found on the [hub](https://huggingface.co/models?search=microsoft/beit).
+- The available checkpoints are either (1) pre-trained on [ImageNet-22k](http://www.image-net.org/) (a collection of
+  14 million images and 22k classes) only, (2) also fine-tuned on ImageNet-22k or (3) also fine-tuned on [ImageNet-1k](http://www.image-net.org/challenges/LSVRC/2012/) (also referred to as ILSVRC 2012, a collection of 1.3 million
+  images and 1,000 classes).
+- BEiT uses relative position embeddings, inspired by the T5 model. During pre-training, the authors shared the
+  relative position bias among the several self-attention layers. During fine-tuning, each layer's relative position
+  bias is initialized with the shared relative position bias obtained after pre-training. Note that, if one wants to
+  pre-train a model from scratch, one needs to either set the `use_relative_position_bias` or the
+  `use_relative_position_bias` attribute of [`BeitConfig`] to `True` in order to add
+  position embeddings.
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr). The JAX/FLAX version of this model was
+contributed by [kamalkraj](https://huggingface.co/kamalkraj). The original code can be found [here](https://github.com/microsoft/unilm/tree/master/beit).
+
+
+## BEiT specific outputs
+
+[[autodoc]] models.beit.modeling_beit.BeitModelOutputWithPooling
+
+[[autodoc]] models.beit.modeling_flax_beit.FlaxBeitModelOutputWithPooling
+
+## BeitConfig
+
+[[autodoc]] BeitConfig
+
+## BeitFeatureExtractor
+
+[[autodoc]] BeitFeatureExtractor
+    - __call__
+
+## BeitModel
+
+[[autodoc]] BeitModel
+    - forward
+
+## BeitForMaskedImageModeling
+
+[[autodoc]] BeitForMaskedImageModeling
+    - forward
+
+## BeitForImageClassification
+
+[[autodoc]] BeitForImageClassification
+    - forward
+
+## BeitForSemanticSegmentation
+
+[[autodoc]] BeitForSemanticSegmentation
+    - forward
+
+## FlaxBeitModel
+
+[[autodoc]] FlaxBeitModel
+    - __call__
+
+## FlaxBeitForMaskedImageModeling
+
+[[autodoc]] FlaxBeitForMaskedImageModeling
+    - __call__
+
+## FlaxBeitForImageClassification
+
+[[autodoc]] FlaxBeitForImageClassification
+    - __call__

docs/source/model_doc/bert.mdx

@@ -0,0 +1,197 @@
+<!--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.
+-->
+
+# BERT
+
+## Overview
+
+The BERT model was proposed in [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. It's a
+bidirectional transformer pretrained using a combination of masked language modeling objective and next sentence
+prediction on a large corpus comprising the Toronto Book Corpus and Wikipedia.
+
+The abstract from the paper is the following:
+
+*We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations
+from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional
+representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result,
+the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models
+for a wide range of tasks, such as question answering and language inference, without substantial task-specific
+architecture modifications.*
+
+*BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural
+language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI
+accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute
+improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).*
+
+Tips:
+
+- BERT is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather than
+  the left.
+- BERT was trained with the masked language modeling (MLM) and next sentence prediction (NSP) objectives. It is
+  efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation.
+
+This model was contributed by [thomwolf](https://huggingface.co/thomwolf). The original code can be found [here](https://github.com/google-research/bert).
+
+## BertConfig
+
+[[autodoc]] BertConfig
+    - all
+
+## BertTokenizer
+
+[[autodoc]] BertTokenizer
+    - build_inputs_with_special_tokens
+    - get_special_tokens_mask
+    - create_token_type_ids_from_sequences
+    - save_vocabulary
+
+## BertTokenizerFast
+
+[[autodoc]] BertTokenizerFast
+
+## Bert specific outputs
+
+[[autodoc]] models.bert.modeling_bert.BertForPreTrainingOutput
+
+[[autodoc]] models.bert.modeling_tf_bert.TFBertForPreTrainingOutput
+
+[[autodoc]] models.bert.modeling_flax_bert.FlaxBertForPreTrainingOutput
+
+## BertModel
+
+[[autodoc]] BertModel
+    - forward
+
+## BertForPreTraining
+
+[[autodoc]] BertForPreTraining
+    - forward
+
+## BertLMHeadModel
+
+[[autodoc]] BertLMHeadModel
+    - forward
+
+## BertForMaskedLM
+
+[[autodoc]] BertForMaskedLM
+    - forward
+
+## BertForNextSentencePrediction
+
+[[autodoc]] BertForNextSentencePrediction
+    - forward
+
+## BertForSequenceClassification
+
+[[autodoc]] BertForSequenceClassification
+    - forward
+
+## BertForMultipleChoice
+
+[[autodoc]] BertForMultipleChoice
+    - forward
+
+## BertForTokenClassification
+
+[[autodoc]] BertForTokenClassification
+    - forward
+
+## BertForQuestionAnswering
+
+[[autodoc]] BertForQuestionAnswering
+    - forward
+
+## TFBertModel
+
+[[autodoc]] TFBertModel
+    - call
+
+## TFBertForPreTraining
+
+[[autodoc]] TFBertForPreTraining
+    - call
+
+## TFBertModelLMHeadModel
+
+[[autodoc]] TFBertLMHeadModel
+    - call
+
+## TFBertForMaskedLM
+
+[[autodoc]] TFBertForMaskedLM
+    - call
+
+## TFBertForNextSentencePrediction
+
+[[autodoc]] TFBertForNextSentencePrediction
+    - call
+
+## TFBertForSequenceClassification
+
+[[autodoc]] TFBertForSequenceClassification
+    - call
+
+## TFBertForMultipleChoice
+
+[[autodoc]] TFBertForMultipleChoice
+    - call
+
+## TFBertForTokenClassification
+
+[[autodoc]] TFBertForTokenClassification
+    - call
+
+## TFBertForQuestionAnswering
+
+[[autodoc]] TFBertForQuestionAnswering
+    - call
+
+## FlaxBertModel
+
+[[autodoc]] FlaxBertModel
+    - __call__
+
+## FlaxBertForPreTraining
+
+[[autodoc]] FlaxBertForPreTraining
+    - __call__
+
+## FlaxBertForMaskedLM
+
+[[autodoc]] FlaxBertForMaskedLM
+    - __call__
+
+## FlaxBertForNextSentencePrediction
+
+[[autodoc]] FlaxBertForNextSentencePrediction
+    - __call__
+
+## FlaxBertForSequenceClassification
+
+[[autodoc]] FlaxBertForSequenceClassification
+    - __call__
+
+## FlaxBertForMultipleChoice
+
+[[autodoc]] FlaxBertForMultipleChoice
+    - __call__
+
+## FlaxBertForTokenClassification
+
+[[autodoc]] FlaxBertForTokenClassification
+    - __call__
+
+## FlaxBertForQuestionAnswering
+
+[[autodoc]] FlaxBertForQuestionAnswering
+    - __call__

docs/source/model_doc/bert.rst

@@ -1,259 +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.
-
-BERT
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BERT model was proposed in `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. It's a
-bidirectional transformer pretrained using a combination of masked language modeling objective and next sentence
-prediction on a large corpus comprising the Toronto Book Corpus and Wikipedia.
-
-The abstract from the paper is the following:
-
-*We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations
-from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional
-representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result,
-the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models
-for a wide range of tasks, such as question answering and language inference, without substantial task-specific
-architecture modifications.*
-
-*BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural
-language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI
-accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute
-improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).*
-
-Tips:
-
-- BERT is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather than
-  the left.
-- BERT was trained with the masked language modeling (MLM) and next sentence prediction (NSP) objectives. It is
-  efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation.
-
-This model was contributed by `thomwolf <https://huggingface.co/thomwolf>`__. The original code can be found `here
-<https://github.com/google-research/bert>`__.
-
-BertConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertConfig
-    :members:
-
-
-BertTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertTokenizer
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
-
-
-BertTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertTokenizerFast
-    :members:
-
-
-Bert specific outputs
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.models.bert.modeling_bert.BertForPreTrainingOutput
-    :members:
-
-.. autoclass:: transformers.models.bert.modeling_tf_bert.TFBertForPreTrainingOutput
-    :members:
-
-
-BertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertModel
-    :members: forward
-
-
-BertForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForPreTraining
-    :members: forward
-
-
-BertLMHeadModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertLMHeadModel
-    :members: forward
-
-
-BertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForMaskedLM
-    :members: forward
-
-
-BertForNextSentencePrediction
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForNextSentencePrediction
-    :members: forward
-
-
-BertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForSequenceClassification
-    :members: forward
-
-
-BertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForMultipleChoice
-    :members: forward
-
-
-BertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForTokenClassification
-    :members: forward
-
-
-BertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertForQuestionAnswering
-    :members: forward
-
-
-TFBertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertModel
-    :members: call
-
-
-TFBertForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForPreTraining
-    :members: call
-
-
-TFBertModelLMHeadModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertLMHeadModel
-    :members: call
-
-
-TFBertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForMaskedLM
-    :members: call
-
-
-TFBertForNextSentencePrediction
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForNextSentencePrediction
-    :members: call
-
-
-TFBertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForSequenceClassification
-    :members: call
-
-
-TFBertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForMultipleChoice
-    :members: call
-
-
-TFBertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForTokenClassification
-    :members: call
-
-
-TFBertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBertForQuestionAnswering
-    :members: call
-
-
-FlaxBertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertModel
-    :members: __call__
-
-
-FlaxBertForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForPreTraining
-    :members: __call__
-
-
-FlaxBertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForMaskedLM
-    :members: __call__
-
-
-FlaxBertForNextSentencePrediction
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForNextSentencePrediction
-    :members: __call__
-
-
-FlaxBertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForSequenceClassification
-    :members: __call__
-
-
-FlaxBertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForMultipleChoice
-    :members: __call__
-
-
-FlaxBertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForTokenClassification
-    :members: __call__
-
-
-FlaxBertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBertForQuestionAnswering
-    :members: __call__

docs/source/model_doc/bert_japanese.mdx

@@ -0,0 +1,74 @@
+<!--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.
+-->
+
+# BertJapanese
+
+## Overview
+
+The BERT models trained on Japanese text.
+
+There are models with two different tokenization methods:
+
+- Tokenize with MeCab and WordPiece. This requires some extra dependencies, [fugashi](https://github.com/polm/fugashi) which is a wrapper around [MeCab](https://taku910.github.io/mecab/).
+- Tokenize into characters.
+
+To use *MecabTokenizer*, you should `pip install transformers["ja"]` (or `pip install -e .["ja"]` if you install
+from source) to install dependencies.
+
+See [details on cl-tohoku repository](https://github.com/cl-tohoku/bert-japanese).
+
+Example of using a model with MeCab and WordPiece tokenization:
+
+```python
+>>> import torch
+>>> from transformers import AutoModel, AutoTokenizer 
+
+>>> bertjapanese = AutoModel.from_pretrained("cl-tohoku/bert-base-japanese")
+>>> tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese")
+
+>>> ## Input Japanese Text
+>>> line = "吾輩は猫である。"
+
+>>> inputs = tokenizer(line, return_tensors="pt")
+
+>>> print(tokenizer.decode(inputs['input_ids'][0]))
+[CLS] 吾輩 は 猫 で ある 。 [SEP]
+
+>>> outputs = bertjapanese(**inputs)
+```
+
+Example of using a model with Character tokenization:
+
+```python
+>>> bertjapanese = AutoModel.from_pretrained("cl-tohoku/bert-base-japanese-char")
+>>> tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese-char")
+
+>>> ## Input Japanese Text
+>>> line = "吾輩は猫である。"
+
+>>> inputs = tokenizer(line, return_tensors="pt")
+
+>>> print(tokenizer.decode(inputs['input_ids'][0]))
+[CLS] 吾 輩 は 猫 で あ る 。 [SEP]
+
+>>> outputs = bertjapanese(**inputs)
+```
+
+Tips:
+
+- This implementation is the same as BERT, except for tokenization method. Refer to the [documentation of BERT](bert) for more usage examples.
+
+This model was contributed by [cl-tohoku](https://huggingface.co/cl-tohoku).
+
+## BertJapaneseTokenizer
+
+[[autodoc]] BertJapaneseTokenizer

docs/source/model_doc/bert_japanese.rst

@@ -1,80 +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.
-
-BertJapanese
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BERT models trained on Japanese text.
-
-There are models with two different tokenization methods:
-
-- Tokenize with MeCab and WordPiece. This requires some extra dependencies, `fugashi
-  <https://github.com/polm/fugashi>`__ which is a wrapper around `MeCab <https://taku910.github.io/mecab/>`__.
-- Tokenize into characters.
-
-To use `MecabTokenizer`, you should ``pip install transformers["ja"]`` (or ``pip install -e .["ja"]`` if you install
-from source) to install dependencies.
-
-See `details on cl-tohoku repository <https://github.com/cl-tohoku/bert-japanese>`__.
-
-Example of using a model with MeCab and WordPiece tokenization:
-
-.. code-block::
-
-    >>> import torch
-    >>> from transformers import AutoModel, AutoTokenizer 
-
-    >>> bertjapanese = AutoModel.from_pretrained("cl-tohoku/bert-base-japanese")
-    >>> tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese")
-
-    >>> ## Input Japanese Text
-    >>> line = "吾輩は猫である。"
-
-    >>> inputs = tokenizer(line, return_tensors="pt")
-
-    >>> print(tokenizer.decode(inputs['input_ids'][0]))
-    [CLS] 吾輩 は 猫 で ある 。 [SEP]
-
-    >>> outputs = bertjapanese(**inputs)
-
-Example of using a model with Character tokenization:
-
-.. code-block::
-
-    >>> bertjapanese = AutoModel.from_pretrained("cl-tohoku/bert-base-japanese-char")
-    >>> tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese-char")
-
-    >>> ## Input Japanese Text
-    >>> line = "吾輩は猫である。"
-
-    >>> inputs = tokenizer(line, return_tensors="pt")
-
-    >>> print(tokenizer.decode(inputs['input_ids'][0]))
-    [CLS] 吾 輩 は 猫 で あ る 。 [SEP]
-
-    >>> outputs = bertjapanese(**inputs)
-
-Tips:
-
-- This implementation is the same as BERT, except for tokenization method. Refer to the :doc:`documentation of BERT
-  <bert>` for more usage examples.
-
-This model was contributed by `cl-tohoku <https://huggingface.co/cl-tohoku>`__.
-
-BertJapaneseTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertJapaneseTokenizer
-    :members: 

docs/source/model_doc/bertgeneration.mdx

@@ -0,0 +1,98 @@
+<!--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.
+-->
+
+# BertGeneration
+
+## Overview
+
+The BertGeneration model is a BERT model that can be leveraged for sequence-to-sequence tasks using
+[`EncoderDecoderModel`] as proposed in [Leveraging Pre-trained Checkpoints for Sequence Generation
+Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+
+The abstract from the paper is the following:
+
+*Unsupervised pretraining of large neural models has recently revolutionized Natural Language Processing. By
+warm-starting from the publicly released checkpoints, NLP practitioners have pushed the state-of-the-art on multiple
+benchmarks while saving significant amounts of compute time. So far the focus has been mainly on the Natural Language
+Understanding tasks. In this paper, we demonstrate the efficacy of pre-trained checkpoints for Sequence Generation. We
+developed a Transformer-based sequence-to-sequence model that is compatible with publicly available pre-trained BERT,
+GPT-2 and RoBERTa checkpoints and conducted an extensive empirical study on the utility of initializing our model, both
+encoder and decoder, with these checkpoints. Our models result in new state-of-the-art results on Machine Translation,
+Text Summarization, Sentence Splitting, and Sentence Fusion.*
+
+Usage:
+
+- The model can be used in combination with the [`EncoderDecoderModel`] to leverage two pretrained
+  BERT checkpoints for subsequent fine-tuning.
+
+```python
+>>> # leverage checkpoints for Bert2Bert model...
+>>> # use BERT's cls token as BOS token and sep token as EOS token
+>>> encoder = BertGenerationEncoder.from_pretrained("bert-large-uncased", bos_token_id=101, eos_token_id=102)
+>>> # add cross attention layers and use BERT's cls token as BOS token and sep token as EOS token
+>>> decoder = BertGenerationDecoder.from_pretrained("bert-large-uncased", add_cross_attention=True, is_decoder=True, bos_token_id=101, eos_token_id=102)
+>>> bert2bert = EncoderDecoderModel(encoder=encoder, decoder=decoder)
+
+>>> # create tokenizer...
+>>> tokenizer = BertTokenizer.from_pretrained("bert-large-uncased")
+
+>>> input_ids = tokenizer('This is a long article to summarize', add_special_tokens=False, return_tensors="pt").input_ids
+>>> labels = tokenizer('This is a short summary', return_tensors="pt").input_ids
+
+>>> # train...
+>>> loss = bert2bert(input_ids=input_ids, decoder_input_ids=labels, labels=labels).loss
+>>> loss.backward()
+```
+
+- Pretrained [`EncoderDecoderModel`] are also directly available in the model hub, e.g.,
+
+
+```python
+>>> # instantiate sentence fusion model
+>>> sentence_fuser = EncoderDecoderModel.from_pretrained("google/roberta2roberta_L-24_discofuse")
+>>> tokenizer = AutoTokenizer.from_pretrained("google/roberta2roberta_L-24_discofuse")
+
+>>> input_ids = tokenizer('This is the first sentence. This is the second sentence.', add_special_tokens=False, return_tensors="pt").input_ids
+
+>>> outputs = sentence_fuser.generate(input_ids)
+
+>>> print(tokenizer.decode(outputs[0]))
+```
+
+Tips:
+
+- [`BertGenerationEncoder`] and [`BertGenerationDecoder`] should be used in
+  combination with [`EncoderDecoder`].
+- For summarization, sentence splitting, sentence fusion and translation, no special tokens are required for the input.
+  Therefore, no EOS token should be added to the end of the input.
+
+This model was contributed by [patrickvonplaten](https://huggingface.co/patrickvonplaten). The original code can be
+found [here](https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder).
+
+## BertGenerationConfig
+
+[[autodoc]] BertGenerationConfig
+
+## BertGenerationTokenizer
+
+[[autodoc]] BertGenerationTokenizer
+    - save_vocabulary
+
+## BertGenerationEncoder
+
+[[autodoc]] BertGenerationEncoder
+    - forward
+
+## BertGenerationDecoder
+
+[[autodoc]] BertGenerationDecoder
+    - forward

docs/source/model_doc/bertgeneration.rst

@@ -1,109 +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.
-
-BertGeneration
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BertGeneration model is a BERT model that can be leveraged for sequence-to-sequence tasks using
-:class:`~transformers.EncoderDecoderModel` as proposed in `Leveraging Pre-trained Checkpoints for Sequence Generation
-Tasks <https://arxiv.org/abs/1907.12461>`__ by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-
-The abstract from the paper is the following:
-
-*Unsupervised pretraining of large neural models has recently revolutionized Natural Language Processing. By
-warm-starting from the publicly released checkpoints, NLP practitioners have pushed the state-of-the-art on multiple
-benchmarks while saving significant amounts of compute time. So far the focus has been mainly on the Natural Language
-Understanding tasks. In this paper, we demonstrate the efficacy of pre-trained checkpoints for Sequence Generation. We
-developed a Transformer-based sequence-to-sequence model that is compatible with publicly available pre-trained BERT,
-GPT-2 and RoBERTa checkpoints and conducted an extensive empirical study on the utility of initializing our model, both
-encoder and decoder, with these checkpoints. Our models result in new state-of-the-art results on Machine Translation,
-Text Summarization, Sentence Splitting, and Sentence Fusion.*
-
-Usage:
-
-- The model can be used in combination with the :class:`~transformers.EncoderDecoderModel` to leverage two pretrained
-  BERT checkpoints for subsequent fine-tuning.
-
-.. code-block::
-
-    >>> # leverage checkpoints for Bert2Bert model...
-    >>> # use BERT's cls token as BOS token and sep token as EOS token
-    >>> encoder = BertGenerationEncoder.from_pretrained("bert-large-uncased", bos_token_id=101, eos_token_id=102)
-    >>> # add cross attention layers and use BERT's cls token as BOS token and sep token as EOS token
-    >>> decoder = BertGenerationDecoder.from_pretrained("bert-large-uncased", add_cross_attention=True, is_decoder=True, bos_token_id=101, eos_token_id=102)
-    >>> bert2bert = EncoderDecoderModel(encoder=encoder, decoder=decoder)
-
-    >>> # create tokenizer...
-    >>> tokenizer = BertTokenizer.from_pretrained("bert-large-uncased")
-
-    >>> input_ids = tokenizer('This is a long article to summarize', add_special_tokens=False, return_tensors="pt").input_ids
-    >>> labels = tokenizer('This is a short summary', return_tensors="pt").input_ids
-
-    >>> # train...
-    >>> loss = bert2bert(input_ids=input_ids, decoder_input_ids=labels, labels=labels).loss
-    >>> loss.backward()
-
-
-- Pretrained :class:`~transformers.EncoderDecoderModel` are also directly available in the model hub, e.g.,
-
-
-.. code-block::
-
-    >>> # instantiate sentence fusion model
-    >>> sentence_fuser = EncoderDecoderModel.from_pretrained("google/roberta2roberta_L-24_discofuse")
-    >>> tokenizer = AutoTokenizer.from_pretrained("google/roberta2roberta_L-24_discofuse")
-
-    >>> input_ids = tokenizer('This is the first sentence. This is the second sentence.', add_special_tokens=False, return_tensors="pt").input_ids
-
-    >>> outputs = sentence_fuser.generate(input_ids)
-
-    >>> print(tokenizer.decode(outputs[0]))
-
-
-Tips:
-
-- :class:`~transformers.BertGenerationEncoder` and :class:`~transformers.BertGenerationDecoder` should be used in
-  combination with :class:`~transformers.EncoderDecoder`.
-- For summarization, sentence splitting, sentence fusion and translation, no special tokens are required for the input.
-  Therefore, no EOS token should be added to the end of the input.
-
-This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__. The original code can be
-found `here <https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder>`__.
-
-BertGenerationConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertGenerationConfig
-    :members:
-
-
-BertGenerationTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertGenerationTokenizer
-    :members: save_vocabulary
-
-BertGenerationEncoder
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertGenerationEncoder
-    :members: forward
-
-
-BertGenerationDecoder
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertGenerationDecoder
-    :members: forward

docs/source/model_doc/bertweet.mdx

@@ -0,0 +1,58 @@
+<!--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.
+-->
+
+# BERTweet
+
+## Overview
+
+The BERTweet model was proposed in [BERTweet: A pre-trained language model for English Tweets](https://www.aclweb.org/anthology/2020.emnlp-demos.2.pdf) by Dat Quoc Nguyen, Thanh Vu, Anh Tuan Nguyen.
+
+The abstract from the paper is the following:
+
+*We present BERTweet, the first public large-scale pre-trained language model for English Tweets. Our BERTweet, having
+the same architecture as BERT-base (Devlin et al., 2019), is trained using the RoBERTa pre-training procedure (Liu et
+al., 2019). Experiments show that BERTweet outperforms strong baselines RoBERTa-base and XLM-R-base (Conneau et al.,
+2020), producing better performance results than the previous state-of-the-art models on three Tweet NLP tasks:
+Part-of-speech tagging, Named-entity recognition and text classification.*
+
+Example of use:
+
+```python
+>>> import torch
+>>> from transformers import AutoModel, AutoTokenizer 
+
+>>> bertweet = AutoModel.from_pretrained("vinai/bertweet-base")
+
+>>> # For transformers v4.x+: 
+>>> tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base", use_fast=False)
+
+>>> # For transformers v3.x: 
+>>> # tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base")
+
+>>> # INPUT TWEET IS ALREADY NORMALIZED!
+>>> line = "SC has first two presumptive cases of coronavirus , DHEC confirms HTTPURL via @USER :cry:"
+
+>>> input_ids = torch.tensor([tokenizer.encode(line)])
+
+>>> with torch.no_grad():
+...     features = bertweet(input_ids)  # Models outputs are now tuples
+
+>>> # With TensorFlow 2.0+:
+>>> # from transformers import TFAutoModel
+>>> # bertweet = TFAutoModel.from_pretrained("vinai/bertweet-base")
+```
+
+This model was contributed by [dqnguyen](https://huggingface.co/dqnguyen). The original code can be found [here](https://github.com/VinAIResearch/BERTweet).
+
+## BertweetTokenizer
+
+[[autodoc]] BertweetTokenizer

docs/source/model_doc/bertweet.rst

@@ -1,64 +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.
-
-Bertweet
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BERTweet model was proposed in `BERTweet: A pre-trained language model for English Tweets
-<https://www.aclweb.org/anthology/2020.emnlp-demos.2.pdf>`__ by Dat Quoc Nguyen, Thanh Vu, Anh Tuan Nguyen.
-
-The abstract from the paper is the following:
-
-*We present BERTweet, the first public large-scale pre-trained language model for English Tweets. Our BERTweet, having
-the same architecture as BERT-base (Devlin et al., 2019), is trained using the RoBERTa pre-training procedure (Liu et
-al., 2019). Experiments show that BERTweet outperforms strong baselines RoBERTa-base and XLM-R-base (Conneau et al.,
-2020), producing better performance results than the previous state-of-the-art models on three Tweet NLP tasks:
-Part-of-speech tagging, Named-entity recognition and text classification.*
-
-Example of use:
-
-.. code-block::
-
-    >>> import torch
-    >>> from transformers import AutoModel, AutoTokenizer 
-
-    >>> bertweet = AutoModel.from_pretrained("vinai/bertweet-base")
-
-    >>> # For transformers v4.x+: 
-    >>> tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base", use_fast=False)
-
-    >>> # For transformers v3.x: 
-    >>> # tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base")
-
-    >>> # INPUT TWEET IS ALREADY NORMALIZED!
-    >>> line = "SC has first two presumptive cases of coronavirus , DHEC confirms HTTPURL via @USER :cry:"
-
-    >>> input_ids = torch.tensor([tokenizer.encode(line)])
-
-    >>> with torch.no_grad():
-    ...     features = bertweet(input_ids)  # Models outputs are now tuples
-
-    >>> # With TensorFlow 2.0+:
-    >>> # from transformers import TFAutoModel
-    >>> # bertweet = TFAutoModel.from_pretrained("vinai/bertweet-base")
-
-This model was contributed by `dqnguyen <https://huggingface.co/dqnguyen>`__. The original code can be found `here
-<https://github.com/VinAIResearch/BERTweet>`__.
-
-BertweetTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BertweetTokenizer
-    :members: 

docs/source/model_doc/bigbird.mdx

@@ -0,0 +1,146 @@
+<!--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.
+-->
+
+# BigBird
+
+## Overview
+
+The BigBird model was proposed in [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by
+Zaheer, Manzil and Guruganesh, Guru and Dubey, Kumar Avinava and Ainslie, Joshua and Alberti, Chris and Ontanon,
+Santiago and Pham, Philip and Ravula, Anirudh and Wang, Qifan and Yang, Li and others. BigBird, is a sparse-attention
+based transformer which extends Transformer based models, such as BERT to much longer sequences. In addition to sparse
+attention, BigBird also applies global attention as well as random attention to the input sequence. Theoretically, it
+has been shown that applying sparse, global, and random attention approximates full attention, while being
+computationally much more efficient for longer sequences. As a consequence of the capability to handle longer context,
+BigBird has shown improved performance on various long document NLP tasks, such as question answering and
+summarization, compared to BERT or RoBERTa.
+
+The abstract from the paper is the following:
+
+*Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP.
+Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence
+length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that
+reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and
+is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our
+theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire
+sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to
+8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context,
+BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also
+propose novel applications to genomics data.*
+
+Tips:
+
+- For an in-detail explanation on how BigBird's attention works, see [this blog post](https://huggingface.co/blog/big-bird).
+- BigBird comes with 2 implementations: **original_full** & **block_sparse**. For the sequence length < 1024, using
+  **original_full** is advised as there is no benefit in using **block_sparse** attention.
+- The code currently uses window size of 3 blocks and 2 global blocks.
+- Sequence length must be divisible by block size.
+- Current implementation supports only **ITC**.
+- Current implementation doesn't support **num_random_blocks = 0**
+
+This model was contributed by [vasudevgupta](https://huggingface.co/vasudevgupta). The original code can be found
+[here](https://github.com/google-research/bigbird).
+
+## BigBirdConfig
+
+[[autodoc]] BigBirdConfig
+
+## BigBirdTokenizer
+
+[[autodoc]] BigBirdTokenizer
+    - build_inputs_with_special_tokens
+    - get_special_tokens_mask
+    - create_token_type_ids_from_sequences
+    - save_vocabulary
+
+## BigBirdTokenizerFast
+
+[[autodoc]] BigBirdTokenizerFast
+
+## BigBird specific outputs
+
+[[autodoc]] models.big_bird.modeling_big_bird.BigBirdForPreTrainingOutput
+
+## BigBirdModel
+
+[[autodoc]] BigBirdModel
+    - forward
+
+## BigBirdForPreTraining
+
+[[autodoc]] BigBirdForPreTraining
+    - forward
+
+## BigBirdForCausalLM
+
+[[autodoc]] BigBirdForCausalLM
+    - forward
+
+## BigBirdForMaskedLM
+
+[[autodoc]] BigBirdForMaskedLM
+    - forward
+
+## BigBirdForSequenceClassification
+
+[[autodoc]] BigBirdForSequenceClassification
+    - forward
+
+## BigBirdForMultipleChoice
+
+[[autodoc]] BigBirdForMultipleChoice
+    - forward
+
+## BigBirdForTokenClassification
+
+[[autodoc]] BigBirdForTokenClassification
+    - forward
+
+## BigBirdForQuestionAnswering
+
+[[autodoc]] BigBirdForQuestionAnswering
+    - forward
+
+## FlaxBigBirdModel
+
+[[autodoc]] FlaxBigBirdModel
+    - __call__
+
+## FlaxBigBirdForPreTraining
+
+[[autodoc]] FlaxBigBirdForPreTraining
+    - __call__
+
+## FlaxBigBirdForMaskedLM
+
+[[autodoc]] FlaxBigBirdForMaskedLM
+    - __call__
+
+## FlaxBigBirdForSequenceClassification
+
+[[autodoc]] FlaxBigBirdForSequenceClassification
+    - __call__
+
+## FlaxBigBirdForMultipleChoice
+
+[[autodoc]] FlaxBigBirdForMultipleChoice
+    - __call__
+
+## FlaxBigBirdForTokenClassification
+
+[[autodoc]] FlaxBigBirdForTokenClassification
+    - __call__
+
+## FlaxBigBirdForQuestionAnswering
+
+[[autodoc]] FlaxBigBirdForQuestionAnswering
+    - __call__

docs/source/model_doc/bigbird.rst

@@ -1,185 +0,0 @@
-.. 
-    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.
-
-BigBird
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BigBird model was proposed in `Big Bird: Transformers for Longer Sequences <https://arxiv.org/abs/2007.14062>`__ by
-Zaheer, Manzil and Guruganesh, Guru and Dubey, Kumar Avinava and Ainslie, Joshua and Alberti, Chris and Ontanon,
-Santiago and Pham, Philip and Ravula, Anirudh and Wang, Qifan and Yang, Li and others. BigBird, is a sparse-attention
-based transformer which extends Transformer based models, such as BERT to much longer sequences. In addition to sparse
-attention, BigBird also applies global attention as well as random attention to the input sequence. Theoretically, it
-has been shown that applying sparse, global, and random attention approximates full attention, while being
-computationally much more efficient for longer sequences. As a consequence of the capability to handle longer context,
-BigBird has shown improved performance on various long document NLP tasks, such as question answering and
-summarization, compared to BERT or RoBERTa.
-
-The abstract from the paper is the following:
-
-*Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP.
-Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence
-length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that
-reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and
-is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our
-theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire
-sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to
-8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context,
-BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also
-propose novel applications to genomics data.*
-
-Tips:
-
-- For an in-detail explanation on how BigBird's attention works, see `this blog post
-  <https://huggingface.co/blog/big-bird>`__.
-- BigBird comes with 2 implementations: **original_full** & **block_sparse**. For the sequence length < 1024, using
-  **original_full** is advised as there is no benefit in using **block_sparse** attention.
-- The code currently uses window size of 3 blocks and 2 global blocks.
-- Sequence length must be divisible by block size.
-- Current implementation supports only **ITC**.
-- Current implementation doesn't support **num_random_blocks = 0**
-
-This model was contributed by `vasudevgupta <https://huggingface.co/vasudevgupta>`__. The original code can be found
-`here <https://github.com/google-research/bigbird>`__.
-
-BigBirdConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdConfig
-    :members:
-
-
-BigBirdTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdTokenizer
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
-
-BigBirdTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdTokenizerFast
-    :members:
-
-BigBird specific outputs
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.models.big_bird.modeling_big_bird.BigBirdForPreTrainingOutput
-    :members:
-
-
-BigBirdModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdModel
-    :members: forward
-
-
-BigBirdForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForPreTraining
-    :members: forward
-
-
-BigBirdForCausalLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForCausalLM
-    :members: forward
-
-
-BigBirdForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForMaskedLM
-    :members: forward
-
-
-BigBirdForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForSequenceClassification
-    :members: forward
-
-
-BigBirdForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForMultipleChoice
-    :members: forward
-
-
-BigBirdForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForTokenClassification
-    :members: forward
-
-
-BigBirdForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdForQuestionAnswering
-    :members: forward
-
-
-FlaxBigBirdModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdModel
-    :members: __call__
-
-
-FlaxBigBirdForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdForPreTraining
-    :members: __call__
-
-
-FlaxBigBirdForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdForMaskedLM
-    :members: __call__
-
-
-FlaxBigBirdForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdForSequenceClassification
-    :members: __call__
-
-
-FlaxBigBirdForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdForMultipleChoice
-    :members: __call__
-
-
-FlaxBigBirdForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdForTokenClassification
-    :members: __call__
-
-
-FlaxBigBirdForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxBigBirdForQuestionAnswering
-    :members: __call__

docs/source/model_doc/bigbird_pegasus.mdx

@@ -0,0 +1,81 @@
+<!--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.
+-->
+
+# BigBirdPegasus
+
+## Overview
+
+The BigBird model was proposed in [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by
+Zaheer, Manzil and Guruganesh, Guru and Dubey, Kumar Avinava and Ainslie, Joshua and Alberti, Chris and Ontanon,
+Santiago and Pham, Philip and Ravula, Anirudh and Wang, Qifan and Yang, Li and others. BigBird, is a sparse-attention
+based transformer which extends Transformer based models, such as BERT to much longer sequences. In addition to sparse
+attention, BigBird also applies global attention as well as random attention to the input sequence. Theoretically, it
+has been shown that applying sparse, global, and random attention approximates full attention, while being
+computationally much more efficient for longer sequences. As a consequence of the capability to handle longer context,
+BigBird has shown improved performance on various long document NLP tasks, such as question answering and
+summarization, compared to BERT or RoBERTa.
+
+The abstract from the paper is the following:
+
+*Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP.
+Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence
+length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that
+reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and
+is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our
+theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire
+sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to
+8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context,
+BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also
+propose novel applications to genomics data.*
+
+Tips:
+
+- For an in-detail explanation on how BigBird's attention works, see [this blog post](https://huggingface.co/blog/big-bird).
+- BigBird comes with 2 implementations: **original_full** & **block_sparse**. For the sequence length < 1024, using
+  **original_full** is advised as there is no benefit in using **block_sparse** attention.
+- The code currently uses window size of 3 blocks and 2 global blocks.
+- Sequence length must be divisible by block size.
+- Current implementation supports only **ITC**.
+- Current implementation doesn't support **num_random_blocks = 0**.
+- BigBirdPegasus uses the [PegasusTokenizer](https://github.com/huggingface/transformers/blob/master/src/transformers/models/pegasus/tokenization_pegasus.py).
+
+The original code can be found [here](https://github.com/google-research/bigbird).
+
+## BigBirdPegasusConfig
+
+[[autodoc]] BigBirdPegasusConfig
+    - all
+
+## BigBirdPegasusModel
+
+[[autodoc]] BigBirdPegasusModel
+    - forward
+
+## BigBirdPegasusForConditionalGeneration
+
+[[autodoc]] BigBirdPegasusForConditionalGeneration
+    - forward
+
+## BigBirdPegasusForSequenceClassification
+
+[[autodoc]] BigBirdPegasusForSequenceClassification
+    - forward
+
+## BigBirdPegasusForQuestionAnswering
+
+[[autodoc]] BigBirdPegasusForQuestionAnswering
+    - forward
+
+## BigBirdPegasusForCausalLM
+
+[[autodoc]] BigBirdPegasusForCausalLM
+    - forward

docs/source/model_doc/bigbird_pegasus.rst

@@ -1,98 +0,0 @@
-.. 
-    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.
-
-BigBirdPegasus
------------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BigBird model was proposed in `Big Bird: Transformers for Longer Sequences <https://arxiv.org/abs/2007.14062>`__ by
-Zaheer, Manzil and Guruganesh, Guru and Dubey, Kumar Avinava and Ainslie, Joshua and Alberti, Chris and Ontanon,
-Santiago and Pham, Philip and Ravula, Anirudh and Wang, Qifan and Yang, Li and others. BigBird, is a sparse-attention
-based transformer which extends Transformer based models, such as BERT to much longer sequences. In addition to sparse
-attention, BigBird also applies global attention as well as random attention to the input sequence. Theoretically, it
-has been shown that applying sparse, global, and random attention approximates full attention, while being
-computationally much more efficient for longer sequences. As a consequence of the capability to handle longer context,
-BigBird has shown improved performance on various long document NLP tasks, such as question answering and
-summarization, compared to BERT or RoBERTa.
-
-The abstract from the paper is the following:
-
-*Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP.
-Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence
-length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that
-reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and
-is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our
-theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire
-sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to
-8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context,
-BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also
-propose novel applications to genomics data.*
-
-Tips:
-
-- For an in-detail explanation on how BigBird's attention works, see `this blog post
-  <https://huggingface.co/blog/big-bird>`__.
-- BigBird comes with 2 implementations: **original_full** & **block_sparse**. For the sequence length < 1024, using
-  **original_full** is advised as there is no benefit in using **block_sparse** attention.
-- The code currently uses window size of 3 blocks and 2 global blocks.
-- Sequence length must be divisible by block size.
-- Current implementation supports only **ITC**.
-- Current implementation doesn't support **num_random_blocks = 0**.
-- BigBirdPegasus uses the `PegasusTokenizer
-  <https://github.com/huggingface/transformers/blob/master/src/transformers/models/pegasus/tokenization_pegasus.py>`__.
-
-The original code can be found `here <https://github.com/google-research/bigbird>`__.
-
-BigBirdPegasusConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdPegasusConfig
-    :members:
-
-
-BigBirdPegasusModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdPegasusModel
-    :members: forward
-
-
-BigBirdPegasusForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdPegasusForConditionalGeneration
-    :members: forward
-
-
-BigBirdPegasusForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdPegasusForSequenceClassification
-    :members: forward
-
-
-BigBirdPegasusForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdPegasusForQuestionAnswering
-    :members: forward
-
-
-BigBirdPegasusForCausalLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BigBirdPegasusForCausalLM
-    :members: forward
-
-

docs/source/model_doc/blenderbot.mdx

@@ -0,0 +1,118 @@
+<!--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.
+-->
+
+# Blenderbot
+
+**DISCLAIMER:** If you see something strange, file a [Github Issue](https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title) .
+
+## Overview
+
+The Blender chatbot model was proposed in [Recipes for building an open-domain chatbot](https://arxiv.org/pdf/2004.13637.pdf) 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 on 30 Apr 2020.
+
+The abstract of the paper is the following:
+
+*Building open-domain chatbots is a challenging area for machine learning research. While prior work has shown that
+scaling neural models in the number of parameters and the size of the data they are trained on gives improved results,
+we show that other ingredients are important for a high-performing chatbot. Good conversation requires a number of
+skills that an expert conversationalist blends in a seamless way: providing engaging talking points and listening to
+their partners, and displaying knowledge, empathy and personality appropriately, while maintaining a consistent
+persona. We show that large scale models can learn these skills when given appropriate training data and choice of
+generation strategy. We build variants of these recipes with 90M, 2.7B and 9.4B parameter models, and make our models
+and code publicly available. Human evaluations show our best models are superior to existing approaches in multi-turn
+dialogue in terms of engagingness and humanness measurements. We then discuss the limitations of this work by analyzing
+failure cases of our models.*
+
+This model was contributed by [sshleifer](https://huggingface.co/sshleifer). The authors' code can be found [here](https://github.com/facebookresearch/ParlAI) .
+
+
+## Implementation Notes
+
+- Blenderbot uses a standard [seq2seq model transformer](https://arxiv.org/pdf/1706.03762.pdf) based architecture.
+- Available checkpoints can be found in the [model hub](https://huggingface.co/models?search=blenderbot).
+- This is the *default* Blenderbot model class. However, some smaller checkpoints, such as
+  `facebook/blenderbot_small_90M`, have a different architecture and consequently should be used with
+  [BlenderbotSmall](blenderbot_small).
+
+
+## Usage
+
+Here is an example of model usage:
+
+```python
+>>> from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration
+>>> mname = 'facebook/blenderbot-400M-distill'
+>>> model = BlenderbotForConditionalGeneration.from_pretrained(mname)
+>>> tokenizer = BlenderbotTokenizer.from_pretrained(mname)
+>>> UTTERANCE = "My friends are cool but they eat too many carbs."
+>>> inputs = tokenizer([UTTERANCE], return_tensors='pt')
+>>> reply_ids = model.generate(**inputs)
+>>> print(tokenizer.batch_decode(reply_ids))
+["<s> That's unfortunate. Are they trying to lose weight or are they just trying to be healthier?</s>"]
+```
+
+## BlenderbotConfig
+
+[[autodoc]] BlenderbotConfig
+
+## BlenderbotTokenizer
+
+[[autodoc]] BlenderbotTokenizer
+    - build_inputs_with_special_tokens
+
+## BlenderbotTokenizerFast
+
+[[autodoc]] BlenderbotTokenizerFast
+    - build_inputs_with_special_tokens
+
+## BlenderbotModel
+
+See `transformers.BartModel` for arguments to *forward* and *generate*
+
+[[autodoc]] BlenderbotModel
+    - forward
+
+## BlenderbotForConditionalGeneration
+
+See [`~transformers.BartForConditionalGeneration`] for arguments to *forward* and *generate*
+
+[[autodoc]] BlenderbotForConditionalGeneration
+    - forward
+
+## BlenderbotForCausalLM
+
+[[autodoc]] BlenderbotForCausalLM
+    - forward
+
+## TFBlenderbotModel
+
+[[autodoc]] TFBlenderbotModel
+    - call
+
+## TFBlenderbotForConditionalGeneration
+
+[[autodoc]] TFBlenderbotForConditionalGeneration
+    - call
+
+## FlaxBlenderbotModel
+
+[[autodoc]] FlaxBlenderbotModel
+    - __call__
+    - encode
+    - decode
+
+## FlaxBlenderbotForConditionalGeneration
+
+[[autodoc]] FlaxBlenderbotForConditionalGeneration
+    - __call__
+    - encode
+    - decode

docs/source/model_doc/blenderbot.rst

@@ -1,120 +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.
-
-Blenderbot
------------------------------------------------------------------------------------------------------------------------
-
-**DISCLAIMER:** If you see something strange, file a `Github Issue
-<https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title>`__ .
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The Blender chatbot model was proposed in `Recipes for building an open-domain chatbot
-<https://arxiv.org/pdf/2004.13637.pdf>`__ 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 on 30 Apr 2020.
-
-The abstract of the paper is the following:
-
-*Building open-domain chatbots is a challenging area for machine learning research. While prior work has shown that
-scaling neural models in the number of parameters and the size of the data they are trained on gives improved results,
-we show that other ingredients are important for a high-performing chatbot. Good conversation requires a number of
-skills that an expert conversationalist blends in a seamless way: providing engaging talking points and listening to
-their partners, and displaying knowledge, empathy and personality appropriately, while maintaining a consistent
-persona. We show that large scale models can learn these skills when given appropriate training data and choice of
-generation strategy. We build variants of these recipes with 90M, 2.7B and 9.4B parameter models, and make our models
-and code publicly available. Human evaluations show our best models are superior to existing approaches in multi-turn
-dialogue in terms of engagingness and humanness measurements. We then discuss the limitations of this work by analyzing
-failure cases of our models.*
-
-This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__. The authors' code can be found `here
-<https://github.com/facebookresearch/ParlAI>`__ .
-
-
-Implementation Notes
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-- Blenderbot uses a standard `seq2seq model transformer <https://arxiv.org/pdf/1706.03762.pdf>`__ based architecture.
-- Available checkpoints can be found in the `model hub <https://huggingface.co/models?search=blenderbot>`__.
-- This is the `default` Blenderbot model class. However, some smaller checkpoints, such as
-  ``facebook/blenderbot_small_90M``, have a different architecture and consequently should be used with
-  `BlenderbotSmall <https://huggingface.co/transformers/master/model_doc/blenderbot_small.html>`__.
-
-
-Usage
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-Here is an example of model usage:
-
-.. code-block::
-
-        >>> from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration
-        >>> mname = 'facebook/blenderbot-400M-distill'
-        >>> model = BlenderbotForConditionalGeneration.from_pretrained(mname)
-        >>> tokenizer = BlenderbotTokenizer.from_pretrained(mname)
-        >>> UTTERANCE = "My friends are cool but they eat too many carbs."
-        >>> inputs = tokenizer([UTTERANCE], return_tensors='pt')
-        >>> reply_ids = model.generate(**inputs)
-        >>> print(tokenizer.batch_decode(reply_ids))
-        ["<s> That's unfortunate. Are they trying to lose weight or are they just trying to be healthier?</s>"]
-
-
-BlenderbotConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BlenderbotConfig
-    :members:
-
-BlenderbotTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BlenderbotTokenizer
-    :members: build_inputs_with_special_tokens
-
-
-BlenderbotModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-See :obj:`transformers.BartModel` for arguments to `forward` and `generate`
-
-.. autoclass:: transformers.BlenderbotModel
-    :members: forward
-
-
-BlenderbotForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-See :obj:`transformers.BartForConditionalGeneration` for arguments to `forward` and `generate`
-
-.. autoclass:: transformers.BlenderbotForConditionalGeneration
-    :members: forward
-
-
-BlenderbotForCausalLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BlenderbotForCausalLM
-    :members: forward
-
-
-TFBlenderbotModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBlenderbotModel
-    :members: call
-
-
-TFBlenderbotForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFBlenderbotForConditionalGeneration
-    :members: call