v1

* Create index.md

* Create _toctree.yml

* Updated index.md in telugu

* Update _toctree.yml

* Create quicktour.md

* Update quicktour.md

* Create index.md

* Update quicktour.md

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Delete docs/source/hi/index.md

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update docs/source/te/quicktour.md

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Update build_documentation.yml

Added telugu [te]

* Update build_pr_documentation.yml

Added Telugu [te]

* Update _toctree.yml

---------

Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

.circleci/config.yml

@@ -209,6 +209,7 @@ jobs:
             - run: make deps_table_check_updated
             - run: python utils/update_metadata.py --check-only
             - run: python utils/check_task_guides.py
+            - run: python utils/check_docstrings.py
 
 workflows:
     version: 2

.circleci/create_circleci_config.py

@@ -32,7 +32,8 @@ COMMON_ENV_VARIABLES = {
     "RUN_PT_TF_CROSS_TESTS": False,
     "RUN_PT_FLAX_CROSS_TESTS": False,
 }
-COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "dist": "loadfile", "s": None}
+# Disable the use of {"s": None} as the output is way too long, causing the navigation on CircleCI impractical
+COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "dist": "loadfile"}
 DEFAULT_DOCKER_IMAGE = [{"image": "cimg/python:3.8.12"}]
 
 
@@ -126,6 +127,7 @@ class CircleCIJob:
             },
         ]
         steps.extend([{"run": l} for l in self.install_steps])
+        steps.extend([{"run": "pip install pytest-subtests"}])
         steps.append(
             {
                 "save_cache": {
@@ -150,10 +152,13 @@ class CircleCIJob:
         pytest_flags.append(
             f"--make-reports={self.name}" if "examples" in self.name else f"--make-reports=tests_{self.name}"
         )
+
+        steps.append({"run": {"name": "Create `test-results` directory", "command": "mkdir test-results"}})
+
         test_command = ""
         if self.command_timeout:
             test_command = f"timeout {self.command_timeout} "
-        test_command += f"python -m pytest -n {self.pytest_num_workers} " + " ".join(pytest_flags)
+        test_command += f"python -m pytest --junitxml=test-results/junit.xml -n {self.pytest_num_workers} " + " ".join(pytest_flags)
 
         if self.parallelism == 1:
             if self.tests_to_run is None:
@@ -222,18 +227,40 @@ class CircleCIJob:
             # failure.
             test_command = f"({test_command}) || true"
         else:
-            test_command += " | tee tests_output.txt"
+            test_command += " || true"
         steps.append({"run": {"name": "Run tests", "command": test_command}})
 
+        # Deal with errors
+        check_test_command = f'if [ -s reports/{self.job_name}/errors.txt ]; '
+        check_test_command += 'then echo "Some tests errored out!"; echo ""; '
+        check_test_command += f'cat reports/{self.job_name}/errors.txt; '
+        check_test_command += 'echo ""; echo ""; '
+
+        py_command = f'import os; fp = open("reports/{self.job_name}/summary_short.txt"); failed = os.linesep.join([x for x in fp.read().split(os.linesep) if x.startswith("ERROR ")]); fp.close(); fp = open("summary_short.txt", "w"); fp.write(failed); fp.close()'
+        check_test_command += f"$(python3 -c '{py_command}'); "
+        check_test_command += f'cat summary_short.txt; echo ""; exit -1; '
+
+        # Deeal with failed tests
+        check_test_command += f'elif [ -s reports/{self.job_name}/failures_short.txt ]; '
+        check_test_command += 'then echo "Some tests failed!"; echo ""; '
+        check_test_command += f'cat reports/{self.job_name}/failures_short.txt; '
+        check_test_command += 'echo ""; echo ""; '
+
+        py_command = f'import os; fp = open("reports/{self.job_name}/summary_short.txt"); failed = os.linesep.join([x for x in fp.read().split(os.linesep) if x.startswith("FAILED ")]); fp.close(); fp = open("summary_short.txt", "w"); fp.write(failed); fp.close()'
+        check_test_command += f"$(python3 -c '{py_command}'); "
+        check_test_command += f'cat summary_short.txt; echo ""; exit -1; '
+
+        check_test_command += f'elif [ -s reports/{self.job_name}/stats.txt ]; then echo "All tests pass!"; '
+
         # return code `124` means the previous (pytest run) step is timeout
         if self.name == "pr_documentation_tests":
-            checkout_doctest_command = 'if [ -s reports/tests_pr_documentation_tests/failures_short.txt ]; '
-            checkout_doctest_command += 'then echo "some test failed"; '
-            checkout_doctest_command += 'cat reports/tests_pr_documentation_tests/failures_short.txt; '
-            checkout_doctest_command += 'cat reports/tests_pr_documentation_tests/summary_short.txt; exit -1; '
-            checkout_doctest_command += 'elif [ -s reports/tests_pr_documentation_tests/stats.txt ]; then echo "All tests pass!"; '
-            checkout_doctest_command += 'elif [ -f 124.txt ]; then echo "doctest timeout!"; else echo "other fatal error)"; exit -1; fi;'
-            steps.append({"run": {"name": "Check doctest results", "command": checkout_doctest_command}})
+            check_test_command += 'elif [ -f 124.txt ]; then echo "doctest timeout!"; '
+
+        check_test_command += 'else echo "other fatal error"; echo ""; exit -1; fi;'
+
+        steps.append({"run": {"name": "Check test results", "command": check_test_command}})
+
+        steps.append({"store_test_results": {"path": "test-results"}})
 
         steps.append({"store_artifacts": {"path": "~/transformers/tests_output.txt"}})
         steps.append({"store_artifacts": {"path": "~/transformers/reports"}})
@@ -285,7 +312,7 @@ torch_job = CircleCIJob(
         "pip install -U --upgrade-strategy eager git+https://github.com/huggingface/accelerate",
     ],
     parallelism=1,
-    pytest_num_workers=8,
+    pytest_num_workers=6,
 )
 
 
@@ -321,6 +348,7 @@ pipelines_torch_job = CircleCIJob(
         "pip install -U --upgrade-strategy eager .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm,video]",
     ],
     marker="is_pipeline_test",
+    pytest_num_workers=6,
 )
 
 
@@ -440,13 +468,15 @@ exotic_models_job = CircleCIJob(
         "sudo apt install tesseract-ocr",
         "pip install -U --upgrade-strategy eager pytesseract",
         "pip install -U --upgrade-strategy eager natten",
-        # TODO (ydshieh): Remove this line once `https://github.com/facebookresearch/detectron2/issues/5010` is resolved
-        'pip install -U --upgrade-strategy eager "Pillow<10.0.0"',
+        "pip install -U --upgrade-strategy eager python-Levenshtein",
+        "pip install -U --upgrade-strategy eager opencv-python",
+        "pip install -U --upgrade-strategy eager nltk",
     ],
     tests_to_run=[
         "tests/models/*layoutlmv*",
         "tests/models/*nat",
         "tests/models/deta",
+        "tests/models/nougat",
     ],
     pytest_num_workers=1,
     pytest_options={"durations": 100},
@@ -593,7 +623,7 @@ def create_circleci_config(folder=None):
                 job.tests_to_run = [f"examples/{framework}"]
             else:
                 job.tests_to_run = [f for f in example_tests.split(" ") if f.startswith(f"examples/{framework}")]
-            
+
             if len(job.tests_to_run) > 0:
                 jobs.append(job)
 

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -37,7 +37,7 @@ body:
           - pipelines: @Narsil
           - tensorflow: @gante and @Rocketknight1
           - tokenizers: @ArthurZucker
-          - trainer: @muellerz and @pacman100
+          - trainer: @muellerzr and @pacman100
         
         Integrations:
         

.github/PULL_REQUEST_TEMPLATE.md

@@ -51,7 +51,7 @@ Library:
 - pipelines: @Narsil
 - tensorflow: @gante and @Rocketknight1
 - tokenizers: @ArthurZucker
-- trainer: @muellerz and @pacman100
+- trainer: @muellerzr and @pacman100
 
 Integrations:
 

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

@@ -34,19 +34,19 @@ jobs:
           sudo du -sh /usr/share/
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-all-latest-gpu
           build-args: |
@@ -59,7 +59,7 @@ jobs:
         # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
         # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
         if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-all-latest-gpu
           build-args: |
@@ -83,19 +83,19 @@ jobs:
           sudo du -sh /usr/share/
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-pytorch-deepspeed-latest-gpu
           build-args: |
@@ -120,13 +120,13 @@ jobs:
           sudo du -sh /usr/share/
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -136,7 +136,7 @@ jobs:
         # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
         # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
         if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-pytorch-deepspeed-latest-gpu
           build-args: |
@@ -152,19 +152,19 @@ jobs:
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-doc-builder
           push: true
@@ -188,19 +188,19 @@ jobs:
           sudo du -sh /usr/share/
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-pytorch-gpu
           build-args: |
@@ -208,6 +208,41 @@ jobs:
           push: true
           tags: huggingface/transformers-pytorch-gpu
 
+  latest-pytorch-amd:
+    name: "Latest PyTorch (AMD) [dev]"
+    runs-on: [self-hosted, docker-gpu, amd-gpu, single-gpu, mi210]
+    steps:
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Check out code
+        uses: actions/checkout@v3
+      - name: Login to DockerHub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_PASSWORD }}
+      - name: Build and push
+        uses: docker/build-push-action@v5
+        with:
+          context: ./docker/transformers-pytorch-amd-gpu
+          build-args: |
+            REF=main
+          push: true
+          tags: huggingface/transformers-pytorch-amd-gpu${{ inputs.image_postfix }}
+      # Push CI images still need to be re-built daily
+      -
+        name: Build and push (for Push CI) in a daily basis
+        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
+        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
+        if: inputs.image_postfix != '-push-ci'
+        uses: docker/build-push-action@v5
+        with:
+          context: ./docker/transformers-pytorch-amd-gpu
+          build-args: |
+            REF=main
+          push: true
+          tags: huggingface/transformers-pytorch-amd-gpu-push-ci
+
   latest-tensorflow:
     name: "Latest TensorFlow [dev]"
     # Push CI doesn't need this image
@@ -216,19 +251,19 @@ jobs:
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: ./docker/transformers-tensorflow-gpu
           build-args: |

.github/workflows/build_documentation.yml

@@ -15,7 +15,7 @@ jobs:
       commit_sha: ${{ github.sha }}
       package: transformers
       notebook_folder: transformers_doc
-      languages: de en es fr it ko pt zh
+      languages: de en es fr it ko pt zh ja te
     secrets:
       token: ${{ secrets.HUGGINGFACE_PUSH }}
       hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}

.github/workflows/build_pr_documentation.yml

@@ -14,4 +14,4 @@ jobs:
       commit_sha: ${{ github.event.pull_request.head.sha }}
       pr_number: ${{ github.event.number }}
       package: transformers
-      languages: de en es fr it ko pt zh
+      languages: de en es fr it ko pt zh ja te

.github/workflows/doctests.yml

@@ -20,7 +20,7 @@ env:
 
 jobs:
   run_doctests:
-    runs-on: [self-hosted, doc-tests-gpu]
+    runs-on: [single-gpu, nvidia-gpu, t4, doctest-ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/

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

@@ -39,7 +39,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -54,7 +54,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -94,7 +94,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -155,7 +155,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -215,7 +215,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     needs: setup
     container:
       image: huggingface/transformers-pytorch-deepspeed-nightly-gpu

.github/workflows/self-past.yml

@@ -50,7 +50,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -65,7 +65,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -101,7 +101,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -177,7 +177,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     container:
       image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -253,7 +253,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
     needs: setup
     container:
       image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu

.github/workflows/self-push-amd-mi210-caller.yml

@@ -0,0 +1,25 @@
+name: Self-hosted runner (AMD mi210 CI caller)
+
+on:
+  workflow_run:
+    workflows: ["Self-hosted runner (push-caller)"]
+    branches: ["main"]
+    types: [completed]
+  push:
+    branches:
+      - run_amd_push_ci_caller*
+    paths:
+      - "src/**"
+      - "tests/**"
+      - ".github/**"
+      - "templates/**"
+      - "utils/**"
+
+jobs:
+  run_amd_ci:
+    name: AMD mi210
+    if: (cancelled() != true) && ((github.event_name != 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_amd_push_ci_caller')))
+    uses: ./.github/workflows/self-push-amd.yml
+    with:
+      gpu_flavor: mi210
+    secrets: inherit

.github/workflows/self-push-amd-mi250-caller.yml

@@ -0,0 +1,25 @@
+name: Self-hosted runner (AMD mi250 CI caller)
+
+on:
+  workflow_run:
+    workflows: ["Self-hosted runner (push-caller)"]
+    branches: ["main"]
+    types: [completed]
+  push:
+    branches:
+      - run_amd_push_ci_caller*
+    paths:
+      - "src/**"
+      - "tests/**"
+      - ".github/**"
+      - "templates/**"
+      - "utils/**"
+
+jobs:
+  run_amd_ci:
+    name: AMD mi250
+    if: (cancelled() != true) && ((github.event_name != 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_amd_push_ci_caller')))
+    uses: ./.github/workflows/self-push-amd.yml
+    with:
+      gpu_flavor: mi250
+    secrets: inherit

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

@@ -0,0 +1,324 @@
+name: Self-hosted runner AMD GPU (push)
+
+on:
+  workflow_call:
+    inputs:
+      gpu_flavor:
+        required: true
+        type: string
+
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  OMP_NUM_THREADS: 8
+  MKL_NUM_THREADS: 8
+  PYTEST_TIMEOUT: 60
+  TF_FORCE_GPU_ALLOW_GROWTH: true
+  RUN_PT_TF_CROSS_TESTS: 1
+
+jobs:
+  check_runner_status:
+    name: Check Runner Status
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout transformers
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 2
+
+      - name: Check Runner Status
+        run: python utils/check_self_hosted_runner.py --target_runners amd-mi210-single-gpu-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+
+  check_runners:
+    name: Check Runners
+    needs: check_runner_status
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+    container:
+      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
+      options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: ROCM-SMI
+        run: |
+          rocminfo  | grep "Agent" -A 14
+      - name: Show HIP environment
+        run: |
+          echo "HIP: $HIP_VISIBLE_DEVICES"
+          echo "ROCR: $ROCR_VISIBLE_DEVICES"
+
+  setup_gpu:
+    name: Setup
+    needs: check_runners
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+    container:
+      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
+      options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
+      test_map: ${{ steps.set-matrix.outputs.test_map }}
+    steps:
+      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
+      # We also take into account the `push` event (we might want to test some changes in a branch)
+      - name: Prepare custom environment variables
+        shell: bash
+        # `CI_BRANCH_PUSH`: The branch name from the push event
+        # `CI_BRANCH_WORKFLOW_RUN`: The name of the branch on which this workflow is triggered by `workflow_run` event
+        # `CI_BRANCH`: The non-empty branch name from the above two (one and only one of them is empty)
+        # `CI_SHA_PUSH`: The commit SHA from the push event
+        # `CI_SHA_WORKFLOW_RUN`: The commit SHA that triggers this workflow by `workflow_run` event
+        # `CI_SHA`: The non-empty commit SHA from the above two (one and only one of them is empty)
+        run: |
+          CI_BRANCH_PUSH=${{ github.event.ref }}
+          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
+          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
+          CI_SHA_PUSH=${{ github.event.head_commit.id }}
+          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
+          echo $CI_BRANCH_PUSH
+          echo $CI_BRANCH_WORKFLOW_RUN
+          echo $CI_SHA_PUSH
+          echo $CI_SHA_WORKFLOW_RUN
+          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
+          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
+
+      - name: print environment variables
+        run: |
+          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
+          echo "env.CI_SHA = ${{ env.CI_SHA }}"
+
+      - name: Update clone using environment variables
+        working-directory: /transformers
+        run: |
+          echo "original branch = $(git branch --show-current)"
+          git fetch && git checkout ${{ env.CI_BRANCH }}
+          echo "updated branch = $(git branch --show-current)"
+          git checkout ${{ env.CI_SHA }}
+          echo "log = $(git log -n 1)"
+
+      - name: Cleanup
+        working-directory: /transformers
+        run: |
+          rm -rf tests/__pycache__
+          rm -rf tests/models/__pycache__
+          rm -rf reports
+
+      - name: Show installed libraries and their versions
+        working-directory: /transformers
+        run: pip freeze
+
+      - name: Fetch the tests to run
+        working-directory: /transformers
+        # TODO: add `git-python` in the docker images
+        run: |
+          pip install --upgrade git-python
+          python3 utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
+
+      - name: Report fetched tests
+        uses: actions/upload-artifact@v3
+        with:
+          name: test_fetched
+          path: /transformers/test_preparation.txt
+
+      - id: set-matrix
+        name: Organize tests into models
+        working-directory: /transformers
+        # The `keys` is used as GitHub actions matrix for jobs, i.e. `models/bert`, `tokenization`, `pipeline`, etc.
+        # The `test_map` is used to get the actual identified test files under each key.
+        # If no test to run (so no `test_map.json` file), create a dummy map (empty matrix will fail)
+        run: |
+          if [ -f test_map.json ]; then
+              keys=$(python3 -c 'import json; fp = open("test_map.json"); test_map = json.load(fp); fp.close(); d = list(test_map.keys()); print(d)')
+              test_map=$(python3 -c 'import json; fp = open("test_map.json"); test_map = json.load(fp); fp.close(); print(test_map)')
+          else
+              keys=$(python3 -c 'keys = ["dummy"]; print(keys)')
+              test_map=$(python3 -c 'test_map = {"dummy": []}; print(test_map)')
+          fi
+          echo $keys
+          echo $test_map
+          echo "matrix=$keys" >> $GITHUB_OUTPUT
+          echo "test_map=$test_map" >> $GITHUB_OUTPUT
+
+  run_tests_amdgpu:
+    name: Model tests
+    needs: setup_gpu
+    # `dummy` means there is no test to run
+    if: contains(fromJson(needs.setup_gpu.outputs.matrix), 'dummy') != true
+    strategy:
+      fail-fast: false
+      matrix:
+        folders: ${{ fromJson(needs.setup_gpu.outputs.matrix) }}
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+    container:
+      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
+      options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
+      # We also take into account the `push` event (we might want to test some changes in a branch)
+      - name: Prepare custom environment variables
+        shell: bash
+        # For the meaning of these environment variables, see the job `Setup`
+        run: |
+          CI_BRANCH_PUSH=${{ github.event.ref }}
+          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
+          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
+          CI_SHA_PUSH=${{ github.event.head_commit.id }}
+          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
+          echo $CI_BRANCH_PUSH
+          echo $CI_BRANCH_WORKFLOW_RUN
+          echo $CI_SHA_PUSH
+          echo $CI_SHA_WORKFLOW_RUN
+          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
+          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
+
+      - name: print environment variables
+        run: |
+          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
+          echo "env.CI_SHA = ${{ env.CI_SHA }}"
+
+      - name: Update clone using environment variables
+        working-directory: /transformers
+        run: |
+          echo "original branch = $(git branch --show-current)"
+          git fetch && git checkout ${{ env.CI_BRANCH }}
+          echo "updated branch = $(git branch --show-current)"
+          git checkout ${{ env.CI_SHA }}
+          echo "log = $(git log -n 1)"
+
+      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
+        working-directory: /transformers
+        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
+
+      - name: Echo folder ${{ matrix.folders }}
+        shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
+        run: |
+          echo "${{ matrix.folders }}"
+          echo "${{ fromJson(needs.setup_gpu.outputs.test_map)[matrix.folders] }}"
+          matrix_folders=${{ matrix.folders }}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
+          echo "$matrix_folders"
+          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+
+      - name: ROCM-SMI
+        run: |
+          rocminfo  | grep "Agent" -A 14
+      - name: Show HIP environment
+        run: |
+          echo "HIP: $HIP_VISIBLE_DEVICES"
+          echo "ROCR: $ROCR_VISIBLE_DEVICES"
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Show installed libraries and their versions
+        working-directory: /transformers
+        run: pip freeze
+
+      - name: Run all non-slow selected tests on GPU
+        working-directory: /transformers
+        run: |
+          python3 -m pytest -n 2 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} ${{ fromJson(needs.setup_gpu.outputs.test_map)[matrix.folders] }}
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+
+  send_results:
+    name: Send results to webhook
+    runs-on: ubuntu-latest
+    if: always()
+    needs: [
+        check_runner_status,
+        check_runners,
+        setup_gpu,
+        run_tests_amdgpu,
+#        run_tests_torch_cuda_extensions_single_gpu,
+#        run_tests_torch_cuda_extensions_multi_gpu
+    ]
+    steps:
+      - name: Preliminary job status
+        shell: bash
+        # For the meaning of these environment variables, see the job `Setup`
+        run: |
+          echo "Runner availability: ${{ needs.check_runner_status.result }}"
+          echo "Setup status: ${{ needs.setup_gpu.result }}"
+          echo "Runner status: ${{ needs.check_runners.result }}"
+
+      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
+      # We also take into account the `push` event (we might want to test some changes in a branch)
+      - name: Prepare custom environment variables
+        shell: bash
+        # For the meaning of these environment variables, see the job `Setup`
+        run: |
+          CI_BRANCH_PUSH=${{ github.event.ref }}
+          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
+          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
+          CI_SHA_PUSH=${{ github.event.head_commit.id }}
+          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
+          echo $CI_BRANCH_PUSH
+          echo $CI_BRANCH_WORKFLOW_RUN
+          echo $CI_SHA_PUSH
+          echo $CI_SHA_WORKFLOW_RUN
+          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
+          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
+
+      - name: print environment variables
+        run: |
+          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
+          echo "env.CI_SHA = ${{ env.CI_SHA }}"
+
+      - uses: actions/checkout@v3
+        # To avoid failure when multiple commits are merged into `main` in a short period of time.
+        # Checking out to an old commit beyond the fetch depth will get an error `fatal: reference is not a tree: ...
+        # (Only required for `workflow_run` event, where we get the latest HEAD on `main` instead of the event commit)
+        with:
+          fetch-depth: 20
+
+      - name: Update clone using environment variables
+        run: |
+          echo "original branch = $(git branch --show-current)"
+          git fetch && git checkout ${{ env.CI_BRANCH }}
+          echo "updated branch = $(git branch --show-current)"
+          git checkout ${{ env.CI_SHA }}
+          echo "log = $(git log -n 1)"
+
+      - uses: actions/download-artifact@v3
+      - 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_AMD: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
+          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
+          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
+          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+          CI_EVENT: Push CI (AMD) - ${{ inputs.gpu_flavor }}
+          CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
+          CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
+          CI_SHA: ${{ env.CI_SHA }}
+          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
+          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
+          SETUP_STATUS: ${{ needs.setup_gpu.result }}
+
+        # We pass `needs.setup_gpu.outputs.matrix` as the argument. A processing in `notification_service.py` to change
+        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
+        run: |
+          pip install slack_sdk
+          pip show slack_sdk
+          python utils/notification_service.py "${{ needs.setup_gpu.outputs.matrix }}"

.github/workflows/self-push.yml

@@ -45,7 +45,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
     container:
       image: huggingface/transformers-all-latest-gpu-push-ci
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -60,7 +60,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
     container:
       image: huggingface/transformers-all-latest-gpu-push-ci
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -158,7 +158,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
     container:
       image: huggingface/transformers-all-latest-gpu-push-ci
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -251,7 +251,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
     container:
       image: huggingface/transformers-all-latest-gpu-push-ci
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -344,7 +344,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
     container:
       image: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -434,7 +434,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [multi-gpu]
-    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
     container:
       image: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/

.github/workflows/self-scheduled.yml

@@ -43,7 +43,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -58,7 +58,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -98,7 +98,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -159,7 +159,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -219,7 +219,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -270,7 +270,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-pytorch-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -320,7 +320,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     container:
       image: huggingface/transformers-tensorflow-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -371,7 +371,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
     needs: setup
     container:
       image: huggingface/transformers-pytorch-deepspeed-latest-gpu

.github/workflows/update_metdata.yml

@@ -19,7 +19,7 @@ jobs:
       - name: Setup environment
         run: |
           pip install --upgrade pip
-          pip install datasets pandas
+          pip install datasets pandas==2.0.3
           pip install .[torch,tf,flax]
 
       - name: Update metadata

.gitignore

@@ -166,4 +166,4 @@ tags
 .DS_Store
 
 # ruff
-.ruff_cache
+.ruff_cache

Makefile

@@ -43,6 +43,7 @@ repo-consistency:
 	python utils/check_doctest_list.py
 	python utils/update_metadata.py --check-only
 	python utils/check_task_guides.py
+	python utils/check_docstrings.py
 
 # this target runs checks on all files
 
@@ -82,6 +83,7 @@ fix-copies:
 	python utils/check_dummies.py --fix_and_overwrite
 	python utils/check_doctest_list.py --fix_and_overwrite
 	python utils/check_task_guides.py --fix_and_overwrite
+	python utils/check_docstrings.py --fix_and_overwrite
 
 # Run tests for the library
 

README.md

@@ -51,8 +51,10 @@ limitations under the License.
         <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-    <p>
+        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_pt-br.md">Рortuguês</a> |
+    </p>
 </h4>
 
 <h3 align="center">
@@ -310,6 +312,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
 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.
@@ -360,19 +363,21 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 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. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://openai.com/research/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
 1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
 1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
 1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://openai.com/research/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-Sw3](https://huggingface.co/docs/transformers/model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
 1. **[GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
 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. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh.
@@ -388,7 +393,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
 1. **[LiLT](https://huggingface.co/docs/transformers/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
 1. **[LLaMA](https://huggingface.co/docs/transformers/model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
-1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/XXX) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
+1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
 1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
 1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
 1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
@@ -406,6 +411,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The [Mistral AI](https://mistral.ai) team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
 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. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
@@ -423,19 +429,22 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (from ADEPT) released in a [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
 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. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
 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. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
 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.
@@ -486,9 +495,11 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) rreleased with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
 1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.

README_es.md

@@ -18,7 +18,7 @@ limitations under the License.
     <br>
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
-<p>
+</p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
         <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
@@ -47,7 +47,7 @@ limitations under the License.
         <b>Español</b> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-    <p>
+    </p>
 </h4>
 
 <h3 align="center">
@@ -287,6 +287,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
 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.
@@ -337,6 +338,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 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. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -350,6 +352,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
 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. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh. 
@@ -383,6 +386,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The Mistral AI team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.. 
 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. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
@@ -400,19 +404,22 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (from ADEPT) released with the paper [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
 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. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
 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. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
 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.
@@ -463,9 +470,11 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) released with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
 1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.

README_hd.md

@@ -43,7 +43,7 @@ checkpoint: जाँच बिंदु
     <br>
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
-<p>
+</p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
         <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
@@ -72,7 +72,7 @@ checkpoint: जाँच बिंदु
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <b>हिन्दी</b> |
-    <p>
+    </p>
 </h4>
 
 <h3 align="center">
@@ -85,13 +85,13 @@ checkpoint: जाँच बिंदु
 
 🤗 Transformers 100 से अधिक भाषाओं में पाठ वर्गीकरण, सूचना निष्कर्षण, प्रश्न उत्तर, सारांशीकरण, अनुवाद, पाठ निर्माण का समर्थन करने के लिए हजारों पूर्व-प्रशिक्षित मॉडल प्रदान करता है। इसका उद्देश्य सबसे उन्नत एनएलपी तकनीक को सभी के लिए सुलभ बनाना है।
 
-🤗 Transformers त्वरित डाउनलोड और उपयोग के लिए एक एपीआई प्रदान करता है, जिससे आप किसी दिए गए पाठ पर एक पूर्व-प्रशिक्षित मॉडल ले सकते हैं, इसे अपने डेटासेट पर ठीक कर सकते हैं और इसे [मॉडल हब] (https://huggingface.co/models) के माध्यम से समुदाय के साथ साझा कर सकते हैं। ) . इसी समय, प्रत्येक परिभाषित पायथन मॉड्यूल पूरी तरह से स्वतंत्र है, जो संशोधन और तेजी से अनुसंधान प्रयोगों के लिए सुविधाजनक है।
+🤗 Transformers त्वरित डाउनलोड और उपयोग के लिए एक एपीआई प्रदान करता है, जिससे आप किसी दिए गए पाठ पर एक पूर्व-प्रशिक्षित मॉडल ले सकते हैं, इसे अपने डेटासेट पर ठीक कर सकते हैं और इसे [मॉडल हब](https://huggingface.co/models) के माध्यम से समुदाय के साथ साझा कर सकते हैं। इसी समय, प्रत्येक परिभाषित पायथन मॉड्यूल पूरी तरह से स्वतंत्र है, जो संशोधन और तेजी से अनुसंधान प्रयोगों के लिए सुविधाजनक है।
 
 🤗 Transformers तीन सबसे लोकप्रिय गहन शिक्षण पुस्तकालयों का समर्थन करता है： [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — और इसके साथ निर्बाध रूप से एकीकृत होता है। आप अपने मॉडल को सीधे एक ढांचे के साथ प्रशिक्षित कर सकते हैं और दूसरे के साथ लोड और अनुमान लगा सकते हैं।
 
 ## ऑनलाइन डेमो
 
-आप सबसे सीधे मॉडल पृष्ठ पर परीक्षण कर सकते हैं [model hub](https://huggingface.co/models) मॉडल पर। हम [निजी मॉडल होस्टिंग, मॉडल संस्करण, और अनुमान एपीआई] भी प्रदान करते हैं।(https://huggingface.co/pricing)。
+आप सबसे सीधे मॉडल पृष्ठ पर परीक्षण कर सकते हैं [model hub](https://huggingface.co/models) मॉडल पर। हम [निजी मॉडल होस्टिंग, मॉडल संस्करण, और अनुमान एपीआई](https://huggingface.co/pricing) भी प्रदान करते हैं।。
 
 यहाँ कुछ उदाहरण हैं：
 - [शब्द को भरने के लिए मास्क के रूप में BERT का प्रयोग करें](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
@@ -165,7 +165,7 @@ checkpoint: जाँच बिंदु
 
 टोकननाइज़र सभी पूर्व-प्रशिक्षित मॉडलों के लिए प्रीप्रोसेसिंग प्रदान करता है और इसे सीधे एक स्ट्रिंग (जैसे ऊपर दिए गए उदाहरण) या किसी सूची पर बुलाया जा सकता है। यह एक डिक्शनरी (तानाशाही) को आउटपुट करता है जिसे आप डाउनस्ट्रीम कोड में उपयोग कर सकते हैं या `**` अनपैकिंग एक्सप्रेशन के माध्यम से सीधे मॉडल को पास कर सकते हैं।
 
-मॉडल स्वयं एक नियमित [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) या [TensorFlow `tf.keras.Model`](https ://pytorch.org/docs/stable/nn.html#torch.nn.Module) ://www.tensorflow.org/api_docs/python/tf/keras/Model) (आपके बैकएंड के आधार पर), जो हो सकता है सामान्य तरीके से उपयोग किया जाता है। [यह ट्यूटोरियल](https://huggingface.co/transformers/training.html) बताता है कि इस तरह के मॉडल को क्लासिक PyTorch या TensorFlow प्रशिक्षण लूप में कैसे एकीकृत किया जाए, या हमारे `ट्रेनर` एपीआई का उपयोग कैसे करें ताकि इसे जल्दी से फ़ाइन ट्यून किया जा सके।एक नया डेटासेट पे।
+मॉडल स्वयं एक नियमित [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) या [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (आपके बैकएंड के आधार पर), जो हो सकता है सामान्य तरीके से उपयोग किया जाता है। [यह ट्यूटोरियल](https://huggingface.co/transformers/training.html) बताता है कि इस तरह के मॉडल को क्लासिक PyTorch या TensorFlow प्रशिक्षण लूप में कैसे एकीकृत किया जाए, या हमारे `ट्रेनर` एपीआई का उपयोग कैसे करें ताकि इसे जल्दी से फ़ाइन ट्यून किया जा सके।एक नया डेटासेट पे।
 
 ## ट्रांसफार्मर का उपयोग क्यों करें?
 
@@ -194,7 +194,7 @@ checkpoint: जाँच बिंदु
 
 - यह लाइब्रेरी मॉड्यूलर न्यूरल नेटवर्क टूलबॉक्स नहीं है। मॉडल फ़ाइल में कोड जानबूझकर अल्पविकसित है, बिना अतिरिक्त सार इनकैप्सुलेशन के, ताकि शोधकर्ता अमूर्तता और फ़ाइल जंपिंग में शामिल हुए जल्दी से पुनरावृति कर सकें।
 - `ट्रेनर` एपीआई किसी भी मॉडल के साथ संगत नहीं है, यह केवल इस पुस्तकालय के मॉडल के लिए अनुकूलित है। यदि आप सामान्य मशीन लर्निंग के लिए उपयुक्त प्रशिक्षण लूप कार्यान्वयन की तलाश में हैं, तो कहीं और देखें।
-- हमारे सर्वोत्तम प्रयासों के बावजूद, [उदाहरण निर्देशिका] (https://github.com/huggingface/transformers/tree/main/examples) में स्क्रिप्ट केवल उपयोग के मामले हैं। आपकी विशिष्ट समस्या के लिए, वे जरूरी नहीं कि बॉक्स से बाहर काम करें, और आपको कोड की कुछ पंक्तियों को सूट करने की आवश्यकता हो सकती है।
+- हमारे सर्वोत्तम प्रयासों के बावजूद, [उदाहरण निर्देशिका](https://github.com/huggingface/transformers/tree/main/examples) में स्क्रिप्ट केवल उपयोग के मामले हैं। आपकी विशिष्ट समस्या के लिए, वे जरूरी नहीं कि बॉक्स से बाहर काम करें, और आपको कोड की कुछ पंक्तियों को सूट करने की आवश्यकता हो सकती है।
 
 ## स्थापित करना
 
@@ -202,11 +202,13 @@ checkpoint: जाँच बिंदु
 
 इस रिपॉजिटरी का परीक्षण Python 3.8+, Flax 0.4.1+, PyTorch 1.10+ और TensorFlow 2.6+ के तहत किया गया है।
 
-आप [वर्चुअल एनवायरनमेंट] (https://docs.python.org/3/library/venv.html) में 🤗 ट्रांसफॉर्मर इंस्टॉल कर सकते हैं। यदि आप अभी तक पायथन के वर्चुअल एनवायरनमेंट से परिचित नहीं हैं, तो कृपया इसे [उपयोगकर्ता निर्देश] (https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) पढ़ें।
+आप [वर्चुअल एनवायरनमेंट](https://docs.python.org/3/library/venv.html) में 🤗 ट्रांसफॉर्मर इंस्टॉल कर सकते हैं। यदि आप अभी तक पायथन के वर्चुअल एनवायरनमेंट से परिचित नहीं हैं, तो कृपया इसे [उपयोगकर्ता निर्देश](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) पढ़ें।
 
 सबसे पहले, पायथन के उस संस्करण के साथ एक आभासी वातावरण बनाएं जिसका आप उपयोग करने और उसे सक्रिय करने की योजना बना रहे हैं।
 
-फिर, आपको Flax, PyTorch या TensorFlow में से किसी एक को स्थापित करने की आवश्यकता है। अपने प्लेटफ़ॉर्म पर इन फ़्रेमवर्क को स्थापित करने के लिए, [TensorFlow स्थापना पृष्ठ](https://www.tensorflow.org/install/), [PyTorch स्थापना पृष्ठ](https://pytorch.org/get-started /locally/# देखें) start-locally) या [Flax स्थापना पृष्ठ](https://github.com/google/flax#quick-install).
+फिर, आपको Flax, PyTorch या TensorFlow में से किसी एक को स्थापित करने की आवश्यकता है। अपने प्लेटफ़ॉर्म पर इन फ़्रेमवर्क को स्थापित करने के लिए, [TensorFlow स्थापना पृष्ठ](https://www.tensorflow.org/install/), [PyTorch स्थापना पृष्ठ](https://pytorch.org/get-started/locally) 
+
+देखें start-locally या [Flax स्थापना पृष्ठ](https://github.com/google/flax#quick-install).
 
 जब इनमें से कोई एक बैकएंड सफलतापूर्वक स्थापित हो जाता है, तो ट्रांसफॉर्मर निम्नानुसार स्थापित किए जा सकते हैं:
 
@@ -214,7 +216,7 @@ checkpoint: जाँच बिंदु
 pip install transformers
 ```
 
-यदि आप उपयोग के मामलों को आज़माना चाहते हैं या आधिकारिक रिलीज़ से पहले नवीनतम इन-डेवलपमेंट कोड का उपयोग करना चाहते हैं, तो आपको [सोर्स से इंस्टॉल करना होगा](https://huggingface.co/docs/transformers/installation#installing-from- स्रोत)।
+यदि आप उपयोग के मामलों को आज़माना चाहते हैं या आधिकारिक रिलीज़ से पहले नवीनतम इन-डेवलपमेंट कोड का उपयोग करना चाहते हैं, तो आपको [सोर्स से इंस्टॉल करना होगा](https://huggingface.co/docs/transformers/installation#installing-from-) स्रोत।
 
 ### कोंडा का उपयोग करना
 
@@ -229,7 +231,7 @@ conda install -c huggingface transformers
 कोंडा के माध्यम से Flax, PyTorch, या TensorFlow में से किसी एक को स्थापित करने के लिए, निर्देशों के लिए उनके संबंधित स्थापना पृष्ठ देखें।
 
 ## मॉडल आर्किटेक्चर
-[उपयोगकर्ता](https://huggingface.co/users) और [organization](https://huggingface.co) द्वारा ट्रांसफॉर्मर समर्थित [**सभी मॉडल चौकियों**](https://huggingface.co/models) /users) हगिंगफेस.को/ऑर्गनाइजेशन), सभी को बिना किसी बाधा के हगिंगफेस.को [मॉडल हब](https://huggingface.co) के साथ एकीकृत किया गया है।
+[उपयोगकर्ता](https://huggingface.co/users) और [organization](https://huggingface.co) द्वारा ट्रांसफॉर्मर समर्थित [**सभी मॉडल चौकियों**](https://huggingface.co/models/users) हगिंगफेस.को/ऑर्गनाइजेशन), सभी को बिना किसी बाधा के हगिंगफेस.को [मॉडल हब](https://huggingface.co) के साथ एकीकृत किया गया है।
 
 चौकियों की वर्तमान संख्या: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
 
@@ -241,13 +243,13 @@ conda install -c huggingface transformers
 1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
 1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
 1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (फेसबुक) साथ थीसिस [बार्ट: प्राकृतिक भाषा निर्माण, अनुवाद के लिए अनुक्रम-से-अनुक्रम पूर्व प्रशिक्षण , और समझ] (https://arxiv.org/pdf/1910.13461.pdf) पर निर्भर माइक लुईस, यिनहान लियू, नमन गोयल, मार्जन ग़ज़विनिनेजाद, अब्देलरहमान मोहम्मद, ओमर लेवी, वेस स्टोयानोव और ल्यूक ज़ेटलमॉयर
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (फेसबुक) साथ थीसिस [बार्ट: प्राकृतिक भाषा निर्माण, अनुवाद के लिए अनुक्रम-से-अनुक्रम पूर्व प्रशिक्षण , और समझ](https://arxiv.org/pdf/1910.13461.pdf) पर निर्भर माइक लुईस, यिनहान लियू, नमन गोयल, मार्जन ग़ज़विनिनेजाद, अब्देलरहमान मोहम्मद, ओमर लेवी, वेस स्टोयानोव और ल्यूक ज़ेटलमॉयर
 1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (से École polytechnique) साथ थीसिस [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) पर निर्भर Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis रिहाई।
 1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (VinAI Research से) साथ में पेपर [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701)गुयेन लुओंग ट्रान, डुओंग मिन्ह ले और डाट क्वोक गुयेन द्वारा पोस्ट किया गया।
 1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (Microsoft से) साथ में कागज [BEiT: BERT इमेज ट्रांसफॉर्मर्स का प्री-ट्रेनिंग](https://arxiv.org/abs/2106.08254) Hangbo Bao, Li Dong, Furu Wei द्वारा।
 1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (गूगल से) साथ वाला पेपर [बीईआरटी: प्री-ट्रेनिंग ऑफ डीप बिडायरेक्शनल ट्रांसफॉर्मर्स फॉर लैंग्वेज अंडरस्टैंडिंग](https://arxiv.org/abs/1810.04805) जैकब डेवलिन, मिंग-वेई चांग, ​​केंटन ली और क्रिस्टीना टौटानोवा द्वारा प्रकाशित किया गया था। .
 1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (गूगल से) साथ देने वाला पेपर [सीक्वेंस जेनरेशन टास्क के लिए प्री-ट्रेंड चेकपॉइंट का इस्तेमाल करना](https ://arxiv.org/abs/1907.12461) साशा रोठे, शशि नारायण, अलियाक्सि सेवेरिन द्वारा।
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (VinAI Research से) साथ में पेपर [BERTweet: अंग्रेजी ट्वीट्स के लिए एक पूर्व-प्रशिक्षित भाषा मॉडल] (https://aclanthology.org/2020.emnlp-demos.2/) डाट क्वोक गुयेन, थान वु और अन्ह तुआन गुयेन द्वारा प्रकाशित।
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (VinAI Research से) साथ में पेपर [BERTweet: अंग्रेजी ट्वीट्स के लिए एक पूर्व-प्रशिक्षित भाषा मॉडल](https://aclanthology.org/2020.emnlp-demos.2/) डाट क्वोक गुयेन, थान वु और अन्ह तुआन गुयेन द्वारा प्रकाशित।
 1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (गूगल रिसर्च से) साथ वाला पेपर [बिग बर्ड: ट्रांसफॉर्मर्स फॉर लॉन्गर सीक्वेंस](https://arxiv .org/abs/2007.14062) मंज़िल ज़हीर, गुरु गुरुगणेश, अविनावा दुबे, जोशुआ आइंस्ली, क्रिस अल्बर्टी, सैंटियागो ओंटानोन, फिलिप फाम, अनिरुद्ध रावुला, किफ़ान वांग, ली यांग, अमर अहमद द्वारा।
 1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (गूगल रिसर्च से) साथ में पेपर [बिग बर्ड: ट्रांसफॉर्मर्स फॉर लॉन्गर सीक्वेंस](https://arxiv.org/abs/2007.14062) मंज़िल ज़हीर, गुरु गुरुगणेश, अविनावा दुबे, जोशुआ आइंस्ली, क्रिस अल्बर्टी, सैंटियागो ओंटानन, फिलिप फाम द्वारा , अनिरुद्ध रावुला, किफ़ान वांग, ली यांग, अमर अहमद द्वारा पोस्ट किया गया।
 1. **[BioGpt](https://huggingface.co/docs/transformers/model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
@@ -259,6 +261,7 @@ conda install -c huggingface transformers
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (एलेक्सा से) कागज के साथ [बीईआरटी के लिए ऑप्टिमल सबआर्किटेक्चर एक्सट्रैक्शन](https://arxiv.org/abs/ 2010.10499) एड्रियन डी विंटर और डैनियल जे पेरी द्वारा।
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (हरबिन इंस्टिट्यूट ऑफ़ टेक्नोलॉजी/माइक्रोसॉफ्ट रिसर्च एशिया/इंटेल लैब्स से) कागज के साथ [ब्रिजटॉवर: विजन-लैंग्वेज रिप्रेजेंटेशन लर्निंग में एनकोडर्स के बीच ब्रिज बनाना](<https://arxiv.org/abs/2206.08657>) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (NAVER CLOVA से) Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park. द्वाराअनुसंधान पत्र [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) के साथ जारी किया गया
 1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (Google अनुसंधान से) साथ में कागज [ByT5: पूर्व-प्रशिक्षित बाइट-टू-बाइट मॉडल के साथ एक टोकन-मुक्त भविष्य की ओर] (https://arxiv.org/abs/2105.13626) Linting Xue, Aditya Barua, Noah Constant, रामी अल-रफू, शरण नारंग, मिहिर काले, एडम रॉबर्ट्स, कॉलिन रैफेल द्वारा पोस्ट किया गया।
 1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (इनरिया/फेसबुक/सोरबोन से) साथ में कागज [CamemBERT: एक टेस्टी फ्रेंच लैंग्वेज मॉडल](https:// arxiv.org/abs/1911.03894) लुई मार्टिन*, बेंजामिन मुलर*, पेड्रो जेवियर ऑर्टिज़ सुआरेज़*, योआन ड्यूपॉन्ट, लॉरेंट रोमरी, एरिक विलेमोन्टे डे ला क्लर्जरी, जैमे सेडाह और बेनोइट सगोट द्वारा।
 1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (Google रिसर्च से) साथ में दिया गया पेपर [कैनाइन: प्री-ट्रेनिंग ए एफिशिएंट टोकनाइजेशन-फ्री एनकोडर फॉर लैंग्वेज रिप्रेजेंटेशन]( https://arxiv.org/abs/2103.06874) जोनाथन एच क्लार्क, डैन गैरेट, यूलिया टर्क, जॉन विएटिंग द्वारा।
@@ -309,6 +312,7 @@ conda install -c huggingface transformers
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (गूगल रिसर्च से) साथ वाला पेपर [FNet: मिक्सिंग टोकन विद फूरियर ट्रांसफॉर्म्स](https://arxiv.org /abs/2105.03824) जेम्स ली-थॉर्प, जोशुआ आइंस्ली, इल्या एकस्टीन, सैंटियागो ओंटानन द्वारा।
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (Microsoft Research से) Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao. द्वाराअनुसंधान पत्र [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) के साथ जारी किया गया
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (सीएमयू/गूगल ब्रेन से) साथ में कागज [फ़नल-ट्रांसफॉर्मर: कुशल भाषा प्रसंस्करण के लिए अनुक्रमिक अतिरेक को छानना](https://arxiv.org/abs/2006.03236) जिहांग दाई, गुओकुन लाई, यिमिंग यांग, क्वोक वी. ले ​​द्वारा रिहाई।
+1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (ADEPT से) रोहन बाविशी, एरिच एलसेन, कर्टिस हॉथोर्न, मैक्सवेल नी, ऑगस्टस ओडेना, अरुशी सोमानी, सागनाक तासिरलार  [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (KAIST से) साथ वाला पेपर [वर्टिकल कटडेप्थ के साथ मोनोकुलर डेप्थ एस्टीमेशन के लिए ग्लोबल-लोकल पाथ नेटवर्क्स](https:/ /arxiv.org/abs/2201.07436) डोयोन किम, वूंगह्युन गा, प्युंगवान आह, डोंगग्यू जू, सेहवान चुन, जुनमो किम द्वारा।
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (OpenAI से) साथ में दिया गया पेपर [जेनरेटिव प्री-ट्रेनिंग द्वारा भाषा की समझ में सुधार](https://blog .openai.com/language-unsupervised/) एलेक रैडफोर्ड, कार्तिक नरसिम्हन, टिम सालिमन्स और इल्या सुत्स्केवर द्वारा।
@@ -322,6 +326,7 @@ conda install -c huggingface transformers
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (UCSD, NVIDIA से) साथ में कागज [GroupViT: टेक्स्ट सुपरविजन से सिमेंटिक सेगमेंटेशन इमर्जेस](https://arxiv .org/abs/2202.11094) जियारुई जू, शालिनी डी मेलो, सिफ़ी लियू, वोनमिन बायन, थॉमस ब्रेउएल, जान कौट्ज़, ज़ियाओलोंग वांग द्वारा।
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (Allegro.pl, AGH University of Science and Technology से) Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik. द्वाराअनुसंधान पत्र [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) के साथ जारी किया गया
 1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (फेसबुक से) साथ में पेपर [ह्यूबर्ट: सेल्फ सुपरवाइज्ड स्पीच रिप्रेजेंटेशन लर्निंग बाय मास्क्ड प्रेडिक्शन ऑफ हिडन यूनिट्स](https ://arxiv.org/abs/2106.07447) वेई-निंग सू, बेंजामिन बोल्टे, याओ-हंग ह्यूबर्ट त्साई, कुशाल लखोटिया, रुस्लान सालाखुतदीनोव, अब्देलरहमान मोहम्मद द्वारा।
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (बर्कले से) साथ में कागज [I-BERT: Integer-only BERT Quantization](https:// arxiv.org/abs/2101.01321) सेहून किम, अमीर घोलमी, ज़ेवेई याओ, माइकल डब्ल्यू महोनी, कर्ट केटज़र द्वारा।
 1. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh. 
@@ -355,6 +360,7 @@ conda install -c huggingface transformers
 1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (NVIDIA से) कागज के साथ [Megatron-LM: मॉडल का उपयोग करके बहु-अरब पैरामीटर भाषा मॉडल का प्रशिक्षण Parallelism](https://arxiv.org/abs/1909.08053) मोहम्मद शोएबी, मोस्टोफा पटवारी, राउल पुरी, पैट्रिक लेग्रेस्ले, जेरेड कैस्पर और ब्रायन कैटानज़ारो द्वारा।
 1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (NVIDIA से) साथ वाला पेपर [Megatron-LM: ट्रेनिंग मल्टी-बिलियन पैरामीटर लैंग्वेज मॉडल्स यूजिंग मॉडल पैरेललिज़्म] (https://arxiv.org/abs/1909.08053) मोहम्मद शोएबी, मोस्टोफा पटवारी, राउल पुरी, पैट्रिक लेग्रेस्ले, जेरेड कैस्पर और ब्रायन कैटानज़ारो द्वारा पोस्ट किया गया।
 1. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (Alibaba Research से) Peng Wang, Cheng Da, and Cong Yao. द्वाराअनुसंधान पत्र [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) के साथ जारी किया गया
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The Mistral AI team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.. 
 1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (फ्रॉम Studio Ousia) साथ में पेपर [mLUKE: द पावर ऑफ एंटिटी रिप्रेजेंटेशन इन मल्टीलिंगुअल प्रीट्रेन्ड लैंग्वेज मॉडल्स](https://arxiv.org/abs/2110.08151) रयोकन री, इकुया यामाडा, और योशिमासा त्सुरोका द्वारा।
 1. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (Facebook से) Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli. द्वाराअनुसंधान पत्र [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) के साथ जारी किया गया
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (सीएमयू/गूगल ब्रेन से) साथ में कागज [मोबाइलबर्ट: संसाधन-सीमित उपकरणों के लिए एक कॉम्पैक्ट टास्क-अज्ञेय बीईआरटी] (https://arxiv.org/abs/2004.02984) Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, और Denny Zhou द्वारा पोस्ट किया गया।
@@ -372,19 +378,22 @@ conda install -c huggingface transformers
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (हुआवेई नूह के आर्क लैब से) साथ में कागज़ [NEZHA: चीनी भाषा समझ के लिए तंत्रिका प्रासंगिक प्रतिनिधित्व](https :/ /arxiv.org/abs/1909.00204) जुन्किउ वेई, ज़ियाओज़े रेन, ज़िआओगुआंग ली, वेनयोंग हुआंग, यी लियाओ, याशेंग वांग, जियाशू लिन, शिन जियांग, जिओ चेन और कुन लियू द्वारा।
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (फ्रॉम मेटा) साथ में पेपर [नो लैंग्वेज लेफ्ट बिहाइंड: स्केलिंग ह्यूमन-सेंटेड मशीन ट्रांसलेशन] (https://arxiv.org/abs/2207.04672) एनएलएलबी टीम द्वारा प्रकाशित।
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (Meta से) the NLLB team. द्वाराअनुसंधान पत्र [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) के साथ जारी किया गया
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (Meta AI से) Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. द्वाराअनुसंधान पत्र [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) के साथ जारी किया गया
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (विस्कॉन्सिन विश्वविद्यालय - मैडिसन से) साथ में कागज [Nyströmformer: A Nyström- आधारित एल्गोरिथम आत्म-ध्यान का अनुमान लगाने के लिए ](https://arxiv.org/abs/2102.03902) युनयांग ज़िओंग, झानपेंग ज़ेंग, रुद्रसिस चक्रवर्ती, मिंगक्सिंग टैन, ग्लेन फंग, यिन ली, विकास सिंह द्वारा पोस्ट किया गया।
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs से) पेपर [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) जितेश जैन, जिआचेन ली, मांगटिक चिउ, अली हसनी, निकिता ओरलोव, हम्फ्री शि के द्वारा जारी किया गया है।
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (Google AI से) साथ में कागज [विज़न ट्रांसफॉर्मर्स के साथ सिंपल ओपन-वोकैबुलरी ऑब्जेक्ट डिटेक्शन](https:/ /arxiv.org/abs/2205.06230) मैथियास मिंडरर, एलेक्सी ग्रिट्सेंको, ऑस्टिन स्टोन, मैक्सिम न्यूमैन, डिर्क वीसेनबोर्न, एलेक्सी डोसोवित्स्की, अरविंद महेंद्रन, अनुराग अर्नब, मुस्तफा देहघानी, ज़ुओरन शेन, जिओ वांग, ज़ियाओहुआ झाई, थॉमस किफ़, और नील हॉल्सबी द्वारा पोस्ट किया गया।
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (Google AI से) Matthias Minderer, Alexey Gritsenko, Neil Houlsby. द्वाराअनुसंधान पत्र [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) के साथ जारी किया गया
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (Google की ओर से) साथ में दिया गया पेपर [लंबे इनपुट सारांश के लिए ट्रांसफ़ॉर्मरों को बेहतर तरीके से एक्सटेंड करना](https://arxiv .org/abs/2208.04347) जेसन फांग, याओ झाओ, पीटर जे लियू द्वारा।
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (दीपमाइंड से) साथ में पेपर [पर्सीवर आईओ: संरचित इनपुट और आउटपुट के लिए एक सामान्य वास्तुकला] (https://arxiv.org/abs/2107.14795) एंड्रयू जेगल, सेबेस्टियन बोरग्यूड, जीन-बैप्टिस्ट अलायराक, कार्ल डोर्श, कैटलिन इओनेस्कु, डेविड द्वारा डिंग, स्कंद कोप्पुला, डैनियल ज़ोरान, एंड्रयू ब्रॉक, इवान शेलहैमर, ओलिवियर हेनाफ, मैथ्यू एम। बोट्विनिक, एंड्रयू ज़िसरमैन, ओरिओल विनियल्स, जोआओ कैरेरा द्वारा पोस्ट किया गया।
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (ADEPT से) Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani. द्वाराअनुसंधान पत्र [blog post](https://www.adept.ai/blog/persimmon-8b) के साथ जारी किया गया
 1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (VinAI Research से) कागज के साथ [PhoBERT: वियतनामी के लिए पूर्व-प्रशिक्षित भाषा मॉडल](https://www .aclweb.org/anthology/2020.findings-emnlp.92/) डैट क्वोक गुयेन और अन्ह तुआन गुयेन द्वारा पोस्ट किया गया।
 1. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (Google से) Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova. द्वाराअनुसंधान पत्र [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) के साथ जारी किया गया
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (UCLA NLP से) साथ वाला पेपर [प्रोग्राम अंडरस्टैंडिंग एंड जेनरेशन के लिए यूनिफाइड प्री-ट्रेनिंग](https://arxiv .org/abs/2103.06333) वसी उद्दीन अहमद, सैकत चक्रवर्ती, बैशाखी रे, काई-वेई चांग द्वारा।
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
 1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (माइक्रोसॉफ्ट रिसर्च से) साथ में पेपर [ProphetNet: प्रेडिक्टिंग फ्यूचर एन-ग्राम फॉर सीक्वेंस-टू-सीक्वेंस प्री-ट्रेनिंग ](https://arxiv.org/abs/2001.04063) यू यान, वीज़ेन क्यूई, येयुन गोंग, दयाहेंग लियू, नान डुआन, जिउशेंग चेन, रुओफ़ेई झांग और मिंग झोउ द्वारा पोस्ट किया गया।
 1. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (Nanjing University, The University of Hong Kong etc. से) Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao. द्वाराअनुसंधान पत्र [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) के साथ जारी किया गया
 1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (NVIDIA से) साथ वाला पेपर [डीप लर्निंग इंफ़ेक्शन के लिए इंटीजर क्वांटिज़ेशन: प्रिंसिपल्स एंड एम्पिरिकल इवैल्यूएशन](https:// arxiv.org/abs/2004.09602) हाओ वू, पैट्रिक जुड, जिआओजी झांग, मिखाइल इसेव और पॉलियस माइकेविसियस द्वारा।
@@ -435,16 +444,18 @@ conda install -c huggingface transformers
 1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (गूगल एआई से) कागज के साथ [एक इमेज इज़ वर्थ 16x16 वर्ड्स: ट्रांसफॉर्मर्स फॉर इमेज रिकॉग्निशन एट स्केल](https://arxiv.org/abs/2010.11929) एलेक्सी डोसोवित्स्की, लुकास बेयर, अलेक्जेंडर कोलेसनिकोव, डिर्क वीसेनबोर्न, शियाओहुआ झाई, थॉमस अनटरथिनर, मुस्तफा देहघानी, मैथियास मिंडरर, जॉर्ज हेगोल्ड, सिल्वेन गेली, जैकब उस्ज़कोरेइट द्वारा हॉल्सबी द्वारा पोस्ट किया गया।
 1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (UCLA NLP से) साथ वाला पेपर [VisualBERT: A Simple and Performant Baseline for Vision and Language](https:/ /arxiv.org/pdf/1908.03557) लियुनियन हेरोल्ड ली, मार्क यात्स्कर, दा यिन, चो-जुई हसीह, काई-वेई चांग द्वारा।
 1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (Meta AI से) Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He. द्वाराअनुसंधान पत्र [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) के साथ जारी किया गया
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (Meta AI से) Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He. द्वाराअनुसंधान पत्र [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) के साथ जारी किया गया
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (मेटा एआई से) साथ में कागज [मास्कड ऑटोएन्कोडर स्केलेबल विजन लर्नर्स हैं](https://arxiv.org/ एब्स/2111.06377) कैमिंग हे, ज़िनेली चेन, सेनिंग ज़ी, यांगहो ली, पिओट्र डॉलर, रॉस गिर्शिक द्वारा।
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (HUST-VL से) Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang. द्वाराअनुसंधान पत्र [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) के साथ जारी किया गया
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (मेटा एआई से) साथ में कागज [लेबल-कुशल सीखने के लिए मास्क्ड स्याम देश के नेटवर्क](https://arxiv. org/abs/2204.07141) महमूद असरान, मथिल्डे कैरन, ईशान मिश्रा, पियोट्र बोजानोवस्की, फ्लोरियन बोर्डेस, पास्कल विंसेंट, आर्मंड जौलिन, माइकल रब्बत, निकोलस बल्लास द्वारा।
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (Kakao Enterprise से) Jaehyeon Kim, Jungil Kong, Juhee Son. द्वाराअनुसंधान पत्र [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) के साथ जारी किया गया
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (फेसबुक एआई से) साथ में पेपर [wav2vec 2.0: ए फ्रेमवर्क फॉर सेल्फ-सुपरवाइज्ड लर्निंग ऑफ स्पीच रिप्रेजेंटेशन] (https://arxiv.org/abs/2006.11477) एलेक्सी बेवस्की, हेनरी झोउ, अब्देलरहमान मोहम्मद, माइकल औली द्वारा।
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (फेसबुक एआई से) साथ में पेपर [wav2vec 2.0: ए फ्रेमवर्क फॉर सेल्फ-सुपरवाइज्ड लर्निंग ऑफ स्पीच रिप्रेजेंटेशन](https://arxiv.org/abs/2006.11477) एलेक्सी बेवस्की, हेनरी झोउ, अब्देलरहमान मोहम्मद, माइकल औली द्वारा।
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (Facebook AI से) साथ वाला पेपर [FAIRSEQ S2T: FAIRSEQ के साथ फास्ट स्पीच-टू-टेक्स्ट मॉडलिंग ](https://arxiv.org/abs/2010.05171) चांगहान वांग, यूं तांग, जुताई मा, ऐनी वू, सरव्या पोपुरी, दिमित्रो ओखोनको, जुआन पिनो द्वारा पोस्ट किया गया।
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (Facebook AI से) साथ वाला पेपर [सरल और प्रभावी जीरो-शॉट क्रॉस-लिंगुअल फोनेम रिकॉग्निशन](https:/ /arxiv.org/abs/2109.11680) कियानटोंग जू, एलेक्सी बाएव्स्की, माइकल औली द्वारा।
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (माइक्रोसॉफ्ट रिसर्च से) पेपर के साथ जारी किया गया [WavLM: फुल स्टैक के लिए बड़े पैमाने पर स्व-पर्यवेक्षित पूर्व-प्रशिक्षण स्पीच प्रोसेसिंग] (https://arxiv.org/abs/2110.13900) सानयुआन चेन, चेंगयी वांग, झेंगयांग चेन, यू वू, शुजी लियू, ज़ुओ चेन, जिन्यु ली, नाओयुकी कांडा, ताकुया योशियोका, ज़िओंग जिओ, जियान वू, लॉन्ग झोउ, शुओ रेन, यानमिन कियान, याओ कियान, जियान वू, माइकल ज़ेंग, फुरु वेई।
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (Facebook AI से) साथ वाला पेपर [सरल और प्रभावी जीरो-शॉट क्रॉस-लिंगुअल फोनेम रिकॉग्निशन](https://arxiv.org/abs/2109.11680) कियानटोंग जू, एलेक्सी बाएव्स्की, माइकल औली द्वारा।
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (माइक्रोसॉफ्ट रिसर्च से) पेपर के साथ जारी किया गया [WavLM: फुल स्टैक के लिए बड़े पैमाने पर स्व-पर्यवेक्षित पूर्व-प्रशिक्षण स्पीच प्रोसेसिंग](https://arxiv.org/abs/2110.13900) सानयुआन चेन, चेंगयी वांग, झेंगयांग चेन, यू वू, शुजी लियू, ज़ुओ चेन, जिन्यु ली, नाओयुकी कांडा, ताकुया योशियोका, ज़िओंग जिओ, जियान वू, लॉन्ग झोउ, शुओ रेन, यानमिन कियान, याओ कियान, जियान वू, माइकल ज़ेंग, फुरु वेई।
 1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (OpenAI से) साथ में कागज [बड़े पैमाने पर कमजोर पर्यवेक्षण के माध्यम से मजबूत भाषण पहचान](https://cdn. openai.com/papers/whisper.pdf) एलेक रैडफोर्ड, जोंग वूक किम, ताओ जू, ग्रेग ब्रॉकमैन, क्रिस्टीन मैकलीवे, इल्या सुत्स्केवर द्वारा।
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (माइक्रोसॉफ्ट रिसर्च से) कागज के साथ [एक्सपैंडिंग लैंग्वेज-इमेज प्रीट्रेन्ड मॉडल फॉर जनरल वीडियो रिकग्निशन](https: //arxiv.org/abs/2208.02816) बोलिन नी, होउवेन पेंग, मिंगाओ चेन, सोंगयांग झांग, गाओफेंग मेंग, जियानलोंग फू, शिमिंग जियांग, हैबिन लिंग द्वारा।
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (माइक्रोसॉफ्ट रिसर्च से) कागज के साथ [एक्सपैंडिंग लैंग्वेज-इमेज प्रीट्रेन्ड मॉडल फॉर जनरल वीडियो रिकग्निशन](https://arxiv.org/abs/2208.02816) बोलिन नी, होउवेन पेंग, मिंगाओ चेन, सोंगयांग झांग, गाओफेंग मेंग, जियानलोंग फू, शिमिंग जियांग, हैबिन लिंग द्वारा।
 1. **[X-MOD](https://huggingface.co/docs/transformers/model_doc/xmod)** (Meta AI से) Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe. द्वाराअनुसंधान पत्र [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) के साथ जारी किया गया
 1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
 1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (फेसबुक से) साथ में पेपर [क्रॉस-लिंगुअल लैंग्वेज मॉडल प्रीट्रेनिंग] (https://arxiv.org/abs/1901.07291) गिलाउम लैम्पल और एलेक्सिस कोनो द्वारा।
@@ -457,7 +468,7 @@ conda install -c huggingface transformers
 1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (फेसबुक एआई से) साथ में पेपर [अनसुपरवाइज्ड क्रॉस-लिंगुअल रिप्रेजेंटेशन लर्निंग फॉर स्पीच रिकग्निशन] (https://arxiv.org/abs/2006.13979) एलेक्सिस कोन्यू, एलेक्सी बेवस्की, रोनन कोलोबर्ट, अब्देलरहमान मोहम्मद, माइकल औली द्वारा।
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (हुआझोंग यूनिवर्सिटी ऑफ साइंस एंड टेक्नोलॉजी से) साथ में पेपर [यू ओनली लुक एट वन सीक्वेंस: रीथिंकिंग ट्रांसफॉर्मर इन विज़न थ्रू ऑब्जेक्ट डिटेक्शन](https://arxiv.org/abs/2106.00666) युक्सिन फेंग, बेनचेंग लियाओ, जिंगगैंग वांग, जेमिन फेंग, जियांग क्यूई, रुई वू, जियानवेई नीयू, वेन्यू लियू द्वारा पोस्ट किया गया।
 1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (विस्कॉन्सिन विश्वविद्यालय - मैडिसन से) साथ में पेपर [यू ओनली सैंपल (लगभग) ज़ानपेंग ज़ेंग, युनयांग ज़िओंग द्वारा , सत्य एन. रवि, शैलेश आचार्य, ग्लेन फंग, विकास सिंह द्वारा पोस्ट किया गया।
-1. एक नए मॉडल में योगदान देना चाहते हैं? नए मॉडल जोड़ने में आपका मार्गदर्शन करने के लिए हमारे पास एक **विस्तृत मार्गदर्शिका और टेम्प्लेट** है। आप उन्हें [`टेम्पलेट्स`](./templates) निर्देशिका में पा सकते हैं। पीआर शुरू करने से पहले [योगदान दिशानिर्देश] (./CONTRIBUTING.md) देखना और अनुरक्षकों से संपर्क करना या प्रतिक्रिया प्राप्त करने के लिए एक नया मुद्दा खोलना याद रखें।
+1. एक नए मॉडल में योगदान देना चाहते हैं? नए मॉडल जोड़ने में आपका मार्गदर्शन करने के लिए हमारे पास एक **विस्तृत मार्गदर्शिका और टेम्प्लेट** है। आप उन्हें [`टेम्पलेट्स`](./templates) निर्देशिका में पा सकते हैं। पीआर शुरू करने से पहले [योगदान दिशानिर्देश](./CONTRIBUTING.md) देखना और अनुरक्षकों से संपर्क करना या प्रतिक्रिया प्राप्त करने के लिए एक नया मुद्दा खोलना याद रखें।
 
 यह जांचने के लिए कि क्या किसी मॉडल में पहले से ही Flax, PyTorch या TensorFlow का कार्यान्वयन है, या यदि उसके पास Tokenizers लाइब्रेरी में संबंधित टोकन है, तो [यह तालिका](https://huggingface.co/docs/transformers/index#supported) देखें। -फ्रेमवर्क)।
 

README_ja.md

@@ -53,7 +53,7 @@ user: ユーザ
     <br>
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
-<p>
+</p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
         <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
@@ -82,7 +82,7 @@ user: ユーザ
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <b>日本語</b> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-    <p>
+    </p>
 </h4>
 
 <h3 align="center">
@@ -210,7 +210,7 @@ Hugging Faceチームによって作られた **[トランスフォーマーを
 >>> outputs = model(**inputs)
 ```
 
-And here is the equivalent code for TensorFlow:
+そしてこちらはTensorFlowと同等のコードとなります:
 ```python
 >>> from transformers import AutoTokenizer, TFAutoModel
 
@@ -321,6 +321,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (BigScience workshop から) [BigScience Workshop](https://bigscience.huggingface.co/) から公開されました.
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (Alexa から) Adrian de Wynter and Daniel J. Perry から公開された研究論文: [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499)
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (Harbin Institute of Technology/Microsoft Research Asia/Intel Labs から) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (NAVER CLOVA から) Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park. から公開された研究論文 [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539)
 1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (Google Research から) Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel から公開された研究論文: [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626)
 1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (Inria/Facebook/Sorbonne から) Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot から公開された研究論文: [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894)
 1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (Google Research から) Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting から公開された研究論文: [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874)
@@ -371,6 +372,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (Google Research から) James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon から公開された研究論文: [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824)
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (Microsoft Research から) Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao. から公開された研究論文 [Focal Modulation Networks](https://arxiv.org/abs/2203.11926)
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (CMU/Google Brain から) Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le から公開された研究論文: [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236)
+1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (ADEPT から) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. から公開された研究論文 [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (Microsoft Research から) Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang. から公開された研究論文 [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100)
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (KAIST から) Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim から公開された研究論文: [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436)
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (OpenAI から) Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever から公開された研究論文: [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/)
@@ -384,6 +386,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) 坂本俊之(tanreinama)からリリースされました.
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (Microsoft から) Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu から公開された研究論文: [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234).
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (UCSD, NVIDIA から) Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang から公開された研究論文: [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094)
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (Allegro.pl, AGH University of Science and Technology から) Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik. から公開された研究論文 [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf)
 1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (Facebook から) Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed から公開された研究論文: [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447)
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (Berkeley から) Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer から公開された研究論文: [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321)
 1. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh. 
@@ -417,6 +420,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (NVIDIA から) Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro から公開された研究論文: [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053)
 1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (NVIDIA から) Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro から公開された研究論文: [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053)
 1. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (Alibaba Research から) Peng Wang, Cheng Da, and Cong Yao. から公開された研究論文 [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592)
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The Mistral AI team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.. 
 1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (Studio Ousia から) Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka から公開された研究論文: [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151)
 1. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (Facebook から) Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli. から公開された研究論文 [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516)
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (CMU/Google Brain から) Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou から公開された研究論文: [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984)
@@ -434,19 +438,22 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (Huawei Noah’s Ark Lab から) Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu から公開された研究論文: [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204)
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (Meta から) the NLLB team から公開された研究論文: [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672)
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (Meta から) the NLLB team. から公開された研究論文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672)
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (Meta AI から) Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. から公開された研究論文 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418)
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (the University of Wisconsin - Madison から) Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh から公開された研究論文: [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902)
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs から) Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi から公開された研究論文: [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220)
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (Meta AI から) Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al から公開された研究論文: [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068)
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (Google AI から) Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby から公開された研究論文: [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230)
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (Google AI から) Matthias Minderer, Alexey Gritsenko, Neil Houlsby. から公開された研究論文 [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683)
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (Google から) Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu から公開された研究論文: [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777)
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (Google から) Jason Phang, Yao Zhao, and Peter J. Liu から公開された研究論文: [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347)
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (Deepmind から) 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 から公開された研究論文: [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795)
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (ADEPT から) Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani. から公開された研究論文 [blog post](https://www.adept.ai/blog/persimmon-8b)
 1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (VinAI Research から) Dat Quoc Nguyen and Anh Tuan Nguyen から公開された研究論文: [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/)
 1. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (Google から) Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova. から公開された研究論文 [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347)
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (UCLA NLP から) Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang から公開された研究論文: [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333)
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (Sea AI Labs から) Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng から公開された研究論文: [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418)
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
 1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (Microsoft Research から) Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou から公開された研究論文: [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063)
 1. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (Nanjing University, The University of Hong Kong etc. から) Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao. から公開された研究論文 [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf)
 1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (NVIDIA から) Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius から公開された研究論文: [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602)
@@ -497,9 +504,11 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (Google AI から) Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby から公開された研究論文: [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929)
 1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (UCLA NLP から) Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang から公開された研究論文: [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557)
 1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (Google AI から) Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby から公開された研究論文: [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929)
-1. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (Meta AI から) Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He. から公開された研究論文 [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527)
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (Meta AI から) Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He. から公開された研究論文 [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527)
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (Meta AI から) Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick から公開された研究論文: [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377)
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (HUST-VL から) Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang. から公開された研究論文 [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272)
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (Meta AI から) Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas から公開された研究論文: [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141)
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (Kakao Enterprise から) Jaehyeon Kim, Jungil Kong, Juhee Son. から公開された研究論文 [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103)
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
 1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (Facebook AI から) Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli から公開された研究論文: [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477)
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (Facebook AI から) Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino から公開された研究論文: [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171)

README_ko.md

@@ -18,7 +18,7 @@ limitations under the License.
     <br>
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
-<p>
+</p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
         <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
@@ -47,7 +47,7 @@ limitations under the License.
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-    <p>
+    </p>
 </h4>
 
 <h3 align="center">
@@ -236,6 +236,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (Alexa 에서) Adrian de Wynter and Daniel J. Perry 의 [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) 논문과 함께 발표했습니다.
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (NAVER CLOVA 에서 제공)은 Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.의 [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539)논문과 함께 발표했습니다.
 1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (Google Research 에서) Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel 의 [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) 논문과 함께 발표했습니다.
 1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (Inria/Facebook/Sorbonne 에서) Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot 의 [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) 논문과 함께 발표했습니다.
 1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (Google Research 에서) Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting 의 [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) 논문과 함께 발표했습니다.
@@ -286,6 +287,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 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. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 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. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. 논문과 함께 공개 [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -299,6 +301,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu  의 [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234)  논문과 함께 발표했습니다.
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (UCSD, NVIDIA 에서) Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang 의 [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) 논문과 함께 발표했습니다.
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (Allegro.pl, AGH University of Science and Technology 에서 제공)은 Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.의 [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf)논문과 함께 발표했습니다.
 1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (Facebook 에서) Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed 의 [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) 논문과 함께 발표했습니다.
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (Berkeley 에서) Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer 의 [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) 논문과 함께 발표했습니다.
 1. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh. 
@@ -332,6 +335,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (NVIDIA 에서) Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro 의 [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) 논문과 함께 발표했습니다.
 1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (NVIDIA 에서) Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro 의 [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) 논문과 함께 발표했습니다.
 1. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (Alibaba Research 에서 제공)은 Peng Wang, Cheng Da, and Cong Yao.의 [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592)논문과 함께 발표했습니다.
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The Mistral AI team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.. 
 1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (Studio Ousia 에서) Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka 의 [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) 논문과 함께 발표했습니다.
 1. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (Facebook 에서 제공)은 Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.의 [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516)논문과 함께 발표했습니다.
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (CMU/Google Brain 에서) Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou 의 [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) 논문과 함께 발표했습니다.
@@ -349,19 +353,22 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (Huawei Noah’s Ark Lab 에서) Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 의 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 논문과 함께 발표했습니다.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (Meta 에서) the NLLB team 의 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 논문과 함께 발표했습니다.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (Meta 에서 제공)은 the NLLB team.의 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672)논문과 함께 발표했습니다.
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (Meta AI 에서 제공)은 Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.의 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418)논문과 함께 발표했습니다.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (the University of Wisconsin - Madison 에서) Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 의 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 논문과 함께 발표했습니다.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs 에서) Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi 의 [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) 논문과 함께 발표했습니다.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (Meta AI 에서) Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 의 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 논문과 함께 발표했습니다.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (Google AI 에서) Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 의 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 논문과 함께 발표했습니다.
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (Google AI 에서 제공)은 Matthias Minderer, Alexey Gritsenko, Neil Houlsby.의 [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683)논문과 함께 발표했습니다.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (Google 에서) Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 의 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 논문과 함께 발표했습니다.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (Google 에서) Jason Phang, Yao Zhao, Peter J. Liu 의 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 논문과 함께 발표했습니다.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (Deepmind 에서) 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 의 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 논문과 함께 발표했습니다.
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (ADEPT 에서 제공)은 Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.의 [blog post](https://www.adept.ai/blog/persimmon-8b)논문과 함께 발표했습니다.
 1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (VinAI Research 에서) Dat Quoc Nguyen and Anh Tuan Nguyen 의 [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) 논문과 함께 발표했습니다.
 1. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (Google 에서 제공)은 Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.의 [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347)논문과 함께 발표했습니다.
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (UCLA NLP 에서) Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang 의 [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) 논문과 함께 발표했습니다.
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (Sea AI Labs 에서) Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng 의 [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) 논문과 함께 발표했습니다.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
 1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (Microsoft Research 에서) Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou 의 [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) 논문과 함께 발표했습니다.
 1. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (Nanjing University, The University of Hong Kong etc. 에서 제공)은 Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.의 [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf)논문과 함께 발표했습니다.
 1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (NVIDIA 에서) Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius 의 [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) 논문과 함께 발표했습니다.
@@ -412,9 +419,11 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (Google AI 에서) Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby 의 [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) 논문과 함께 발표했습니다.
 1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (UCLA NLP 에서) Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang 의 [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) 논문과 함께 발표했습니다.
 1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (Google AI 에서) Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby 의 [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) 논문과 함께 발표했습니다.
-1. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (Meta AI 에서 제공)은 Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.의 [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527)논문과 함께 발표했습니다.
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (Meta AI 에서 제공)은 Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.의 [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527)논문과 함께 발표했습니다.
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (Meta AI 에서) Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick 의 [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) 논문과 함께 발표했습니다.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (HUST-VL 에서 제공)은 Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.의 [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272)논문과 함께 발표했습니다.
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (Meta AI 에서) Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas 의 [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) 논문과 함께 발표했습니다.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (Kakao Enterprise 에서 제공)은 Jaehyeon Kim, Jungil Kong, Juhee Son.의 [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103)논문과 함께 발표했습니다.
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
 1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (Facebook AI 에서) Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli 의 [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) 논문과 함께 발표했습니다.
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (Facebook AI 에서) Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino 의 [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) 논문과 함께 발표했습니다.
@@ -433,7 +442,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (Facebook AI 에서) 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 의 [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) 논문과 함께 발표했습니다.
 1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (Facebook AI 에서) Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli 의 [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) 논문과 함께 발표했습니다.
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (Huazhong University of Science & Technology 에서) Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu 의 [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) 논문과 함께 발표했습니다.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (the University of Wisconsin - Madison 에서) Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh 의 [You Only Sample (Almost) 논문과 함께 발표했습니다.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (the University of Wisconsin - Madison 에서) Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh 의 [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) 논문과 함께 발표했습니다.
 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.

README_pt-br.md

@@ -0,0 +1,575 @@
+<!---
+Copyright 2023 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">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-dark.svg">
+    <source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg">
+    <img alt="Hugging Face Transformers Library" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg" width="352" height="59" style="max-width: 100%;">
+  </picture>
+  <br/>
+  <br/>
+</p>
+
+<p align="center">
+    <a href="https://circleci.com/gh/huggingface/transformers">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
+    </a>
+    <a href="https://huggingface.co/docs/transformers/index">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    </a>
+    <a href="https://github.com/huggingface/transformers/releases">
+        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
+    </a>
+    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
+</p>
+
+<h4 align="center">
+    <p>
+        <b>English</b> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_pt-br.md">Рortuguês</a> |
+    </p>
+</h4>
+
+<h3 align="center">
+    <p>Aprendizado de máquina de última geração para JAX, PyTorch e TensorFlow</p>
+</h3>
+
+<h3 align="center">
+    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
+</h3>
+
+
+A biblioteca 🤗 Transformers oferece milhares de modelos pré-treinados para executar tarefas em diferentes modalidades, como texto, visão e áudio.
+
+Esses modelos podem ser aplicados a:
+
+* 📝 Texto, para tarefas como classificação de texto, extração de informações, resposta a perguntas, sumarização, tradução, geração de texto, em mais de 100 idiomas.
+* 🖼️ Imagens, para tarefas como classificação de imagens, detecção de objetos e segmentação.
+* 🗣️ Áudio, para tarefas como reconhecimento de fala e classificação de áudio.
+
+Os modelos Transformer também podem executar tarefas em diversas modalidades combinadas, como responder a perguntas em tabelas, reconhecimento óptico de caracteres, extração de informações de documentos digitalizados, classificação de vídeo e resposta a perguntas visuais.
+
+
+A biblioteca 🤗 Transformers oferece APIs para baixar e usar rapidamente esses modelos pré-treinados em um texto específico, ajustá-los em seus próprios conjuntos de dados e, em seguida, compartilhá-los com a comunidade em nosso [model hub](https://huggingface.co/models). Ao mesmo tempo, cada módulo Python que define uma arquitetura é totalmente independente e pode ser modificado para permitir experimentos de pesquisa rápidos.
+
+A biblioteca 🤗 Transformers é respaldada pelas três bibliotecas de aprendizado profundo mais populares — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) e [TensorFlow](https://www.tensorflow.org/) — com uma integração perfeita entre elas. É simples treinar seus modelos com uma delas antes de carregá-los para inferência com a outra
+
+## Demonstração Online
+
+Você pode testar a maioria de nossos modelos diretamente em suas páginas a partir do [model hub](https://huggingface.co/models). Também oferecemos [hospedagem de modelos privados, versionamento e uma API de inferência](https://huggingface.co/pricing)
+para modelos públicos e privados.
+
+Aqui estão alguns exemplos:
+
+Em Processamento de Linguagem Natural:
+
+- [Completar palavra mascarada com BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
+-  [Reconhecimento de Entidades Nomeadas com Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
+- [Geração de texto com GPT-2](https://huggingface.co/gpt2?text=A+long+time+ago%2C)
+- [Inferência de Linguagem Natural com RoBERTa](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
+- [Sumarização com BART](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
+- [Resposta a perguntas com 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)
+- [Tradução com T5](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
+
+
+Em Visão Computacional:
+- [Classificação de Imagens com ViT](https://huggingface.co/google/vit-base-patch16-224)
+- [Detecção de Objetos com DETR](https://huggingface.co/facebook/detr-resnet-50)
+- [Segmentação Semântica com SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
+- [Segmentação Panóptica com MaskFormer](https://huggingface.co/facebook/maskformer-swin-small-coco)
+- [Estimativa de Profundidade com DPT](https://huggingface.co/docs/transformers/model_doc/dpt)
+- [Classificação de Vídeo com VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)
+- [Segmentação Universal com OneFormer](https://huggingface.co/shi-labs/oneformer_ade20k_dinat_large)
+
+
+Em Áudio:
+- [Reconhecimento Automático de Fala com Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base-960h)
+- [Detecção de Palavras-Chave com Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
+- [Classificação de Áudio com Transformer de Espectrograma de Áudio](https://huggingface.co/MIT/ast-finetuned-audioset-10-10-0.4593)
+
+Em Tarefas Multimodais:
+- [Respostas de Perguntas em Tabelas com TAPAS](https://huggingface.co/google/tapas-base-finetuned-wtq)
+- [Respostas de Perguntas Visuais com ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
+- [Classificação de Imagens sem Anotação com CLIP](https://huggingface.co/openai/clip-vit-large-patch14)
+- [Respostas de Perguntas em Documentos com LayoutLM](https://huggingface.co/impira/layoutlm-document-qa)
+- [Classificação de Vídeo sem Anotação com X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)
+
+## 100 Projetos Usando Transformers
+
+Transformers é mais do que um conjunto de ferramentas para usar modelos pré-treinados: é uma comunidade de projetos construídos ao seu redor e o Hugging Face Hub. Queremos que o Transformers permita que desenvolvedores, pesquisadores, estudantes, professores, engenheiros e qualquer outra pessoa construa seus projetos dos sonhos.
+
+Para celebrar as 100.000 estrelas do Transformers, decidimos destacar a comunidade e criamos a página [awesome-transformers](./awesome-transformers.md), que lista 100 projetos incríveis construídos nas proximidades dos Transformers.
+
+Se você possui ou utiliza um projeto que acredita que deveria fazer parte da lista, abra um PR para adicioná-lo!
+
+## Se você está procurando suporte personalizado da equipe Hugging Face
+
+<a target="_blank" href="https://huggingface.co/support">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+</a><br>
+
+
+## Tour Rápido
+
+Para usar imediatamente um modelo em uma entrada específica (texto, imagem, áudio, ...), oferecemos a API `pipeline`. Os pipelines agrupam um modelo pré-treinado com o pré-processamento que foi usado durante o treinamento desse modelo. Aqui está como usar rapidamente um pipeline para classificar textos como positivos ou negativos:
+
+```python
+from transformers import pipeline
+
+# Carregue o pipeline de classificação de texto
+>>> classifier = pipeline("sentiment-analysis")
+
+# Classifique o texto como positivo ou negativo
+>>> classifier("Estamos muito felizes em apresentar o pipeline no repositório dos transformers.")
+[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
+```
+
+A segunda linha de código baixa e armazena em cache o modelo pré-treinado usado pelo pipeline, enquanto a terceira linha o avalia no texto fornecido. Neste exemplo, a resposta é "positiva" com uma confiança de 99,97%.
+
+Muitas tarefas têm um `pipeline` pré-treinado pronto para uso, não apenas em PNL, mas também em visão computacional e processamento de áudio. Por exemplo, podemos facilmente extrair objetos detectados em uma imagem:
+
+``` python
+>>> import requests
+>>> from PIL import Image
+>>> from transformers import pipeline
+
+# Download an image with cute cats
+>>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
+>>> image_data = requests.get(url, stream=True).raw
+>>> image = Image.open(image_data)
+
+# Allocate a pipeline for object detection
+>>> object_detector = pipeline('object-detection')
+>>> object_detector(image)
+[{'score': 0.9982201457023621,
+  'label': 'remote',
+  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
+ {'score': 0.9960021376609802,
+  'label': 'remote',
+  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
+ {'score': 0.9954745173454285,
+  'label': 'couch',
+  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
+ {'score': 0.9988006353378296,
+  'label': 'cat',
+  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
+ {'score': 0.9986783862113953,
+  'label': 'cat',
+  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
+```
+
+
+Aqui obtemos uma lista de objetos detectados na imagem, com uma caixa envolvendo o objeto e uma pontuação de confiança. Aqui está a imagem original à esquerda, com as previsões exibidas à direita:
+
+<h3 align="center">
+    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
+    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
+</h3>
+
+Você pode aprender mais sobre as tarefas suportadas pela API `pipeline` em [este tutorial](https://huggingface.co/docs/transformers/task_summary).
+
+
+Além do `pipeline`, para baixar e usar qualquer um dos modelos pré-treinados em sua tarefa específica, tudo o que é necessário são três linhas de código. Aqui está a versão em PyTorch:
+
+```python
+>>> from transformers import AutoTokenizer, AutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = AutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="pt")
+>>> outputs = model(**inputs)
+```
+
+E aqui está o código equivalente para TensorFlow:
+
+```python
+>>> from transformers import AutoTokenizer, TFAutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = TFAutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="tf")
+>>> outputs = model(**inputs)
+```
+
+O tokenizador é responsável por todo o pré-processamento que o modelo pré-treinado espera, e pode ser chamado diretamente em uma única string (como nos exemplos acima) ou em uma lista. Ele produzirá um dicionário que você pode usar no código subsequente ou simplesmente passar diretamente para o seu modelo usando o operador de descompactação de argumentos **.
+
+O modelo em si é um [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)  ou um [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)(dependendo do seu back-end) que você pode usar como de costume. [Este tutorial](https://huggingface.co/docs/transformers/training) explica como integrar esse modelo em um ciclo de treinamento clássico do PyTorch ou TensorFlow, ou como usar nossa API `Trainer` para ajuste fino rápido em um novo conjunto de dados.
+
+## Por que devo usar transformers?
+
+1. Modelos state-of-the-art fáceis de usar:
+    - Alto desempenho em compreensão e geração de linguagem natural, visão computacional e tarefas de áudio.
+    - Barreira de entrada baixa para educadores e profissionais.
+    - Poucas abstrações visíveis para o usuário, com apenas três classes para aprender.
+    - Uma API unificada para usar todos os nossos modelos pré-treinados.
+
+1. Menores custos de computação, menor pegada de carbono:
+    - Pesquisadores podem compartilhar modelos treinados em vez de treinar sempre do zero.
+    - Profissionais podem reduzir o tempo de computação e os custos de produção.
+    - Dezenas de arquiteturas com mais de 60.000 modelos pré-treinados em todas as modalidades.
+
+1. Escolha o framework certo para cada parte da vida de um modelo:
+    - Treine modelos state-of-the-art em 3 linhas de código.
+    - Mova um único modelo entre frameworks TF2.0/PyTorch/JAX à vontade.
+    - Escolha o framework certo de forma contínua para treinamento, avaliação e produção.
+
+1. Personalize facilmente um modelo ou um exemplo para atender às suas necessidades:
+    - Fornecemos exemplos para cada arquitetura para reproduzir os resultados publicados pelos autores originais.
+    - Os detalhes internos do modelo são expostos de maneira consistente.
+    - Os arquivos do modelo podem ser usados de forma independente da biblioteca para experimentos rápidos.
+
+## Por que não devo usar transformers?
+
+- Esta biblioteca não é uma caixa de ferramentas modular para construir redes neurais. O código nos arquivos do modelo não é refatorado com abstrações adicionais de propósito, para que os pesquisadores possam iterar rapidamente em cada um dos modelos sem se aprofundar em abstrações/arquivos adicionais.
+- A API de treinamento não é projetada para funcionar com qualquer modelo, mas é otimizada para funcionar com os modelos fornecidos pela biblioteca. Para loops de aprendizado de máquina genéricos, você deve usar outra biblioteca (possivelmente, [Accelerate](https://huggingface.co/docs/accelerate)).
+- Embora nos esforcemos para apresentar o maior número possível de casos de uso, os scripts em nossa [pasta de exemplos](https://github.com/huggingface/transformers/tree/main/examples) são apenas isso: exemplos. É esperado que eles não funcionem prontos para uso em seu problema específico e que seja necessário modificar algumas linhas de código para adaptá-los às suas necessidades.
+
+
+
+### Com pip
+
+Este repositório é testado no Python 3.8+, Flax 0.4.1+, PyTorch 1.10+ e TensorFlow 2.6+.
+
+Você deve instalar o 🤗 Transformers em um [ambiente virtual](https://docs.python.org/3/library/venv.html). Se você não está familiarizado com ambientes virtuais em Python, confira o [guia do usuário](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
+
+Primeiro, crie um ambiente virtual com a versão do Python que você vai usar e ative-o.
+
+Em seguida, você precisará instalar pelo menos um dos back-ends Flax, PyTorch ou TensorFlow.
+Consulte a [página de instalação do TensorFlow](https://www.tensorflow.org/install/), a [página de instalação do PyTorch](https://pytorch.org/get-started/locally/#start-locally) e/ou [Flax](https://github.com/google/flax#quick-install) e [Jax](https://github.com/google/jax#installation) páginas de instalação para obter o comando de instalação específico para a sua plataforma.
+
+Quando um desses back-ends estiver instalado, o 🤗 Transformers pode ser instalado usando pip da seguinte forma:
+
+```bash
+pip install transformers
+```
+Se você deseja experimentar com os exemplos ou precisa da versão mais recente do código e não pode esperar por um novo lançamento, você deve instalar a [biblioteca a partir do código-fonte](https://huggingface.co/docs/transformers/installation#installing-from-source).
+
+### Com conda
+
+Desde a versão v4.0.0 do Transformers, agora temos um canal conda: `huggingface`.
+
+O 🤗 Transformers pode ser instalado com conda da seguinte forma:
+
+```bash
+conda install -c huggingface transformers
+```
+
+Siga as páginas de instalação do Flax, PyTorch ou TensorFlow para ver como instalá-los com conda. 
+
+Siga as páginas de instalação do Flax, PyTorch ou TensorFlow para ver como instalá-los com o conda.
+
+> **_NOTA:_**  No Windows, você pode ser solicitado a ativar o Modo de Desenvolvedor para aproveitar o cache. Se isso não for uma opção para você, por favor nos avise [neste problema](https://github.com/huggingface/huggingface_hub/issues/1062).
+
+## Arquiteturas de Modelos
+
+**[Todos os pontos de verificação de modelo](https://huggingface.co/models)** fornecidos pelo 🤗 Transformers são integrados de forma transparente do [model hub](https://huggingface.co/models) do huggingface.co, onde são carregados diretamente por [usuários](https://huggingface.co/users) e [organizações](https://huggingface.co/organizations).
+
+Número atual de pontos de verificação: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
+
+🤗 Transformers atualmente fornece as seguintes arquiteturas (veja [aqui](https://huggingface.co/docs/transformers/model_summary) para um resumo de alto nível de cada uma delas):
+
+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. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (from Google Research) released with the paper [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.
+1. **[AltCLIP](https://huggingface.co/docs/transformers/model_doc/altclip)** (from BAAI) released with the paper [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679) by Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell.
+1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
+1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
+1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
+1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
+1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
+1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
+1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BioGpt](https://huggingface.co/docs/transformers/model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
+1. **[BiT](https://huggingface.co/docs/transformers/model_doc/bit)** (from Google AI) released with the paper [Big Transfer (BiT): General Visual Representation Learning](https://arxiv.org/abs/1912.11370) by Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil Houlsby.
+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. **[BLIP](https://huggingface.co/docs/transformers/model_doc/blip)** (from Salesforce) released with the paper [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://arxiv.org/abs/2201.12086) by Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi.
+1. **[BLIP-2](https://huggingface.co/docs/transformers/model_doc/blip-2)** (from Salesforce) released with the paper [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://arxiv.org/abs/2301.12597) by Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi.
+1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
+1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
+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. **[Chinese-CLIP](https://huggingface.co/docs/transformers/model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
+1. **[CLAP](https://huggingface.co/docs/transformers/model_doc/clap)** (from LAION-AI) released with the paper [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://arxiv.org/abs/2211.06687) by Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, Shlomo Dubnov.
+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. **[CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)** (from University of Göttingen) released with the paper [Image Segmentation Using Text and Image Prompts](https://arxiv.org/abs/2112.10003) by Timo Lüddecke and Alexander Ecker.
+1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
+1. **[CodeLlama](https://huggingface.co/docs/transformers/model_doc/llama_code)** (from MetaAI) released with the paper [Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/) by Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong, Alexandre Défossez, Jade Copet, Faisal Azhar, Hugo Touvron, Louis Martin, Nicolas Usunier, Thomas Scialom, Gabriel Synnaeve.
+1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
+1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
+1. **[ConvNeXTV2](https://huggingface.co/docs/transformers/model_doc/convnextv2)** (from Facebook AI) released with the paper [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) by Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, Saining Xie.
+1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+1. **[CPM-Ant](https://huggingface.co/docs/transformers/model_doc/cpmant)** (from OpenBMB) released by the [OpenBMB](https://www.openbmb.org/).
+1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
+1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
+1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
+1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
+1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
+1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
+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. **[DiNAT](https://huggingface.co/docs/transformers/model_doc/dinat)** (from SHI Labs) released with the paper [Dilated Neighborhood Attention Transformer](https://arxiv.org/abs/2209.15001) by Ali Hassani and Humphrey Shi.
+1. **[DINOv2](https://huggingface.co/docs/transformers/model_doc/dinov2)** (from Meta AI) released with the paper [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
+1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
+1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
+1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
+1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
+1. **[EfficientFormer](https://huggingface.co/docs/transformers/model_doc/efficientformer)** (from Snap Research) released with the paper [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191) by Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren.
+1. **[EfficientNet](https://huggingface.co/docs/transformers/model_doc/efficientnet)** (from Google Brain) released with the paper [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946) by Mingxing Tan, Quoc V. Le.
+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. **[EnCodec](https://huggingface.co/docs/transformers/model_doc/encodec)** (from Meta AI) released with the paper [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) by Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.
+1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
+1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (from Baidu) released with the paper [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) by Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang.
+1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2 and ESMFold** were released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
+1. **[Falcon](https://huggingface.co/docs/transformers/model_doc/falcon)** (from Technology Innovation Institute) by Almazrouei, Ebtesam and Alobeidli, Hamza and Alshamsi, Abdulaziz and Cappelli, Alessandro and Cojocaru, Ruxandra and Debbah, Merouane and Goffinet, Etienne and Heslow, Daniel and Launay, Julien and Malartic, Quentin and Noune, Badreddine and Pannier, Baptiste and Penedo, Guilherme.
+1. **[FLAN-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-t5-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
+1. **[FLAN-UL2](https://huggingface.co/docs/transformers/model_doc/flan-ul2)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-ul2-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
+1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
+1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
+1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
+1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
+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. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
+1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://openai.com/research/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
+1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://openai.com/research/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-Sw3](https://huggingface.co/docs/transformers/model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
+1. **[GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
+1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
+1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
+1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
+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. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh.
+1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
+1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
+1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
+1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
+1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
+1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
+1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
+1. **[LiLT](https://huggingface.co/docs/transformers/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
+1. **[LLaMA](https://huggingface.co/docs/transformers/model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
+1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
+1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
+1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
+1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
+1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
+1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
+1. **[Mask2Former](https://huggingface.co/docs/transformers/model_doc/mask2former)** (from FAIR and UIUC) released with the paper [Masked-attention Mask Transformer for Universal Image Segmentation](https://arxiv.org/abs/2112.01527) by Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, Rohit Girdhar.
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
+1. **[MatCha](https://huggingface.co/docs/transformers/model_doc/matcha)** (from Google AI) released with the paper [MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering](https://arxiv.org/abs/2212.09662) by Fangyu Liu, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Yasemin Altun, Nigel Collier, Julian Martin Eisenschlos.
+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. **[MEGA](https://huggingface.co/docs/transformers/model_doc/mega)** (from Meta/USC/CMU/SJTU) released with the paper [Mega: Moving Average Equipped Gated Attention](https://arxiv.org/abs/2209.10655) by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer.
+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. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The [Mistral AI](https://mistral.ai) team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
+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. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
+1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
+1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
+1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
+1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
+1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
+1. **[MPT](https://huggingface.co/docs/transformers/model_doc/mpt)** (from MosaiML) released with the repository [llm-foundry](https://github.com/mosaicml/llm-foundry/) by the MosaicML NLP Team.
+1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
+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. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
+1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
+1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
+1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
+1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
+1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
+1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
+1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
+1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (from ADEPT) released in a [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
+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. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
+1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
+1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
+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. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
+1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
+1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
+1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
+1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
+1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
+1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
+1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
+1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng), released on [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
+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. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
+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. **[SpeechT5](https://huggingface.co/docs/transformers/model_doc/speecht5)** (from Microsoft Research) released with the paper [SpeechT5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing](https://arxiv.org/abs/2110.07205) by Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, Zhihua Wei, Yao Qian, Jinyu Li, Furu Wei.
+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. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (from MBZUAI) released with the paper [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) by Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.
+1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
+1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
+1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (from University of Würzburg) released with the paper [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) by Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte.
+1. **[SwitchTransformers](https://huggingface.co/docs/transformers/model_doc/switch_transformers)** (from Google) released with the paper [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961) by William Fedus, Barret Zoph, Noam Shazeer.
+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. **[Table Transformer](https://huggingface.co/docs/transformers/model_doc/table-transformer)** (from Microsoft Research) released with the paper [PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents](https://arxiv.org/abs/2110.00061) by Brandon Smock, Rohith Pesala, Robin Abraham.
+1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
+1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
+1. **[TimeSformer](https://huggingface.co/docs/transformers/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
+1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
+1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
+1. **[TVLT](https://huggingface.co/docs/transformers/model_doc/tvlt)** (from UNC Chapel Hill) released with the paper [TVLT: Textless Vision-Language Transformer](https://arxiv.org/abs/2209.14156) by Zineng Tang, Jaemin Cho, Yixin Nie, Mohit Bansal.
+1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
+1. **[UMT5](https://huggingface.co/docs/transformers/model_doc/umt5)** (from Google Research) released with the paper [UniMax: Fairer and More Effective Language Sampling for Large-Scale Multilingual Pretraining](https://openreview.net/forum?id=kXwdL1cWOAi) by Hyung Won Chung, Xavier Garcia, Adam Roberts, Yi Tay, Orhan Firat, Sharan Narang, Noah Constant.
+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. **[UPerNet](https://huggingface.co/docs/transformers/model_doc/upernet)** (from Peking University) released with the paper [Unified Perceptual Parsing for Scene Understanding](https://arxiv.org/abs/1807.10221) by Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, Jian Sun.
+1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
+1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
+1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
+1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
+1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) rreleased with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
+1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
+1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
+1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
+1. **[X-MOD](https://huggingface.co/docs/transformers/model_doc/xmod)** (from Meta AI) released with the paper [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) by Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe.
+1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
+1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
+1. **[XLM-V](https://huggingface.co/docs/transformers/model_doc/xlm-v)** (from Meta AI) released with the paper [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://arxiv.org/abs/2301.10472) by Davis Liang, Hila Gonen, Yuning Mao, Rui Hou, Naman Goyal, Marjan Ghazvininejad, Luke Zettlemoyer, Madian Khabsa.
+1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
+1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
+1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, 
+Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+
+1. Quer contribuir com um novo modelo? Adicionamos um **guia detalhado e modelos de exemplo** para orientar você no processo de adição de um novo modelo. Você pode encontrá-los na pasta [`templates`](./templates) do repositório. Certifique-se de verificar as [diretrizes de contribuição](./CONTRIBUTING.md) e entrar em contato com os mantenedores ou abrir uma issue para coletar feedback antes de iniciar sua PR.
+
+Para verificar se cada modelo tem uma implementação em Flax, PyTorch ou TensorFlow, ou possui um tokenizador associado com a biblioteca 🤗 Tokenizers, consulte [esta tabela](https://huggingface.co/docs/transformers/index#supported-frameworks).
+
+Essas implementações foram testadas em vários conjuntos de dados (veja os scripts de exemplo) e devem corresponder ao desempenho das implementações originais. Você pode encontrar mais detalhes sobre o desempenho na seção de Exemplos da [documentação](https://github.com/huggingface/transformers/tree/main/examples).
+
+
+## Saiba mais
+
+| Seção | Descrição |
+|-|-|
+| [Documentação](https://huggingface.co/docs/transformers/) | Documentação completa da API e tutoriais |
+| [Resumo de Tarefas](https://huggingface.co/docs/transformers/task_summary) | Tarefas suportadas pelo 🤗 Transformers |
+| [Tutorial de Pré-processamento](https://huggingface.co/docs/transformers/preprocessing) | Usando a classe `Tokenizer` para preparar dados para os modelos |
+| [Treinamento e Ajuste Fino](https://huggingface.co/docs/transformers/training) | Usando os modelos fornecidos pelo 🤗 Transformers em um loop de treinamento PyTorch/TensorFlow e a API `Trainer` |
+| [Tour Rápido: Scripts de Ajuste Fino/Utilização](https://github.com/huggingface/transformers/tree/main/examples) | Scripts de exemplo para ajuste fino de modelos em uma ampla gama de tarefas |
+| [Compartilhamento e Envio de Modelos](https://huggingface.co/docs/transformers/model_sharing) | Envie e compartilhe seus modelos ajustados com a comunidade |
+
+## Citação
+
+Agora temos um [artigo](https://www.aclweb.org/anthology/2020.emnlp-demos.6/) que você pode citar para a biblioteca 🤗 Transformers:
+```bibtex
+@inproceedings{wolf-etal-2020-transformers,
+    title = "Transformers: State-of-the-Art Natural Language Processing",
+    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
+    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
+    month = out,
+    year = "2020",
+    address = "Online",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
+    pages = "38--45"
+}
+```
+
+
+
+
+
+
+
+
+
+
+

README_ru.md

@@ -0,0 +1,550 @@
+<!---
+Copyright 2023 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">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-dark.svg">
+    <source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg">
+    <img alt="Hugging Face Transformers Library" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg" width="352" height="59" style="max-width: 100%;">
+  </picture>
+  <br/>
+  <br/>
+</p>
+
+<p align="center">
+    <a href="https://circleci.com/gh/huggingface/transformers">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
+    </a>
+    <a href="https://huggingface.co/docs/transformers/index">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    </a>
+    <a href="https://github.com/huggingface/transformers/releases">
+        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
+    </a>
+    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
+</p>
+
+<h4 align="center">
+    <p>
+        <a href="https://github.com/huggingface/transformers/blob/main/README.md">English</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
+        <b>Русский</b>
+    <p>
+</h4>
+
+<h3 align="center">
+    <p>Современное машинное обучение для JAX, PyTorch и TensorFlow</p>
+</h3>
+
+<h3 align="center">
+    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
+</h3>
+
+🤗 Transformers предоставляет тысячи предварительно обученных моделей для выполнения различных задач, таких как текст, зрение и аудио.
+
+Эти модели могут быть применены к:
+
+* 📝 Тексту для таких задач, как классификация текстов, извлечение информации, ответы на вопросы, обобщение, перевод, генерация текстов на более чем 100 языках.
+* 🖼️ Изображениям для задач классификации изображений, обнаружения объектов и сегментации.
+* 🗣️ Аудио для задач распознавания речи и классификации аудио.
+
+Модели transformers также могут выполнять несколько задач, такие как ответы на табличные вопросы, распознавание оптических символов, извлечение информации из отсканированных документов, классификация видео и ответы на визуальные вопросы.
+
+🤗 Transformers предоставляет API для быстрой загрузки и использования предварительно обученных моделей, их тонкой настройки на собственных датасетах и последующего взаимодействия ими с сообществом на нашем [сайте](https://huggingface.co/models). В то же время каждый python модуль, определяющий архитектуру, полностью автономен и может быть модифицирован для проведения быстрых исследовательских экспериментов.
+
+🤗 Transformers опирается на три самые популярные библиотеки глубокого обучения - [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) и [TensorFlow](https://www.tensorflow.org/) - и легко интегрируется между ними. Это позволяет легко обучать модели с помощью одной из них, а затем загружать их для выводов с помощью другой.
+
+## Онлайн демонстрация
+
+Большинство наших моделей можно протестировать непосредственно на их страницах с [сайта](https://huggingface.co/models). Мы также предлагаем [привтаный хостинг моделей, контроль версий и API для выводов](https://huggingface.co/pricing) для публичных и частных моделей.
+
+Вот несколько примеров:
+
+В области NLP ( Обработка текстов на естественном языке ):
+- [Маскированное заполнение слов с помощью BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
+- [Распознавание сущностей с помощью Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
+- [Генерация текста с помощью GPT-2](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
+- [Выводы на естественном языке с помощью RoBERTa](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
+- [Обобщение с помощью BART](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
+- [Ответы на вопросы с помощью 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)
+- [Перевод с помощью T5](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
+
+В области компьютерного зрения:
+- [Классификация изображений с помощью ViT](https://huggingface.co/google/vit-base-patch16-224)
+- [Обнаружение объектов с помощью DETR](https://huggingface.co/facebook/detr-resnet-50)
+- [Семантическая сегментация с помощью SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
+- [Сегментация паноптикума с помощью MaskFormer](https://huggingface.co/facebook/maskformer-swin-small-coco)
+- [Оценка глубины с помощью DPT](https://huggingface.co/docs/transformers/model_doc/dpt)
+- [Классификация видео с помощью VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)
+- [Универсальная сегментация с помощью OneFormer](https://huggingface.co/shi-labs/oneformer_ade20k_dinat_large)
+
+В области звука:
+- [Автоматическое распознавание речи с помощью Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base-960h)
+- [Поиск ключевых слов с помощью Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
+- [Классификация аудиоданных с помощью траснформера аудиоспектрограмм](https://huggingface.co/MIT/ast-finetuned-audioset-10-10-0.4593)
+
+В мультимодальных задачах:
+- [Ответы на вопросы по таблице с помощью TAPAS](https://huggingface.co/google/tapas-base-finetuned-wtq)
+- [Визуальные ответы на вопросы с помощью ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
+- [Zero-shot классификация изображений с помощью CLIP](https://huggingface.co/openai/clip-vit-large-patch14)
+- [Ответы на вопросы по документам с помощью LayoutLM](https://huggingface.co/impira/layoutlm-document-qa)
+- [Zero-shot классификация видео с помощью X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)
+
+
+## 100 проектов, использующих Transformers
+
+Transformers - это не просто набор инструментов для использования предварительно обученных моделей: это сообщество проектов, созданное на его основе, и
+Hugging Face Hub. Мы хотим, чтобы Transformers позволил разработчикам, исследователям, студентам, профессорам, инженерам и всем желающим
+создавать проекты своей мечты.
+
+Чтобы отпраздновать 100 тысяч звезд Transformers, мы решили сделать акцент на сообществе, и создали страницу [awesome-transformers](./awesome-transformers.md), на которой перечислены 100
+невероятных проектов, созданных с помощью transformers.
+
+Если вы являетесь владельцем или пользователем проекта, который, по вашему мнению, должен быть включен в этот список, пожалуйста, откройте PR для его добавления!
+
+## Если вы хотите получить индивидуальную поддержку от команды Hugging Face
+
+<a target="_blank" href="https://huggingface.co/support">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+</a><br>
+
+## Быстрый гайд
+
+Для использования модели на заданном входе (текст, изображение, звук, ...) мы предоставляем API `pipeline`. Конвейеры объединяют предварительно обученную модель с препроцессингом, который использовался при ее обучении. Вот как можно быстро использовать конвейер для классификации положительных и отрицательных текстов:
+
+```python
+>>> from transformers import pipeline
+
+# Выделение конвейера для анализа настроений
+>>> classifier = pipeline('sentiment-analysis')
+>>> classifier('Мы очень рады представить конвейер в transformers.')
+[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
+```
+
+Вторая строка кода загружает и кэширует предварительно обученную модель, используемую конвейером, а третья оценивает ее на заданном тексте. Здесь ответ "POSITIVE" с уверенностью 99,97%.
+
+Во многих задачах, как в НЛП, так и в компьютерном зрении и речи, уже есть готовый `pipeline`. Например, мы можем легко извлечь обнаруженные объекты на изображении:
+
+``` python
+>>> import requests
+>>> from PIL import Image
+>>> from transformers import pipeline
+
+# Скачиваем изображение с милыми котиками
+>>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
+>>> image_data = requests.get(url, stream=True).raw
+>>> image = Image.open(image_data)
+
+# Выделение конвейера для обнаружения объектов
+>>> object_detector = pipeline('object-detection')
+>>> object_detector(image)
+[{'score': 0.9982201457023621,
+  'label': 'remote',
+  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
+ {'score': 0.9960021376609802,
+  'label': 'remote',
+  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
+ {'score': 0.9954745173454285,
+  'label': 'couch',
+  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
+ {'score': 0.9988006353378296,
+  'label': 'cat',
+  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
+ {'score': 0.9986783862113953,
+  'label': 'cat',
+  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
+```
+
+Здесь мы получаем список объектов, обнаруженных на изображении, с рамкой вокруг объекта и оценкой достоверности. Слева - исходное изображение, справа прогнозы:
+
+<h3 align="center">
+    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
+    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
+</h3>
+
+Подробнее о задачах, поддерживаемых API `pipeline`, можно узнать в [этом учебном пособии](https://huggingface.co/docs/transformers/task_sum)
+
+В дополнение к `pipeline`, для загрузки и использования любой из предварительно обученных моделей в заданной задаче достаточно трех строк кода. Вот версия для PyTorch:
+```python
+>>> from transformers import AutoTokenizer, AutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = AutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Привет мир!", return_tensors="pt")
+>>> outputs = model(**inputs)
+```
+
+А вот эквивалентный код для TensorFlow:
+```python
+>>> from transformers import AutoTokenizer, TFAutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = TFAutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Привет мир!", return_tensors="tf")
+>>> outputs = model(**inputs)
+```
+
+Токенизатор отвечает за всю предварительную обработку, которую ожидает предварительно обученная модель, и может быть вызван непосредственно с помощью одной строки (как в приведенных выше примерах) или на списке. В результате будет получен словарь, который можно использовать в последующем коде или просто напрямую передать в модель с помощью оператора распаковки аргументов **.
+
+Сама модель представляет собой обычный [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) или [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (в зависимости от используемого бэкенда), который можно использовать как обычно. [В этом руководстве](https://huggingface.co/docs/transformers/training) рассказывается, как интегрировать такую модель в классический цикл обучения PyTorch или TensorFlow, или как использовать наш API `Trainer` для быстрой тонкой настройки на новом датасете.
+
+## Почему необходимо использовать transformers?
+
+1. Простые в использовании современные модели:
+    - Высокая производительность в задачах понимания и генерации естественного языка, компьютерного зрения и аудио.
+    - Низкий входной барьер для преподавателей и практиков.
+    - Небольшое количество абстракций для пользователя и всего три класса для изучения.
+    - Единый API для использования всех наших предварительно обученных моделей.
+
+1. Более низкие вычислительные затраты, меньший "углеродный след":
+    - Исследователи могут обмениваться обученными моделями вместо того, чтобы постоянно их переобучать.
+    - Практики могут сократить время вычислений и производственные затраты.
+    - Десятки архитектур с более чем 60 000 предварительно обученных моделей для всех модальностей.
+
+1. Выбор подходящего фреймворка для каждого этапа жизни модели:
+    - Обучение самых современных моделей за 3 строки кода.
+    - Перемещайте одну модель между фреймворками TF2.0/PyTorch/JAX по своему усмотрению.
+    - Беспрепятственный выбор подходящего фреймворка для обучения, оценки и производства.
+
+1. Легко настроить модель или пример под свои нужды:
+    - Мы предоставляем примеры для каждой архитектуры, чтобы воспроизвести результаты, опубликованные их авторами.
+    - Внутренние компоненты модели раскрываются максимально последовательно.
+    - Файлы моделей можно использовать независимо от библиотеки для проведения быстрых экспериментов.
+
+## Почему я не должен использовать transformers?
+
+- Данная библиотека не является модульным набором строительных блоков для нейронных сетей. Код в файлах моделей специально не рефакторится дополнительными абстракциями, чтобы исследователи могли быстро итеративно работать с каждой из моделей, не погружаясь в дополнительные абстракции/файлы.
+- API обучения не предназначен для работы с любой моделью, а оптимизирован для работы с моделями, предоставляемыми библиотекой. Для работы с общими циклами машинного обучения следует использовать другую библиотеку (возможно, [Accelerate](https://huggingface.co/docs/accelerate)).
+- Несмотря на то, что мы стремимся представить как можно больше примеров использования, скрипты в нашей папке [примеров](https://github.com/huggingface/transformers/tree/main/examples) являются именно примерами. Предполагается, что они не будут работать "из коробки" для решения вашей конкретной задачи, и вам придется изменить несколько строк кода, чтобы адаптировать их под свои нужды.
+
+## Установка
+
+### С помощью pip
+
+Данный репозиторий протестирован на Python 3.8+, Flax 0.4.1+, PyTorch 1.10+ и TensorFlow 2.6+.
+
+Устанавливать 🤗 Transformers следует в [виртуальной среде](https://docs.python.org/3/library/venv.html). Если вы не знакомы с виртуальными средами Python, ознакомьтесь с [руководством пользователя](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
+
+Сначала создайте виртуальную среду с той версией Python, которую вы собираетесь использовать, и активируйте ее.
+
+Затем необходимо установить хотя бы один бекенд из Flax, PyTorch или TensorFlow.
+Пожалуйста, обратитесь к страницам [TensorFlow установочная страница](https://www.tensorflow.org/install/), [PyTorch установочная страница](https://pytorch.org/get-started/locally/#start-locally) и/или [Flax](https://github.com/google/flax#quick-install) и [Jax](https://github.com/google/jax#installation), где описаны команды установки для вашей платформы.
+
+После установки одного из этих бэкендов 🤗 Transformers может быть установлен с помощью pip следующим образом:
+
+```bash
+pip install transformers
+```
+
+Если вы хотите поиграть с примерами или вам нужен самый современный код и вы не можете ждать нового релиза, вы должны [установить библиотеку из исходного кода](https://huggingface.co/docs/transformers/installation#installing-from-source).
+
+### С помощью conda
+
+Начиная с версии Transformers v4.0.0, у нас появилсась поддержка conda: `huggingface`.
+
+Установить Transformers с помощью conda можно следующим образом:
+
+```bash
+conda install -c huggingface transformers
+```
+
+О том, как установить Flax, PyTorch или TensorFlow с помощью conda, читайте на страницах, посвященных их установке.
+
+> **_ЗАМЕТКА:_** В операционной системе Windows вам может быть предложено активировать режим разработчика, чтобы воспользоваться преимуществами кэширования. Если для вас это невозможно, сообщите нам об этом [здесь](https://github.com/huggingface/huggingface_hub/issues/1062).
+
+## Модельные архитектуры
+
+**[Все контрольные точки моделей](https://huggingface.co/models)**, предоставляемые 🤗 Transformers, беспрепятственно интегрируются с huggingface.co [model hub](https://huggingface.co/models), куда они загружаются непосредственно [пользователями](https://huggingface.co/users) и [организациями](https://huggingface.co/organizations).
+
+Текущее количество контрольных точек: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
+
+🤗 В настоящее время Transformers предоставляет следующие архитектуры (подробное описание каждой из них см. [здесь](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. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (from Google Research) released with the paper [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.
+1. **[AltCLIP](https://huggingface.co/docs/transformers/model_doc/altclip)** (from BAAI) released with the paper [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679) by Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell.
+1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
+1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
+1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
+1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
+1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
+1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
+1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BioGpt](https://huggingface.co/docs/transformers/model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
+1. **[BiT](https://huggingface.co/docs/transformers/model_doc/bit)** (from Google AI) released with the paper [Big Transfer (BiT): General Visual Representation Learning](https://arxiv.org/abs/1912.11370) by Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil Houlsby.
+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. **[BLIP](https://huggingface.co/docs/transformers/model_doc/blip)** (from Salesforce) released with the paper [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://arxiv.org/abs/2201.12086) by Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi.
+1. **[BLIP-2](https://huggingface.co/docs/transformers/model_doc/blip-2)** (from Salesforce) released with the paper [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://arxiv.org/abs/2301.12597) by Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi.
+1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
+1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
+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. **[Chinese-CLIP](https://huggingface.co/docs/transformers/model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
+1. **[CLAP](https://huggingface.co/docs/transformers/model_doc/clap)** (from LAION-AI) released with the paper [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://arxiv.org/abs/2211.06687) by Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, Shlomo Dubnov.
+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. **[CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)** (from University of Göttingen) released with the paper [Image Segmentation Using Text and Image Prompts](https://arxiv.org/abs/2112.10003) by Timo Lüddecke and Alexander Ecker.
+1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
+1. **[CodeLlama](https://huggingface.co/docs/transformers/model_doc/llama_code)** (from MetaAI) released with the paper [Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/) by Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong, Alexandre Défossez, Jade Copet, Faisal Azhar, Hugo Touvron, Louis Martin, Nicolas Usunier, Thomas Scialom, Gabriel Synnaeve.
+1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
+1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
+1. **[ConvNeXTV2](https://huggingface.co/docs/transformers/model_doc/convnextv2)** (from Facebook AI) released with the paper [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) by Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, Saining Xie.
+1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+1. **[CPM-Ant](https://huggingface.co/docs/transformers/model_doc/cpmant)** (from OpenBMB) released by the [OpenBMB](https://www.openbmb.org/).
+1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
+1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
+1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
+1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
+1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
+1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
+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. **[DiNAT](https://huggingface.co/docs/transformers/model_doc/dinat)** (from SHI Labs) released with the paper [Dilated Neighborhood Attention Transformer](https://arxiv.org/abs/2209.15001) by Ali Hassani and Humphrey Shi.
+1. **[DINOv2](https://huggingface.co/docs/transformers/model_doc/dinov2)** (from Meta AI) released with the paper [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
+1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
+1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
+1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
+1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
+1. **[EfficientFormer](https://huggingface.co/docs/transformers/model_doc/efficientformer)** (from Snap Research) released with the paper [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191) by Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren.
+1. **[EfficientNet](https://huggingface.co/docs/transformers/model_doc/efficientnet)** (from Google Brain) released with the paper [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946) by Mingxing Tan, Quoc V. Le.
+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. **[EnCodec](https://huggingface.co/docs/transformers/model_doc/encodec)** (from Meta AI) released with the paper [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) by Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.
+1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
+1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (from Baidu) released with the paper [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) by Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang.
+1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2 and ESMFold** were released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
+1. **[Falcon](https://huggingface.co/docs/transformers/model_doc/falcon)** (from Technology Innovation Institute) by Almazrouei, Ebtesam and Alobeidli, Hamza and Alshamsi, Abdulaziz and Cappelli, Alessandro and Cojocaru, Ruxandra and Debbah, Merouane and Goffinet, Etienne and Heslow, Daniel and Launay, Julien and Malartic, Quentin and Noune, Badreddine and Pannier, Baptiste and Penedo, Guilherme.
+1. **[FLAN-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-t5-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
+1. **[FLAN-UL2](https://huggingface.co/docs/transformers/model_doc/flan-ul2)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-ul2-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
+1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
+1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
+1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
+1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
+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. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
+1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
+1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
+1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
+1. **[GPT-Sw3](https://huggingface.co/docs/transformers/model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
+1. **[GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
+1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
+1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
+1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
+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. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh.
+1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
+1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
+1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
+1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
+1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
+1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
+1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
+1. **[LiLT](https://huggingface.co/docs/transformers/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
+1. **[LLaMA](https://huggingface.co/docs/transformers/model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
+1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/XXX) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
+1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
+1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
+1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
+1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
+1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
+1. **[Mask2Former](https://huggingface.co/docs/transformers/model_doc/mask2former)** (from FAIR and UIUC) released with the paper [Masked-attention Mask Transformer for Universal Image Segmentation](https://arxiv.org/abs/2112.01527) by Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, Rohit Girdhar.
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
+1. **[MatCha](https://huggingface.co/docs/transformers/model_doc/matcha)** (from Google AI) released with the paper [MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering](https://arxiv.org/abs/2212.09662) by Fangyu Liu, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Yasemin Altun, Nigel Collier, Julian Martin Eisenschlos.
+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. **[MEGA](https://huggingface.co/docs/transformers/model_doc/mega)** (from Meta/USC/CMU/SJTU) released with the paper [Mega: Moving Average Equipped Gated Attention](https://arxiv.org/abs/2209.10655) by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer.
+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. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
+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. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
+1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
+1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
+1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
+1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
+1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
+1. **[MPT](https://huggingface.co/docs/transformers/model_doc/mpt)** (from MosaiML) released with the repository [llm-foundry](https://github.com/mosaicml/llm-foundry/) by the MosaicML NLP Team.
+1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
+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. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
+1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
+1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
+1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
+1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
+1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
+1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
+1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[Persimmon](https://huggingface.co/docs/transformers/main/model_doc/persimmon)** (from ADEPT) released in a [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
+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. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
+1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
+1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
+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. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
+1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
+1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
+1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
+1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
+1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
+1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
+1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
+1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng), released on [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
+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. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
+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. **[SpeechT5](https://huggingface.co/docs/transformers/model_doc/speecht5)** (from Microsoft Research) released with the paper [SpeechT5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing](https://arxiv.org/abs/2110.07205) by Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, Zhihua Wei, Yao Qian, Jinyu Li, Furu Wei.
+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. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (from MBZUAI) released with the paper [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) by Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.
+1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
+1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
+1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (from University of Würzburg) released with the paper [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) by Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte.
+1. **[SwitchTransformers](https://huggingface.co/docs/transformers/model_doc/switch_transformers)** (from Google) released with the paper [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961) by William Fedus, Barret Zoph, Noam Shazeer.
+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. **[Table Transformer](https://huggingface.co/docs/transformers/model_doc/table-transformer)** (from Microsoft Research) released with the paper [PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents](https://arxiv.org/abs/2110.00061) by Brandon Smock, Rohith Pesala, Robin Abraham.
+1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
+1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
+1. **[TimeSformer](https://huggingface.co/docs/transformers/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
+1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
+1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
+1. **[TVLT](https://huggingface.co/docs/transformers/model_doc/tvlt)** (from UNC Chapel Hill) released with the paper [TVLT: Textless Vision-Language Transformer](https://arxiv.org/abs/2209.14156) by Zineng Tang, Jaemin Cho, Yixin Nie, Mohit Bansal.
+1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
+1. **[UMT5](https://huggingface.co/docs/transformers/model_doc/umt5)** (from Google Research) released with the paper [UniMax: Fairer and More Effective Language Sampling for Large-Scale Multilingual Pretraining](https://openreview.net/forum?id=kXwdL1cWOAi) by Hyung Won Chung, Xavier Garcia, Adam Roberts, Yi Tay, Orhan Firat, Sharan Narang, Noah Constant.
+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. **[UPerNet](https://huggingface.co/docs/transformers/model_doc/upernet)** (from Peking University) released with the paper [Unified Perceptual Parsing for Scene Understanding](https://arxiv.org/abs/1807.10221) by Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, Jian Sun.
+1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
+1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
+1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
+1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
+1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/main/model_doc/vitmatte)** (from HUST-VL) rreleased with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
+1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
+1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
+1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
+1. **[X-MOD](https://huggingface.co/docs/transformers/model_doc/xmod)** (from Meta AI) released with the paper [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) by Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe.
+1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
+1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
+1. **[XLM-V](https://huggingface.co/docs/transformers/model_doc/xlm-v)** (from Meta AI) released with the paper [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://arxiv.org/abs/2301.10472) by Davis Liang, Hila Gonen, Yuning Mao, Rui Hou, Naman Goyal, Marjan Ghazvininejad, Luke Zettlemoyer, Madian Khabsa.
+1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
+1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
+1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+1. 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.
+
+Чтобы проверить, есть ли у каждой модели реализация на Flax, PyTorch или TensorFlow, или связанный с ней токенизатор, поддерживаемый библиотекой 🤗 Tokenizers, обратитесь к [этой таблице](https://huggingface.co/docs/transformers/index#supported-frameworks).
+
+Эти реализации были протестированы на нескольких наборах данных (см. примеры скриптов) и должны соответствовать производительности оригинальных реализаций. Более подробную информацию о производительности можно найти в разделе "Примеры" [документации](https://github.com/huggingface/transformers/tree/main/examples).
+
+
+## Изучи больше
+
+| Секция | Описание |
+|-|-|
+| [Документация](https://huggingface.co/docs/transformers/) | Полная документация по API и гайды |
+| [Краткие описания задач](https://huggingface.co/docs/transformers/task_summary) | Задачи поддерживаются 🤗 Transformers |
+| [Пособие по предварительной обработке](https://huggingface.co/docs/transformers/preprocessing) | Использование класса `Tokenizer` для подготовки данных для моделей |
+| [Обучение и доработка](https://huggingface.co/docs/transformers/training) | Использование моделей, предоставляемых 🤗 Transformers, в цикле обучения PyTorch/TensorFlow и API `Trainer`. |
+| [Быстрый тур: Тонкая настройка/скрипты использования](https://github.com/huggingface/transformers/tree/main/examples) | Примеры скриптов для тонкой настройки моделей на широком спектре задач |
+| [Совместное использование и загрузка моделей](https://huggingface.co/docs/transformers/model_sharing) | Загружайте и делитесь с сообществом своими доработанными моделями |
+
+## Цитирование
+
+Теперь у нас есть [статья](https://www.aclweb.org/anthology/2020.emnlp-demos.6/), которую можно цитировать для библиотеки 🤗 Transformers:
+```bibtex
+@inproceedings{wolf-etal-2020-transformers,
+    title = "Transformers: State-of-the-Art Natural Language Processing",
+    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
+    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
+    month = oct,
+    year = "2020",
+    address = "Online",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
+    pages = "38--45"
+}
+```

README_zh-hans.md

@@ -43,7 +43,7 @@ checkpoint: 检查点
     <br>
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
-<p>
+</p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
         <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
@@ -72,7 +72,7 @@ checkpoint: 检查点
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-    <p>
+    </p>
 </h4>
 
 <h3 align="center">
@@ -260,6 +260,7 @@ conda install -c huggingface transformers
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (来自 Alexa) 伴随论文 [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) 由 Adrian de Wynter and Daniel J. Perry 发布。
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (来自 NAVER CLOVA) 伴随论文 [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) 由 Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park 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。
@@ -310,6 +311,7 @@ conda install -c huggingface transformers
 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. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (来自 Microsoft Research) 伴随论文 [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) 由 Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao 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. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (来自 ADEPT) 伴随论文 [blog post](https://www.adept.ai/blog/fuyu-8b 由 Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar 发布。)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (来自 Microsoft Research) 伴随论文 [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) 由 Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang 发布。
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (来自 KAIST) 伴随论文 [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) 由 Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim 发布。
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (来自 OpenAI) 伴随论文 [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) 由 Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever 发布。
@@ -323,6 +325,7 @@ conda install -c huggingface transformers
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by 坂本俊之(tanreinama).
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (来自 UCSD, NVIDIA) 伴随论文 [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) 由 Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang 发布。
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (来自 Allegro.pl, AGH University of Science and Technology) 伴随论文 [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) 由 Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik 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. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh. 
@@ -356,6 +359,7 @@ conda install -c huggingface transformers
 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. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (来自 Alibaba Research) 伴随论文 [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) 由 Peng Wang, Cheng Da, and Cong Yao 发布。
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The Mistral AI team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.. 
 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. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (来自 Facebook) 伴随论文 [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) 由 Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli 发布。
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (来自 CMU/Google Brain) 伴随论文 [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) 由 Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou 发布。
@@ -373,19 +377,22 @@ conda install -c huggingface transformers
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (来自华为诺亚方舟实验室) 伴随论文 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 由 Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 发布。
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (来自 Meta AI) 伴随论文 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) 由 Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic 发布。
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (来自 the University of Wisconsin - Madison) 伴随论文 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 由 Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 发布。
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (来自 SHI Labs)  伴随论文 [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) 由 Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi 发布。
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (来自 [s-JoL](https://huggingface.co/s-JoL)) 由 [Open-Llama](https://github.com/s-JoL/Open-Llama) 发布.
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (来自 Meta AI) 伴随论文 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 由 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 发布。
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (来自 Google AI) 伴随论文 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 由 Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 发布。
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (来自 Google AI) 伴随论文 [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) 由 Matthias Minderer, Alexey Gritsenko, Neil Houlsby 发布。
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (来自 Google) 伴随论文 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 由 Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 发布。
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (来自 Google) 伴随论文 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 由 Jason Phang, Yao Zhao, Peter J. Liu 发布。
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (来自 Deepmind) 伴随论文 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 由 Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira 发布。
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (来自 ADEPT) 伴随论文 [blog post](https://www.adept.ai/blog/persimmon-8b) 由 Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani 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. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (来自 Google) 伴随论文 [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) 由 Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova 发布。
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (来自 UCLA NLP) 伴随论文 [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) 由 Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang 发布。
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (来自 Sea AI Labs) 伴随论文 [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) 由 Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng 发布。
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
 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. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (来自 Nanjing University, The University of Hong Kong etc.) 伴随论文 [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) 由 Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao 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。
@@ -436,9 +443,11 @@ conda install -c huggingface transformers
 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. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (来自 Meta AI) 伴随论文 [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) 由 Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He 发布。
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (来自 Meta AI) 伴随论文 [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) 由 Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He 发布。
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (来自 Meta AI) 伴随论文 [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) 由 Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick 发布。
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (来自 HUST-VL) 伴随论文 [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) 由 Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang 发布。
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (来自 Meta AI) 伴随论文 [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas 发布.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (来自 Kakao Enterprise) 伴随论文 [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) 由 Jaehyeon Kim, Jungil Kong, Juhee Son 发布。
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (来自 Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) 由 Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
 1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (来自 Facebook AI) 伴随论文 [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) 由 Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli 发布。
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (来自 Facebook AI) 伴随论文 [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) 由 Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino 发布。
@@ -457,7 +466,7 @@ conda install -c huggingface transformers
 1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (来自 Facebook AI) 伴随论文 [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) 由 Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli 发布。
 1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (来自 Facebook AI) 伴随论文 [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) 由 Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli 发布。
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (来自 Huazhong University of Science & Technology) 伴随论文 [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) 由 Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu 发布。
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (来自 the University of Wisconsin - Madison) 伴随论文 [You Only Sample (Almost) 由 Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh 发布。
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (来自 the University of Wisconsin - Madison) 伴随论文 [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) 由 Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh 发布。
 1. 想要贡献新的模型？我们这里有一份**详细指引和模板**来引导你添加新的模型。你可以在 [`templates`](./templates) 目录中找到他们。记得查看 [贡献指南](./CONTRIBUTING.md) 并在开始写 PR 前联系维护人员或开一个新的 issue 来获得反馈。
 
 要检查某个模型是否已有 Flax、PyTorch 或 TensorFlow 的实现，或其是否在 🤗 Tokenizers 库中有对应词符化器（tokenizer），敬请参阅[此表](https://huggingface.co/docs/transformers/index#supported-frameworks)。

README_zh-hant.md

@@ -55,7 +55,7 @@ user: 使用者
     <br>
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
     <br>
-<p>
+</p>
 <p align="center">
     <a href="https://circleci.com/gh/huggingface/transformers">
         <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
@@ -84,7 +84,7 @@ user: 使用者
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-    <p>
+    </p>
 </h4>
 
 <h3 align="center">
@@ -272,6 +272,7 @@ conda install -c huggingface transformers
 1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
 1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
 1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
+1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
 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.
@@ -322,6 +323,7 @@ conda install -c huggingface transformers
 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. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 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. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -335,6 +337,7 @@ conda install -c huggingface transformers
 1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by 坂本俊之(tanreinama).
 1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
 1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
 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. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh. 
@@ -368,6 +371,7 @@ conda install -c huggingface transformers
 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. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
+1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The Mistral AI team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.. 
 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. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
 1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
@@ -385,19 +389,22 @@ conda install -c huggingface transformers
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (from ADEPT) released with the paper [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
 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. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
 1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
 1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/main/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
+1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi, Kyogu Lee. 
 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. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
 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.
@@ -448,9 +455,11 @@ conda install -c huggingface transformers
 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. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (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. **[VitDet](https://huggingface.co/docs/transformers/main/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
+1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) released with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
+1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
 1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
@@ -469,7 +478,7 @@ conda install -c huggingface transformers
 1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
 1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
 1. 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)。

docker/transformers-all-latest-gpu/Dockerfile

@@ -9,7 +9,7 @@ SHELL ["sh", "-lc"]
 # The following `ARG` are mainly used to specify the versions explicitly & directly in this docker file, and not meant
 # to be used as arguments for docker build (so far).
 
-ARG PYTORCH='2.0.1'
+ARG PYTORCH='2.1.0'
 # (not always a valid torch version)
 ARG INTEL_TORCH_EXT='1.11.0'
 # Example: `cu102`, `cu113`, etc.

docker/transformers-doc-builder/Dockerfile

@@ -11,7 +11,6 @@ RUN apt-get -y update && apt-get install -y libsndfile1-dev && apt install -y te
 RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed]
 
 RUN python3 -m pip install --no-cache-dir torchvision git+https://github.com/facebookresearch/detectron2.git pytesseract
-RUN python3 -m pip install --no-cache-dir pytorch-quantization --extra-index-url https://pypi.ngc.nvidia.com
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
 
 # Test if the image could successfully build the doc. before publishing the image

docker/transformers-pytorch-amd-gpu/Dockerfile

@@ -0,0 +1,31 @@
+FROM rocm/pytorch:rocm5.6_ubuntu20.04_py3.8_pytorch_2.0.1
+LABEL maintainer="Hugging Face"
+
+ARG DEBIAN_FRONTEND=noninteractive
+
+RUN apt update && \
+    apt install -y --no-install-recommends git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg && \
+    apt clean && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip setuptools ninja git+https://github.com/facebookresearch/detectron2.git pytesseract "itsdangerous<2.1.0"
+
+# If set to nothing, will install the latest version
+ARG PYTORCH='2.0.1'
+ARG TORCH_VISION='0.15.2'
+ARG TORCH_AUDIO='2.0.2'
+ARG ROCM='5.6'
+
+RUN git clone --depth 1 --branch v$TORCH_AUDIO https://github.com/pytorch/audio.git
+RUN cd audio && USE_ROCM=1 USE_CUDA=0 python setup.py install
+
+ARG REF=main
+WORKDIR /
+RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
+RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing,video]
+
+RUN python3 -m pip uninstall -y tensorflow flax
+
+# When installing in editable mode, `transformers` is not recognized as a package.
+# this line must be added in order for python to be aware of transformers.
+RUN cd transformers && python3 setup.py develop

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

@@ -4,7 +4,7 @@ LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
-ARG PYTORCH='2.0.1'
+ARG PYTORCH='2.1.0'
 # Example: `cu102`, `cu113`, etc.
 ARG CUDA='cu118'
 
@@ -36,7 +36,8 @@ RUN python3 -m pip uninstall -y torch-tensorrt
 RUN python3 -m pip uninstall -y apex
 RUN git clone https://github.com/NVIDIA/apex
 #  `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
-RUN cd apex && git checkout 82ee367f3da74b4cd62a1fb47aa9806f0f47b58b && MAX_JOBS=1 python3 -m pip install --global-option="--cpp_ext" --global-option="--cuda_ext" --no-cache -v --disable-pip-version-check .
+# TODO: check if there is alternative way to install latest apex
+# RUN cd apex && MAX_JOBS=1 python3 -m pip install --global-option="--cpp_ext" --global-option="--cuda_ext" --no-cache -v --disable-pip-version-check .
 
 # Pre-build **latest** DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
 RUN python3 -m pip uninstall -y deepspeed

docker/transformers-pytorch-gpu/Dockerfile

@@ -9,10 +9,9 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing,video]
 
 # If set to nothing, will install the latest version
-ARG PYTORCH='2.0.1'
+ARG PYTORCH='2.1.0'
 ARG TORCH_VISION=''
 ARG TORCH_AUDIO=''
 # Example: `cu102`, `cu113`, etc.
@@ -22,6 +21,8 @@ RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch';
 RUN [ ${#TORCH_VISION} -gt 0 ] && VERSION='torchvision=='TORCH_VISION'.*' ||  VERSION='torchvision'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/$CUDA
 RUN [ ${#TORCH_AUDIO} -gt 0 ] && VERSION='torchaudio=='TORCH_AUDIO'.*' ||  VERSION='torchaudio'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/$CUDA
 
+RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing,video]
+
 RUN python3 -m pip uninstall -y tensorflow flax
 
 RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract

docs/README.md

@@ -81,7 +81,7 @@ The `preview` command only works with existing doc files. When you add a complet
 
 ## Adding a new element to the navigation bar
 
-Accepted files are Markdown (.md or .md).
+Accepted files are 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 in the [`_toctree.yml`](https://github.com/huggingface/transformers/blob/main/docs/source/en/_toctree.yml) file.
@@ -364,9 +364,6 @@ We use pytests' [doctest integration](https://docs.pytest.org/doctest.html) to v
 For Transformers, the doctests are run on a daily basis via GitHub Actions as can be 
 seen [here](https://github.com/huggingface/transformers/actions/workflows/doctests.yml).
 
-To include your example in the daily doctests, you need to add the filename that
-contains the example docstring to the [documentation_tests.txt](../utils/documentation_tests.txt).
-
 ### For Python files
 
 Run all the tests in the docstrings of a given file with the following command, here is how we test the modeling file of Wav2Vec2 for instance:

docs/source/de/_toctree.yml

@@ -15,8 +15,28 @@
     title: Vorverarbeiten
   - local: training
     title: Optimierung eines vortrainierten Modells
+  - local: run_scripts
+    title: Trainieren mit einem Skript
   - local: accelerate
     title: Verteiltes Training mit 🤗 Accelerate
+  - local: peft
+    title: Laden und Trainieren von Adaptern mit 🤗 PEFT
   - local: model_sharing
     title: Ein Modell teilen
+  - local: transformers_agents
+    title: Agents
+  - local: llm_tutorial
+    title: Generation with LLMs
   title: Tutorials
+- sections:
+  - local: add_new_model
+    title: Wie fügt man ein Modell zu 🤗 Transformers hinzu?
+  - local: add_tensorflow_model
+    title: Wie konvertiert man ein 🤗 Transformers-Modell in TensorFlow?
+  - local: add_new_pipeline
+    title: Wie fügt man eine Pipeline zu 🤗 Transformers hinzu?
+  - local: testing
+    title: Testen
+  - local: pr_checks
+    title: Überprüfung einer Pull Request
+  title: Contribute

docs/source/de/add_new_model.md

@@ -0,0 +1,895 @@
+<!--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
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Wie kann ich ein Modell zu 🤗 Transformers hinzufügen?
+
+Die 🤗 Transformers-Bibliothek ist dank der Beiträge der Community oft in der Lage, neue Modelle anzubieten. Aber das kann ein anspruchsvolles Projekt sein und erfordert eine eingehende Kenntnis der 🤗 Transformers-Bibliothek und des zu implementierenden Modells. Bei Hugging Face versuchen wir, mehr Mitgliedern der Community die Möglichkeit zu geben, aktiv Modelle hinzuzufügen, und wir haben diese Anleitung zusammengestellt, die Sie durch den Prozess des Hinzufügens eines PyTorch-Modells führt (stellen Sie sicher, dass Sie [PyTorch installiert haben](https://pytorch.org/get-started/locally/)).
+
+<Tip>
+
+Wenn Sie daran interessiert sind, ein TensorFlow-Modell zu implementieren, werfen Sie einen Blick in die Anleitung [How to convert a 🤗 Transformers model to TensorFlow](add_tensorflow_model)!
+
+</Tip>
+
+Auf dem Weg dorthin, werden Sie:
+
+- Einblicke in bewährte Open-Source-Verfahren erhalten
+- die Konstruktionsprinzipien hinter einer der beliebtesten Deep-Learning-Bibliotheken verstehen
+- lernen Sie, wie Sie große Modelle effizient testen können
+- lernen Sie, wie Sie Python-Hilfsprogramme wie `black`, `ruff` und `make fix-copies` integrieren, um sauberen und lesbaren Code zu gewährleisten
+
+Ein Mitglied des Hugging Face-Teams wird Ihnen dabei zur Seite stehen, damit Sie nicht alleine sind. 🤗 ❤️
+
+Um loszulegen, öffnen Sie eine [New model addition](https://github.com/huggingface/transformers/issues/new?assignees=&labels=New+model&template=new-model-addition.yml) Ausgabe für das Modell, das Sie in 🤗 Transformers sehen möchten. Wenn Sie nicht besonders wählerisch sind, wenn es darum geht, ein bestimmtes Modell beizusteuern, können Sie nach dem [New model label](https://github.com/huggingface/transformers/labels/New%20model) filtern, um zu sehen, ob es noch unbeanspruchte Modellanfragen gibt, und daran arbeiten.
+
+Sobald Sie eine neue Modellanfrage eröffnet haben, sollten Sie sich zunächst mit 🤗 Transformers vertraut machen, falls Sie das noch nicht sind!
+
+## Allgemeiner Überblick über 🤗 Transformers
+
+Zunächst sollten Sie sich einen allgemeinen Überblick über 🤗 Transformers verschaffen. 🤗 Transformers ist eine sehr meinungsfreudige Bibliothek, es ist also möglich, dass
+Es besteht also die Möglichkeit, dass Sie mit einigen der Philosophien oder Designentscheidungen der Bibliothek nicht einverstanden sind. Aus unserer Erfahrung heraus haben wir jedoch
+dass die grundlegenden Designentscheidungen und Philosophien der Bibliothek entscheidend sind, um 🤗 Transformers effizient zu skalieren.
+Transformatoren zu skalieren und gleichzeitig die Wartungskosten auf einem vernünftigen Niveau zu halten.
+
+Ein guter erster Ansatzpunkt, um die Bibliothek besser zu verstehen, ist die Lektüre der [Dokumentation unserer Philosophie](Philosophie). Als Ergebnis unserer Arbeitsweise gibt es einige Entscheidungen, die wir versuchen, auf alle Modelle anzuwenden:
+
+- Komposition wird im Allgemeinen gegenüber Abstraktion bevorzugt
+- Die Duplizierung von Code ist nicht immer schlecht, wenn sie die Lesbarkeit oder Zugänglichkeit eines Modells stark verbessert
+- Modelldateien sind so in sich geschlossen wie möglich, so dass Sie, wenn Sie den Code eines bestimmten Modells lesen, idealerweise nur
+  in die entsprechende Datei `modeling_....py` schauen müssen.
+
+Unserer Meinung nach ist der Code der Bibliothek nicht nur ein Mittel, um ein Produkt bereitzustellen, *z.B.* die Möglichkeit, BERT für
+Inferenz zu verwenden, sondern auch als das Produkt selbst, das wir verbessern wollen. Wenn Sie also ein Modell hinzufügen, ist der Benutzer nicht nur die
+Person, die Ihr Modell verwenden wird, sondern auch jeder, der Ihren Code liest, zu verstehen versucht und ihn möglicherweise verbessert.
+
+Lassen Sie uns daher ein wenig tiefer in das allgemeine Design der Bibliothek einsteigen.
+
+### Überblick über die Modelle
+
+Um ein Modell erfolgreich hinzuzufügen, ist es wichtig, die Interaktion zwischen Ihrem Modell und seiner Konfiguration zu verstehen,
+[`PreTrainedModel`] und [`PretrainedConfig`]. Als Beispiel werden wir
+das Modell, das zu 🤗 Transformers hinzugefügt werden soll, `BrandNewBert` nennen.
+
+Schauen wir uns das mal an:
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_overview.png"/>
+
+Wie Sie sehen, machen wir in 🤗 Transformers von der Vererbung Gebrauch, aber wir beschränken die Abstraktionsebene auf ein absolutes Minimum.
+Minimum. Es gibt nie mehr als zwei Abstraktionsebenen für ein Modell in der Bibliothek. `BrandNewBertModel`
+erbt von `BrandNewBertPreTrainedModel`, das wiederum von [`PreTrainedModel`] erbt und
+das war's. In der Regel wollen wir sicherstellen, dass ein neues Modell nur von
+[`PreTrainedModel`] abhängt. Die wichtigen Funktionalitäten, die jedem neuen Modell automatisch zur Verfügung gestellt werden, sind
+Modell automatisch bereitgestellt werden, sind [`~PreTrainedModel.from_pretrained`] und
+[`~PreTrainedModel.save_pretrained`], die für die Serialisierung und Deserialisierung verwendet werden. Alle
+anderen wichtigen Funktionalitäten, wie `BrandNewBertModel.forward` sollten vollständig in der neuen
+Skript `modeling_brand_new_bert.py` definiert werden. Als nächstes wollen wir sicherstellen, dass ein Modell mit einer bestimmten Kopfebene, wie z.B.
+`BrandNewBertForMaskedLM` nicht von `BrandNewBertModel` erbt, sondern `BrandNewBertModel` verwendet
+als Komponente, die im Forward Pass aufgerufen werden kann, um die Abstraktionsebene niedrig zu halten. Jedes neue Modell erfordert eine
+Konfigurationsklasse, genannt `BrandNewBertConfig`. Diese Konfiguration wird immer als ein Attribut in
+[PreTrainedModel] gespeichert und kann daher über das Attribut `config` für alle Klassen aufgerufen werden
+die von `BrandNewBertPreTrainedModel` erben:
+
+```python
+model = BrandNewBertModel.from_pretrained("brandy/brand_new_bert")
+model.config  # model has access to its config
+```
+
+Ähnlich wie das Modell erbt die Konfiguration grundlegende Serialisierungs- und Deserialisierungsfunktionalitäten von
+[`PretrainedConfig`]. Beachten Sie, dass die Konfiguration und das Modell immer in zwei verschiedene Formate serialisiert werden
+unterschiedliche Formate serialisiert werden - das Modell in eine *pytorch_model.bin* Datei und die Konfiguration in eine *config.json* Datei. Aufruf von
+[~PreTrainedModel.save_pretrained`] wird automatisch
+[~PretrainedConfig.save_pretrained`] auf, so dass sowohl das Modell als auch die Konfiguration gespeichert werden.
+
+
+### Code-Stil
+
+Wenn Sie Ihr neues Modell kodieren, sollten Sie daran denken, dass Transformers eine Bibliothek mit vielen Meinungen ist und dass wir selbst ein paar Macken haben
+wie der Code geschrieben werden sollte :-)
+
+1. Der Vorwärtsdurchlauf Ihres Modells sollte vollständig in die Modellierungsdatei geschrieben werden und dabei völlig unabhängig von anderen
+   Modellen in der Bibliothek. Wenn Sie einen Block aus einem anderen Modell wiederverwenden möchten, kopieren Sie den Code und fügen ihn mit einem
+   `# Kopiert von` ein (siehe [hier](https://github.com/huggingface/transformers/blob/v4.17.0/src/transformers/models/roberta/modeling_roberta.py#L160)
+   für ein gutes Beispiel und [hier](pr_checks#check-copies) für weitere Dokumentation zu Copied from). 
+2. Der Code sollte vollständig verständlich sein, auch für einen Nicht-Muttersprachler. Das heißt, Sie sollten
+   beschreibende Variablennamen wählen und Abkürzungen vermeiden. Ein Beispiel: `activation` ist `act` vorzuziehen.
+   Von Variablennamen mit nur einem Buchstaben wird dringend abgeraten, es sei denn, es handelt sich um einen Index in einer for-Schleife.
+3. Generell ziehen wir längeren expliziten Code einem kurzen magischen Code vor.
+4. Vermeiden Sie die Unterklassifizierung von `nn.Sequential` in PyTorch, sondern unterklassifizieren Sie `nn.Module` und schreiben Sie den Vorwärtspass, so dass jeder
+   so dass jeder, der Ihren Code verwendet, ihn schnell debuggen kann, indem er Druckanweisungen oder Haltepunkte hinzufügt.
+5. Ihre Funktionssignatur sollte mit einer Typ-Annotation versehen sein. Im Übrigen sind gute Variablennamen viel lesbarer und verständlicher
+   verständlicher als Typ-Anmerkungen.
+
+### Übersicht der Tokenizer
+
+Noch nicht ganz fertig :-( Dieser Abschnitt wird bald hinzugefügt!
+
+## Schritt-für-Schritt-Rezept zum Hinzufügen eines Modells zu 🤗 Transformers
+
+Jeder hat andere Vorlieben, was die Portierung eines Modells angeht. Daher kann es sehr hilfreich sein, wenn Sie sich Zusammenfassungen ansehen
+wie andere Mitwirkende Modelle auf Hugging Face portiert haben. Hier ist eine Liste von Blogbeiträgen aus der Community, wie man ein Modell portiert:
+
+1. [Portierung eines GPT2-Modells](https://medium.com/huggingface/from-tensorflow-to-pytorch-265f40ef2a28) von [Thomas](https://huggingface.co/thomwolf)
+2. [Portierung des WMT19 MT-Modells](https://huggingface.co/blog/porting-fsmt) von [Stas](https://huggingface.co/stas)
+
+Aus Erfahrung können wir Ihnen sagen, dass die wichtigsten Dinge, die Sie beim Hinzufügen eines Modells beachten müssen, sind:
+
+- Erfinden Sie das Rad nicht neu! Die meisten Teile des Codes, den Sie für das neue 🤗 Transformers-Modell hinzufügen werden, existieren bereits
+  irgendwo in 🤗 Transformers. Nehmen Sie sich etwas Zeit, um ähnliche, bereits vorhandene Modelle und Tokenizer zu finden, die Sie kopieren können
+  von. [grep](https://www.gnu.org/software/grep/) und [rg](https://github.com/BurntSushi/ripgrep) sind Ihre
+  Freunde. Beachten Sie, dass es sehr gut möglich ist, dass der Tokenizer Ihres Modells auf einer Modellimplementierung basiert und
+  und der Modellierungscode Ihres Modells auf einer anderen. *Z.B.* Der Modellierungscode von FSMT basiert auf BART, während der Tokenizer-Code von FSMT
+  auf XLM basiert.
+- Es handelt sich eher um eine technische als um eine wissenschaftliche Herausforderung. Sie sollten mehr Zeit auf die Schaffung einer
+  eine effiziente Debugging-Umgebung zu schaffen, als zu versuchen, alle theoretischen Aspekte des Modells in dem Papier zu verstehen.
+- Bitten Sie um Hilfe, wenn Sie nicht weiterkommen! Modelle sind der Kernbestandteil von 🤗 Transformers, so dass wir bei Hugging Face mehr als
+  mehr als glücklich, Ihnen bei jedem Schritt zu helfen, um Ihr Modell hinzuzufügen. Zögern Sie nicht zu fragen, wenn Sie merken, dass Sie nicht weiterkommen.
+  Fortschritte machen.
+
+Im Folgenden versuchen wir, Ihnen ein allgemeines Rezept an die Hand zu geben, das uns bei der Portierung eines Modells auf 🤗 Transformers am nützlichsten erschien.
+
+Die folgende Liste ist eine Zusammenfassung all dessen, was getan werden muss, um ein Modell hinzuzufügen und kann von Ihnen als To-Do verwendet werden
+Liste verwenden:
+
+☐ (Optional) Verstehen der theoretischen Aspekte des Modells<br>
+☐ Vorbereiten der 🤗 Transformers-Entwicklungsumgebung<br>
+☐ Debugging-Umgebung des ursprünglichen Repositorys eingerichtet<br>
+☐ Skript erstellt, das den Durchlauf `forward()` unter Verwendung des ursprünglichen Repositorys und des Checkpoints erfolgreich durchführt<br>
+☐ Erfolgreich das Modellskelett zu 🤗 Transformers hinzugefügt<br>
+☐ Erfolgreiche Umwandlung des ursprünglichen Prüfpunkts in den 🤗 Transformers-Prüfpunkt<br>
+☐ Erfolgreich den Durchlauf `forward()` in 🤗 Transformers ausgeführt, der eine identische Ausgabe wie der ursprüngliche Prüfpunkt liefert<br>
+☐ Modell-Tests in 🤗 Transformers abgeschlossen<br>
+☐ Erfolgreich Tokenizer in 🤗 Transformers hinzugefügt<br>
+☐ End-to-End-Integrationstests ausgeführt<br>
+☐ Docs fertiggestellt<br>
+☐ Modellgewichte in den Hub hochgeladen<br>
+☐ Die Pull-Anfrage eingereicht<br>
+☐ (Optional) Hinzufügen eines Demo-Notizbuchs
+
+Für den Anfang empfehlen wir in der Regel, mit einem guten theoretischen Verständnis von `BrandNewBert` zu beginnen. Wie auch immer,
+wenn Sie es vorziehen, die theoretischen Aspekte des Modells *on-the-job* zu verstehen, dann ist es völlig in Ordnung, direkt in die
+in die Code-Basis von `BrandNewBert` einzutauchen. Diese Option könnte für Sie besser geeignet sein, wenn Ihre technischen Fähigkeiten besser sind als
+als Ihre theoretischen Fähigkeiten, wenn Sie Schwierigkeiten haben, die Arbeit von `BrandNewBert` zu verstehen, oder wenn Sie einfach Spaß am Programmieren
+mehr Spaß am Programmieren haben als am Lesen wissenschaftlicher Abhandlungen.
+
+### 1. (Optional) Theoretische Aspekte von BrandNewBert
+
+Sie sollten sich etwas Zeit nehmen, um die Abhandlung von *BrandNewBert* zu lesen, falls eine solche Beschreibung existiert. Möglicherweise gibt es große
+Abschnitte des Papiers, die schwer zu verstehen sind. Wenn das der Fall ist, ist das in Ordnung - machen Sie sich keine Sorgen! Das Ziel ist
+ist es nicht, ein tiefes theoretisches Verständnis des Papiers zu erlangen, sondern die notwendigen Informationen zu extrahieren, um
+das Modell effektiv in 🤗 Transformers zu implementieren. Das heißt, Sie müssen nicht zu viel Zeit auf die
+theoretischen Aspekten verbringen, sondern sich lieber auf die praktischen Aspekte konzentrieren, nämlich:
+
+- Welche Art von Modell ist *brand_new_bert*? BERT-ähnliches Modell nur für den Encoder? GPT2-ähnliches reines Decoder-Modell? BART-ähnliches
+  Encoder-Decoder-Modell? Sehen Sie sich die [model_summary](model_summary) an, wenn Sie mit den Unterschieden zwischen diesen Modellen nicht vertraut sind.
+- Was sind die Anwendungen von *brand_new_bert*? Textklassifizierung? Texterzeugung? Seq2Seq-Aufgaben, *z.B.,*
+  Zusammenfassungen?
+- Was ist die neue Eigenschaft des Modells, die es von BERT/GPT-2/BART unterscheidet?
+- Welches der bereits existierenden [🤗 Transformers-Modelle](https://huggingface.co/transformers/#contents) ist am ähnlichsten
+  ähnlich wie *brand_new_bert*?
+- Welche Art von Tokenizer wird verwendet? Ein Satzteil-Tokenisierer? Ein Wortstück-Tokenisierer? Ist es derselbe Tokenisierer, der für
+  für BERT oder BART?
+
+Nachdem Sie das Gefühl haben, einen guten Überblick über die Architektur des Modells erhalten zu haben, können Sie dem
+Hugging Face Team schreiben und Ihre Fragen stellen. Dazu können Fragen zur Architektur des Modells gehören,
+seiner Aufmerksamkeitsebene usw. Wir werden Ihnen gerne weiterhelfen.
+
+### 2. Bereiten Sie als nächstes Ihre Umgebung vor
+
+1. Forken Sie das [Repository](https://github.com/huggingface/transformers), indem Sie auf der Seite des Repositorys auf die Schaltfläche 'Fork' klicken.
+   Seite des Repositorys klicken. Dadurch wird eine Kopie des Codes unter Ihrem GitHub-Benutzerkonto erstellt.
+
+2. Klonen Sie Ihren `transformers` Fork auf Ihre lokale Festplatte und fügen Sie das Basis-Repository als Remote hinzu:
+
+```bash
+git clone https://github.com/[your Github handle]/transformers.git
+cd transformers
+git remote add upstream https://github.com/huggingface/transformers.git
+```
+
+3. Richten Sie eine Entwicklungsumgebung ein, indem Sie z.B. den folgenden Befehl ausführen:
+
+```bash
+python -m venv .env
+source .env/bin/activate
+pip install -e ".[dev]"
+```
+
+Abhängig von Ihrem Betriebssystem und da die Anzahl der optionalen Abhängigkeiten von Transformers wächst, kann es sein, dass Sie bei diesem Befehl einen
+Fehler mit diesem Befehl. Stellen Sie in diesem Fall sicher, dass Sie das Deep Learning Framework, mit dem Sie arbeiten, installieren
+(PyTorch, TensorFlow und/oder Flax) und führen Sie es aus:
+
+```bash
+pip install -e ".[quality]"
+```
+
+was für die meisten Anwendungsfälle ausreichend sein sollte. Sie können dann zum übergeordneten Verzeichnis zurückkehren
+
+```bash
+cd ..
+```
+
+4. Wir empfehlen, die PyTorch-Version von *brand_new_bert* zu Transformers hinzuzufügen. Um PyTorch zu installieren, folgen Sie bitte den
+   Anweisungen auf https://pytorch.org/get-started/locally/.
+
+**Anmerkung:** Sie müssen CUDA nicht installiert haben. Es reicht aus, das neue Modell auf der CPU zum Laufen zu bringen.
+
+5. Um *brand_new_bert* zu portieren, benötigen Sie außerdem Zugriff auf das Original-Repository:
+
+```bash
+git clone https://github.com/org_that_created_brand_new_bert_org/brand_new_bert.git
+cd brand_new_bert
+pip install -e .
+```
+
+Jetzt haben Sie eine Entwicklungsumgebung eingerichtet, um *brand_new_bert* auf 🤗 Transformers zu portieren.
+
+### 3.-4. Führen Sie einen Pre-Training-Checkpoint mit dem Original-Repository durch
+
+Zunächst werden Sie mit dem ursprünglichen *brand_new_bert* Repository arbeiten. Oft ist die ursprüngliche Implementierung sehr
+"forschungslastig". Das bedeutet, dass es an Dokumentation mangeln kann und der Code schwer zu verstehen sein kann. Aber das sollte
+genau Ihre Motivation sein, *brand_new_bert* neu zu implementieren. Eines unserer Hauptziele bei Hugging Face ist es, *die Menschen dazu zu bringen
+auf den Schultern von Giganten zu stehen*, was sich hier sehr gut darin ausdrückt, dass wir ein funktionierendes Modell nehmen und es umschreiben, um es so
+es so **zugänglich, benutzerfreundlich und schön** wie möglich zu machen. Dies ist die wichtigste Motivation für die Neuimplementierung von
+Modelle in 🤗 Transformers umzuwandeln - der Versuch, komplexe neue NLP-Technologie für **jeden** zugänglich zu machen.
+
+Sie sollten damit beginnen, indem Sie in das Original-Repository eintauchen.
+
+Die erfolgreiche Ausführung des offiziellen Pre-Trainingsmodells im Original-Repository ist oft **der schwierigste** Schritt.
+Unserer Erfahrung nach ist es sehr wichtig, dass Sie einige Zeit damit verbringen, sich mit der ursprünglichen Code-Basis vertraut zu machen. Sie müssen
+das Folgende herausfinden:
+
+- Wo finden Sie die vortrainierten Gewichte?
+- Wie lädt man die vorab trainierten Gewichte in das entsprechende Modell?
+- Wie kann der Tokenizer unabhängig vom Modell ausgeführt werden?
+- Verfolgen Sie einen Forward Pass, damit Sie wissen, welche Klassen und Funktionen für einen einfachen Forward Pass erforderlich sind. Normalerweise,
+  müssen Sie nur diese Funktionen reimplementieren.
+- Sie müssen in der Lage sein, die wichtigen Komponenten des Modells zu finden: Wo befindet sich die Klasse des Modells? Gibt es Unterklassen des Modells,
+  *z.B.* EncoderModel, DecoderModel? Wo befindet sich die Selbstaufmerksamkeitsschicht? Gibt es mehrere verschiedene Aufmerksamkeitsebenen,
+  *z.B.* *Selbstaufmerksamkeit*, *Kreuzaufmerksamkeit*...?
+- Wie können Sie das Modell in der ursprünglichen Umgebung des Repo debuggen? Müssen Sie *print* Anweisungen hinzufügen, können Sie
+  mit einem interaktiven Debugger wie *ipdb* arbeiten oder sollten Sie eine effiziente IDE zum Debuggen des Modells verwenden, wie z.B. PyCharm?
+
+Es ist sehr wichtig, dass Sie, bevor Sie mit der Portierung beginnen, den Code im Original-Repository **effizient** debuggen können
+Repository können! Denken Sie auch daran, dass Sie mit einer Open-Source-Bibliothek arbeiten, also zögern Sie nicht, ein Problem oder
+oder sogar eine Pull-Anfrage im Original-Repository zu stellen. Die Betreuer dieses Repositorys sind wahrscheinlich sehr froh darüber
+dass jemand in ihren Code schaut!
+
+An diesem Punkt liegt es wirklich an Ihnen, welche Debugging-Umgebung und Strategie Sie zum Debuggen des ursprünglichen
+Modell zu debuggen. Wir raten dringend davon ab, eine kostspielige GPU-Umgebung einzurichten, sondern arbeiten Sie einfach auf einer CPU, sowohl wenn Sie mit dem
+in das ursprüngliche Repository einzutauchen und auch, wenn Sie beginnen, die 🤗 Transformers-Implementierung des Modells zu schreiben. Nur
+ganz am Ende, wenn das Modell bereits erfolgreich auf 🤗 Transformers portiert wurde, sollte man überprüfen, ob das
+Modell auch auf der GPU wie erwartet funktioniert.
+
+Im Allgemeinen gibt es zwei mögliche Debugging-Umgebungen für die Ausführung des Originalmodells
+
+- [Jupyter notebooks](https://jupyter.org/) / [google colab](https://colab.research.google.com/notebooks/intro.ipynb)
+- Lokale Python-Skripte.
+
+Jupyter-Notebooks haben den Vorteil, dass sie eine zellenweise Ausführung ermöglichen, was hilfreich sein kann, um logische Komponenten besser voneinander zu trennen und
+logische Komponenten voneinander zu trennen und schnellere Debugging-Zyklen zu haben, da Zwischenergebnisse gespeichert werden können. Außerdem,
+Außerdem lassen sich Notebooks oft leichter mit anderen Mitwirkenden teilen, was sehr hilfreich sein kann, wenn Sie das Hugging Face Team um Hilfe bitten möchten.
+Face Team um Hilfe bitten. Wenn Sie mit Jupyter-Notizbüchern vertraut sind, empfehlen wir Ihnen dringend, mit ihnen zu arbeiten.
+
+Der offensichtliche Nachteil von Jupyter-Notizbüchern ist, dass Sie, wenn Sie nicht daran gewöhnt sind, mit ihnen zu arbeiten, einige Zeit damit verbringen müssen
+einige Zeit damit verbringen müssen, sich an die neue Programmierumgebung zu gewöhnen, und dass Sie möglicherweise Ihre bekannten Debugging-Tools nicht mehr verwenden können
+wie z.B. `ipdb` nicht mehr verwenden können.
+
+Für jede Codebasis ist es immer ein guter erster Schritt, einen **kleinen** vortrainierten Checkpoint zu laden und in der Lage zu sein, einen
+einzelnen Vorwärtsdurchlauf mit einem Dummy-Integer-Vektor von Eingabe-IDs als Eingabe zu reproduzieren. Ein solches Skript könnte wie folgt aussehen (in
+Pseudocode):
+
+```python
+model = BrandNewBertModel.load_pretrained_checkpoint("/path/to/checkpoint/")
+input_ids = [0, 4, 5, 2, 3, 7, 9]  # vector of input ids
+original_output = model.predict(input_ids)
+```
+
+Was die Debugging-Strategie anbelangt, so können Sie im Allgemeinen aus mehreren Strategien wählen:
+
+- Zerlegen Sie das ursprüngliche Modell in viele kleine testbare Komponenten und führen Sie für jede dieser Komponenten einen Vorwärtsdurchlauf zur
+  Überprüfung
+- Zerlegen Sie das ursprüngliche Modell nur in den ursprünglichen *Tokenizer* und das ursprüngliche *Modell*, führen Sie einen Vorwärtsdurchlauf für diese Komponenten durch
+  und verwenden Sie dazwischenliegende Druckanweisungen oder Haltepunkte zur Überprüfung.
+
+Auch hier bleibt es Ihnen überlassen, welche Strategie Sie wählen. Oft ist die eine oder die andere Strategie vorteilhaft, je nach der ursprünglichen Codebasis
+Basis.
+
+Wenn die ursprüngliche Codebasis es Ihnen erlaubt, das Modell in kleinere Teilkomponenten zu zerlegen, *z.B.* wenn die ursprüngliche
+Code-Basis problemlos im Eager-Modus ausgeführt werden kann, lohnt es sich in der Regel, dies zu tun. Es gibt einige wichtige Vorteile
+am Anfang den schwierigeren Weg zu gehen:
+
+- Wenn Sie später das ursprüngliche Modell mit der Hugging Face-Implementierung vergleichen, können Sie automatisch überprüfen, ob
+  für jede Komponente einzeln überprüfen, ob die entsprechende Komponente der 🤗 Transformers-Implementierung übereinstimmt, anstatt sich auf
+  anstatt sich auf den visuellen Vergleich über Druckanweisungen zu verlassen
+- können Sie das große Problem der Portierung eines Modells in kleinere Probleme der Portierung einzelner Komponenten zerlegen
+  einzelnen Komponenten zu zerlegen und so Ihre Arbeit besser zu strukturieren
+- Die Aufteilung des Modells in logisch sinnvolle Komponenten hilft Ihnen, einen besseren Überblick über das Design des Modells zu bekommen
+  und somit das Modell besser zu verstehen
+- In einem späteren Stadium helfen Ihnen diese komponentenweisen Tests dabei, sicherzustellen, dass keine Regressionen auftreten, während Sie fortfahren
+  Ihren Code ändern
+
+[Lysandre's](https://gist.github.com/LysandreJik/db4c948f6b4483960de5cbac598ad4ed) Integrationstests für ELECTRA
+gibt ein schönes Beispiel dafür, wie dies geschehen kann.
+
+Wenn die ursprüngliche Codebasis jedoch sehr komplex ist oder nur die Ausführung von Zwischenkomponenten in einem kompilierten Modus erlaubt,
+könnte es zu zeitaufwändig oder sogar unmöglich sein, das Modell in kleinere testbare Teilkomponenten zu zerlegen. Ein gutes
+Beispiel ist die [T5's MeshTensorFlow](https://github.com/tensorflow/mesh/tree/master/mesh_tensorflow) Bibliothek, die sehr komplex ist
+sehr komplex ist und keine einfache Möglichkeit bietet, das Modell in seine Unterkomponenten zu zerlegen. Bei solchen Bibliotheken ist man
+oft auf die Überprüfung von Druckanweisungen angewiesen.
+
+Unabhängig davon, welche Strategie Sie wählen, ist die empfohlene Vorgehensweise oft die gleiche, nämlich dass Sie mit der Fehlersuche in den
+die Anfangsebenen zuerst und die Endebenen zuletzt debuggen.
+
+Es wird empfohlen, dass Sie die Ausgaben der folgenden Ebenen abrufen, entweder durch Druckanweisungen oder Unterkomponentenfunktionen
+Schichten in der folgenden Reihenfolge abrufen:
+
+1. Rufen Sie die Eingabe-IDs ab, die an das Modell übergeben wurden
+2. Rufen Sie die Worteinbettungen ab
+3. Rufen Sie die Eingabe der ersten Transformer-Schicht ab
+4. Rufen Sie die Ausgabe der ersten Transformer-Schicht ab
+5. Rufen Sie die Ausgabe der folgenden n - 1 Transformer-Schichten ab
+6. Rufen Sie die Ausgabe des gesamten BrandNewBert Modells ab
+
+Die Eingabe-IDs sollten dabei aus einem Array von Ganzzahlen bestehen, *z.B.* `input_ids = [0, 4, 4, 3, 2, 4, 1, 7, 19]`
+
+Die Ausgaben der folgenden Schichten bestehen oft aus mehrdimensionalen Float-Arrays und können wie folgt aussehen:
+
+```
+[[
+ [-0.1465, -0.6501,  0.1993,  ...,  0.1451,  0.3430,  0.6024],
+ [-0.4417, -0.5920,  0.3450,  ..., -0.3062,  0.6182,  0.7132],
+ [-0.5009, -0.7122,  0.4548,  ..., -0.3662,  0.6091,  0.7648],
+ ...,
+ [-0.5613, -0.6332,  0.4324,  ..., -0.3792,  0.7372,  0.9288],
+ [-0.5416, -0.6345,  0.4180,  ..., -0.3564,  0.6992,  0.9191],
+ [-0.5334, -0.6403,  0.4271,  ..., -0.3339,  0.6533,  0.8694]]],
+```
+
+Wir erwarten, dass jedes zu 🤗 Transformers hinzugefügte Modell eine Reihe von Integrationstests besteht, was bedeutet, dass das ursprüngliche
+Modell und die neu implementierte Version in 🤗 Transformers exakt dieselbe Ausgabe liefern müssen, und zwar mit einer Genauigkeit von 0,001!
+Da es normal ist, dass das exakt gleiche Modell, das in verschiedenen Bibliotheken geschrieben wurde, je nach Bibliotheksrahmen eine leicht unterschiedliche Ausgabe liefern kann
+eine leicht unterschiedliche Ausgabe liefern kann, akzeptieren wir eine Fehlertoleranz von 1e-3 (0,001). Es reicht nicht aus, wenn das Modell
+fast das gleiche Ergebnis liefert, sie müssen fast identisch sein. Daher werden Sie sicherlich die Zwischenergebnisse
+Zwischenergebnisse der 🤗 Transformers-Version mehrfach mit den Zwischenergebnissen der ursprünglichen Implementierung von
+*brand_new_bert* vergleichen. In diesem Fall ist eine **effiziente** Debugging-Umgebung des ursprünglichen Repositorys absolut
+wichtig ist. Hier sind einige Ratschläge, um Ihre Debugging-Umgebung so effizient wie möglich zu gestalten.
+
+- Finden Sie den besten Weg, um Zwischenergebnisse zu debuggen. Ist das ursprüngliche Repository in PyTorch geschrieben? Dann sollten Sie
+  dann sollten Sie sich wahrscheinlich die Zeit nehmen, ein längeres Skript zu schreiben, das das ursprüngliche Modell in kleinere Unterkomponenten zerlegt, um
+  Zwischenwerte abzurufen. Ist das ursprüngliche Repository in Tensorflow 1 geschrieben? Dann müssen Sie sich möglicherweise auf die
+  TensorFlow Druckoperationen wie [tf.print](https://www.tensorflow.org/api_docs/python/tf/print) verlassen, um die
+  Zwischenwerte auszugeben. Ist das ursprüngliche Repository in Jax geschrieben? Dann stellen Sie sicher, dass das Modell **nicht jitted** ist, wenn
+  wenn Sie den Vorwärtsdurchlauf ausführen, *z.B.* schauen Sie sich [dieser Link](https://github.com/google/jax/issues/196) an.
+- Verwenden Sie den kleinsten vortrainierten Prüfpunkt, den Sie finden können. Je kleiner der Prüfpunkt ist, desto schneller wird Ihr Debugging-Zyklus
+  wird. Es ist nicht effizient, wenn Ihr vorab trainiertes Modell so groß ist, dass Ihr Vorwärtsdurchlauf mehr als 10 Sekunden dauert.
+  Falls nur sehr große Checkpoints verfügbar sind, kann es sinnvoller sein, ein Dummy-Modell in der neuen
+  Umgebung mit zufällig initialisierten Gewichten zu erstellen und diese Gewichte zum Vergleich mit der 🤗 Transformers-Version
+  Ihres Modells
+- Vergewissern Sie sich, dass Sie den einfachsten Weg wählen, um einen Forward Pass im ursprünglichen Repository aufzurufen. Idealerweise sollten Sie
+  die Funktion im originalen Repository finden, die **nur** einen einzigen Vorwärtspass aufruft, *d.h.* die oft aufgerufen wird
+  Vorhersagen", "Auswerten", "Vorwärts" oder "Aufruf" genannt wird. Sie wollen keine Funktion debuggen, die `forward` aufruft
+  mehrfach aufruft, *z.B.* um Text zu erzeugen, wie `autoregressive_sample`, `generate`.
+- Versuchen Sie, die Tokenisierung vom *Forward*-Pass des Modells zu trennen. Wenn das Original-Repository Beispiele zeigt, bei denen
+  Sie eine Zeichenkette eingeben müssen, dann versuchen Sie herauszufinden, an welcher Stelle im Vorwärtsaufruf die Zeichenketteneingabe in Eingabe-IDs geändert wird
+  geändert wird und beginnen Sie an dieser Stelle. Das könnte bedeuten, dass Sie möglicherweise selbst ein kleines Skript schreiben oder den
+  Originalcode so ändern müssen, dass Sie die ids direkt eingeben können, anstatt eine Zeichenkette einzugeben.
+- Vergewissern Sie sich, dass sich das Modell in Ihrem Debugging-Setup **nicht** im Trainingsmodus befindet, der oft dazu führt, dass das Modell
+  Dies führt häufig zu zufälligen Ergebnissen, da das Modell mehrere Dropout-Schichten enthält. Stellen Sie sicher, dass der Vorwärtsdurchlauf in Ihrer Debugging
+  Umgebung **deterministisch** ist, damit die Dropout-Schichten nicht verwendet werden. Oder verwenden Sie *transformers.utils.set_seed*.
+  wenn sich die alte und die neue Implementierung im selben Framework befinden.
+
+Im folgenden Abschnitt finden Sie genauere Details/Tipps, wie Sie dies für *brand_new_bert* tun können.
+
+### 5.-14. Portierung von BrandNewBert auf 🤗 Transformatoren
+
+Als nächstes können Sie endlich damit beginnen, neuen Code zu 🤗 Transformers hinzuzufügen. Gehen Sie in den Klon Ihres 🤗 Transformers Forks:
+
+```bash
+cd transformers
+```
+
+In dem speziellen Fall, dass Sie ein Modell hinzufügen, dessen Architektur genau mit der Modellarchitektur eines
+Modells übereinstimmt, müssen Sie nur ein Konvertierungsskript hinzufügen, wie in [diesem Abschnitt](#write-a-conversion-script) beschrieben.
+In diesem Fall können Sie einfach die gesamte Modellarchitektur des bereits vorhandenen Modells wiederverwenden.
+
+Andernfalls beginnen wir mit der Erstellung eines neuen Modells. Sie haben hier zwei Möglichkeiten:
+
+- `transformers-cli add-new-model-like`, um ein neues Modell wie ein bestehendes hinzuzufügen
+- `transformers-cli add-new-model`, um ein neues Modell aus unserer Vorlage hinzuzufügen (sieht dann aus wie BERT oder Bart, je nachdem, welche Art von Modell Sie wählen)
+
+In beiden Fällen werden Sie mit einem Fragebogen aufgefordert, die grundlegenden Informationen zu Ihrem Modell auszufüllen. Für den zweiten Befehl müssen Sie `cookiecutter` installieren, weitere Informationen dazu finden Sie [hier](https://github.com/huggingface/transformers/tree/main/templates/adding_a_new_model).
+
+**Eröffnen Sie einen Pull Request auf dem Haupt-Repositorium huggingface/transformers**
+
+Bevor Sie mit der Anpassung des automatisch generierten Codes beginnen, ist es nun an der Zeit, einen "Work in progress (WIP)" Pull
+Anfrage, *z.B.* "[WIP] Add *brand_new_bert*", in 🤗 Transformers zu öffnen, damit Sie und das Hugging Face Team
+Seite an Seite an der Integration des Modells in 🤗 Transformers arbeiten können.
+
+Sie sollten Folgendes tun:
+
+1. Erstellen Sie eine Verzweigung mit einem beschreibenden Namen von Ihrer Hauptverzweigung
+
+```bash
+git checkout -b add_brand_new_bert
+```
+
+2. Bestätigen Sie den automatisch generierten Code:
+
+```bash
+git add .
+git commit
+```
+
+3. Abrufen und zurücksetzen auf die aktuelle Haupt
+
+```bash
+git fetch upstream
+git rebase upstream/main
+```
+
+4. Übertragen Sie die Änderungen auf Ihr Konto mit:
+
+```bash
+git push -u origin a-descriptive-name-for-my-changes
+```
+
+5. Wenn Sie zufrieden sind, gehen Sie auf die Webseite Ihrer Abspaltung auf GitHub. Klicken Sie auf "Pull request". Stellen Sie sicher, dass Sie das
+   GitHub-Handle einiger Mitglieder des Hugging Face-Teams als Reviewer hinzuzufügen, damit das Hugging Face-Team über zukünftige Änderungen informiert wird.
+   zukünftige Änderungen benachrichtigt wird.
+
+6. Ändern Sie den PR in einen Entwurf, indem Sie auf der rechten Seite der GitHub-Pull-Request-Webseite auf "In Entwurf umwandeln" klicken.
+
+Vergessen Sie im Folgenden nicht, wenn Sie Fortschritte gemacht haben, Ihre Arbeit zu committen und in Ihr Konto zu pushen, damit sie in der Pull-Anfrage erscheint.
+damit sie in der Pull-Anfrage angezeigt wird. Außerdem sollten Sie darauf achten, dass Sie Ihre Arbeit von Zeit zu Zeit mit dem aktuellen main
+von Zeit zu Zeit zu aktualisieren, indem Sie dies tun:
+
+```bash
+git fetch upstream
+git merge upstream/main
+```
+
+Generell sollten Sie alle Fragen, die Sie in Bezug auf das Modell oder Ihre Implementierung haben, in Ihrem PR stellen und
+in der PR diskutiert/gelöst werden. Auf diese Weise wird das Hugging Face Team immer benachrichtigt, wenn Sie neuen Code einreichen oder
+wenn Sie eine Frage haben. Es ist oft sehr hilfreich, das Hugging Face-Team auf Ihren hinzugefügten Code hinzuweisen, damit das Hugging Face-Team Ihr Problem oder Ihre Frage besser verstehen kann.
+Face-Team Ihr Problem oder Ihre Frage besser verstehen kann.
+
+Gehen Sie dazu auf die Registerkarte "Geänderte Dateien", auf der Sie alle Ihre Änderungen sehen, gehen Sie zu einer Zeile, zu der Sie eine Frage stellen möchten
+eine Frage stellen möchten, und klicken Sie auf das "+"-Symbol, um einen Kommentar hinzuzufügen. Wenn eine Frage oder ein Problem gelöst wurde,
+können Sie auf die Schaltfläche "Lösen" des erstellten Kommentars klicken.
+
+Auf dieselbe Weise wird das Hugging Face-Team Kommentare öffnen, wenn es Ihren Code überprüft. Wir empfehlen, die meisten Fragen
+auf GitHub in Ihrem PR zu stellen. Für einige sehr allgemeine Fragen, die für die Öffentlichkeit nicht sehr nützlich sind, können Sie das
+Hugging Face Team per Slack oder E-Mail zu stellen.
+
+**5. Passen Sie den Code der generierten Modelle für brand_new_bert** an.
+
+Zunächst werden wir uns nur auf das Modell selbst konzentrieren und uns nicht um den Tokenizer kümmern. Den gesamten relevanten Code sollten Sie
+finden Sie in den generierten Dateien `src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` und
+`src/transformers/models/brand_new_bert/configuration_brand_new_bert.py`.
+
+Jetzt können Sie endlich mit dem Programmieren beginnen :). Der generierte Code in
+`src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` wird entweder die gleiche Architektur wie BERT haben, wenn
+wenn es sich um ein reines Encoder-Modell handelt oder BART, wenn es sich um ein Encoder-Decoder-Modell handelt. An diesem Punkt sollten Sie sich daran erinnern, was
+was Sie am Anfang über die theoretischen Aspekte des Modells gelernt haben: *Wie unterscheidet sich das Modell von BERT oder
+BART?*". Implementieren Sie diese Änderungen, was oft bedeutet, dass Sie die *Selbstaufmerksamkeitsschicht*, die Reihenfolge der Normalisierungsschicht usw. ändern müssen.
+Schicht usw... Auch hier ist es oft nützlich, sich die ähnliche Architektur bereits bestehender Modelle in Transformers anzusehen, um ein besseres Gefühl dafür zu bekommen
+ein besseres Gefühl dafür zu bekommen, wie Ihr Modell implementiert werden sollte.
+
+**Beachten Sie**, dass Sie an diesem Punkt nicht sehr sicher sein müssen, dass Ihr Code völlig korrekt oder sauber ist. Vielmehr ist es
+Sie sollten vielmehr eine erste *unbereinigte*, kopierte Version des ursprünglichen Codes in
+src/transformers/models/brand_new_bert/modeling_brand_new_bert.py" hinzuzufügen, bis Sie das Gefühl haben, dass der gesamte notwendige Code
+hinzugefügt wurde. Unserer Erfahrung nach ist es viel effizienter, schnell eine erste Version des erforderlichen Codes hinzuzufügen und
+den Code iterativ mit dem Konvertierungsskript zu verbessern/korrigieren, wie im nächsten Abschnitt beschrieben. Das einzige, was
+zu diesem Zeitpunkt funktionieren muss, ist, dass Sie die 🤗 Transformers-Implementierung von *brand_new_bert* instanziieren können, *d.h.* der
+folgende Befehl sollte funktionieren:
+
+```python
+from transformers import BrandNewBertModel, BrandNewBertConfig
+
+model = BrandNewBertModel(BrandNewBertConfig())
+```
+
+Der obige Befehl erstellt ein Modell gemäß den Standardparametern, die in `BrandNewBertConfig()` definiert sind, mit
+zufälligen Gewichten und stellt damit sicher, dass die `init()` Methoden aller Komponenten funktionieren.
+
+Beachten Sie, dass alle zufälligen Initialisierungen in der Methode `_init_weights` Ihres `BrandnewBertPreTrainedModel` stattfinden sollten.
+Klasse erfolgen sollte. Sie sollte alle Blattmodule in Abhängigkeit von den Variablen der Konfiguration initialisieren. Hier ist ein Beispiel mit der
+BERT `_init_weights` Methode:
+
+```py
+def _init_weights(self, module):
+    """Initialize the weights"""
+    if isinstance(module, nn.Linear):
+        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        if module.bias is not None:
+            module.bias.data.zero_()
+    elif isinstance(module, nn.Embedding):
+        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        if module.padding_idx is not None:
+            module.weight.data[module.padding_idx].zero_()
+    elif isinstance(module, nn.LayerNorm):
+        module.bias.data.zero_()
+        module.weight.data.fill_(1.0)
+```
+
+Sie können weitere benutzerdefinierte Schemata verwenden, wenn Sie eine spezielle Initialisierung für einige Module benötigen. Zum Beispiel in
+`Wav2Vec2ForPreTraining` müssen die letzten beiden linearen Schichten die Initialisierung des regulären PyTorch `nn.Linear` haben.
+aber alle anderen sollten eine Initialisierung wie oben verwenden. Dies ist wie folgt kodiert:
+
+```py
+def _init_weights(self, module):
+    """Initialize the weights"""
+    if isinstnace(module, Wav2Vec2ForPreTraining):
+        module.project_hid.reset_parameters()
+        module.project_q.reset_parameters()
+        module.project_hid._is_hf_initialized = True
+        module.project_q._is_hf_initialized = True
+    elif isinstance(module, nn.Linear):
+        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        if module.bias is not None:
+            module.bias.data.zero_()
+```
+
+Das Flag `_is_hf_initialized` wird intern verwendet, um sicherzustellen, dass wir ein Submodul nur einmal initialisieren. Wenn Sie es auf
+True` für `module.project_q` und `module.project_hid` setzen, stellen wir sicher, dass die benutzerdefinierte Initialisierung, die wir vorgenommen haben, später nicht überschrieben wird,
+die Funktion `_init_weights` nicht auf sie angewendet wird.
+
+**6. Schreiben Sie ein Konvertierungsskript**
+
+Als nächstes sollten Sie ein Konvertierungsskript schreiben, mit dem Sie den Checkpoint, den Sie zum Debuggen von *brand_new_bert* im
+im ursprünglichen Repository in einen Prüfpunkt konvertieren, der mit Ihrer gerade erstellten 🤗 Transformers-Implementierung von
+*brand_new_bert*. Es ist nicht ratsam, das Konvertierungsskript von Grund auf neu zu schreiben, sondern die bereits
+bestehenden Konvertierungsskripten in 🤗 Transformers nach einem Skript zu suchen, das für die Konvertierung eines ähnlichen Modells verwendet wurde, das im
+demselben Framework wie *brand_new_bert* geschrieben wurde. Normalerweise reicht es aus, ein bereits vorhandenes Konvertierungsskript zu kopieren und
+es für Ihren Anwendungsfall leicht anzupassen. Zögern Sie nicht, das Hugging Face Team zu bitten, Sie auf ein ähnliches, bereits vorhandenes
+Konvertierungsskript für Ihr Modell zu finden.
+
+- Wenn Sie ein Modell von TensorFlow nach PyTorch portieren, ist ein guter Ausgangspunkt das Konvertierungsskript von BERT [hier] (https://github.com/huggingface/transformers/blob/7acfa95afb8194f8f9c1f4d2c6028224dbed35a2/src/transformers/models/bert/modeling_bert.py#L91)
+- Wenn Sie ein Modell von PyTorch nach PyTorch portieren, ist ein guter Ausgangspunkt das Konvertierungsskript von BART [hier](https://github.com/huggingface/transformers/blob/main/src/transformers/models/bart/convert_bart_original_pytorch_checkpoint_to_pytorch.py)
+
+Im Folgenden werden wir kurz erklären, wie PyTorch-Modelle Ebenengewichte speichern und Ebenennamen definieren. In PyTorch wird der
+Name einer Ebene durch den Namen des Klassenattributs definiert, das Sie der Ebene geben. Lassen Sie uns ein Dummy-Modell in
+PyTorch, das wir `SimpleModel` nennen, wie folgt:
+
+```python
+from torch import nn
+
+
+class SimpleModel(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.dense = nn.Linear(10, 10)
+        self.intermediate = nn.Linear(10, 10)
+        self.layer_norm = nn.LayerNorm(10)
+```
+
+Jetzt können wir eine Instanz dieser Modelldefinition erstellen, die alle Gewichte ausfüllt: `dense`, `intermediate`,
+`layer_norm` mit zufälligen Gewichten. Wir können das Modell ausdrucken, um seine Architektur zu sehen
+
+```python
+model = SimpleModel()
+
+print(model)
+```
+
+Dies gibt folgendes aus:
+
+```
+SimpleModel(
+  (dense): Linear(in_features=10, out_features=10, bias=True)
+  (intermediate): Linear(in_features=10, out_features=10, bias=True)
+  (layer_norm): LayerNorm((10,), eps=1e-05, elementwise_affine=True)
+)
+```
+
+Wir können sehen, dass die Ebenennamen durch den Namen des Klassenattributs in PyTorch definiert sind. Sie können die Gewichtswerte
+Werte einer bestimmten Ebene anzeigen lassen:
+
+```python
+print(model.dense.weight.data)
+```
+
+um zu sehen, dass die Gewichte zufällig initialisiert wurden
+
+```
+tensor([[-0.0818,  0.2207, -0.0749, -0.0030,  0.0045, -0.1569, -0.1598,  0.0212,
+         -0.2077,  0.2157],
+        [ 0.1044,  0.0201,  0.0990,  0.2482,  0.3116,  0.2509,  0.2866, -0.2190,
+          0.2166, -0.0212],
+        [-0.2000,  0.1107, -0.1999, -0.3119,  0.1559,  0.0993,  0.1776, -0.1950,
+         -0.1023, -0.0447],
+        [-0.0888, -0.1092,  0.2281,  0.0336,  0.1817, -0.0115,  0.2096,  0.1415,
+         -0.1876, -0.2467],
+        [ 0.2208, -0.2352, -0.1426, -0.2636, -0.2889, -0.2061, -0.2849, -0.0465,
+          0.2577,  0.0402],
+        [ 0.1502,  0.2465,  0.2566,  0.0693,  0.2352, -0.0530,  0.1859, -0.0604,
+          0.2132,  0.1680],
+        [ 0.1733, -0.2407, -0.1721,  0.1484,  0.0358, -0.0633, -0.0721, -0.0090,
+          0.2707, -0.2509],
+        [-0.1173,  0.1561,  0.2945,  0.0595, -0.1996,  0.2988, -0.0802,  0.0407,
+          0.1829, -0.1568],
+        [-0.1164, -0.2228, -0.0403,  0.0428,  0.1339,  0.0047,  0.1967,  0.2923,
+          0.0333, -0.0536],
+        [-0.1492, -0.1616,  0.1057,  0.1950, -0.2807, -0.2710, -0.1586,  0.0739,
+          0.2220,  0.2358]]).
+```
+
+Im Konvertierungsskript sollten Sie diese zufällig initialisierten Gewichte mit den genauen Gewichten der
+entsprechenden Ebene im Kontrollpunkt. *Z.B.*
+
+```python
+# retrieve matching layer weights, e.g. by
+# recursive algorithm
+layer_name = "dense"
+pretrained_weight = array_of_dense_layer
+
+model_pointer = getattr(model, "dense")
+
+model_pointer.weight.data = torch.from_numpy(pretrained_weight)
+```
+
+Dabei müssen Sie sicherstellen, dass jedes zufällig initialisierte Gewicht Ihres PyTorch-Modells und sein entsprechendes
+Checkpoint-Gewicht in **Form und Name** genau übereinstimmen. Zu diesem Zweck ist es **notwendig**, assert
+Anweisungen für die Form hinzuzufügen und die Namen der Checkpoint-Gewichte auszugeben. Sie sollten z.B. Anweisungen hinzufügen wie:
+
+```python
+assert (
+    model_pointer.weight.shape == pretrained_weight.shape
+), f"Pointer shape of random weight {model_pointer.shape} and array shape of checkpoint weight {pretrained_weight.shape} mismatched"
+```
+
+Außerdem sollten Sie die Namen der beiden Gewichte ausdrucken, um sicherzustellen, dass sie übereinstimmen, *z.B.*.
+
+```python
+logger.info(f"Initialize PyTorch weight {layer_name} from {pretrained_weight.name}")
+```
+
+Wenn entweder die Form oder der Name nicht übereinstimmt, haben Sie wahrscheinlich das falsche Kontrollpunktgewicht einer zufällig
+Ebene der 🤗 Transformers-Implementierung zugewiesen.
+
+Eine falsche Form ist höchstwahrscheinlich auf eine falsche Einstellung der Konfigurationsparameter in `BrandNewBertConfig()` zurückzuführen, die
+nicht genau mit denen übereinstimmen, die für den zu konvertierenden Prüfpunkt verwendet wurden. Es könnte aber auch sein, dass
+die PyTorch-Implementierung eines Layers erfordert, dass das Gewicht vorher transponiert wird.
+
+Schließlich sollten Sie auch überprüfen, ob **alle** erforderlichen Gewichte initialisiert sind und alle Checkpoint-Gewichte ausgeben, die
+die nicht zur Initialisierung verwendet wurden, um sicherzustellen, dass das Modell korrekt konvertiert wurde. Es ist völlig normal, dass die
+Konvertierungsversuche entweder mit einer falschen Shape-Anweisung oder einer falschen Namenszuweisung fehlschlagen. Das liegt höchstwahrscheinlich daran, dass entweder
+Sie haben falsche Parameter in `BrandNewBertConfig()` verwendet, haben eine falsche Architektur in der 🤗 Transformers
+Implementierung, Sie haben einen Fehler in den `init()` Funktionen einer der Komponenten der 🤗 Transformers
+Implementierung oder Sie müssen eine der Kontrollpunktgewichte transponieren.
+
+Dieser Schritt sollte mit dem vorherigen Schritt wiederholt werden, bis alle Gewichte des Kontrollpunkts korrekt in das
+Transformers-Modell geladen sind. Nachdem Sie den Prüfpunkt korrekt in die 🤗 Transformers-Implementierung geladen haben, können Sie das Modell
+das Modell unter einem Ordner Ihrer Wahl `/path/to/converted/checkpoint/folder` speichern, der dann sowohl ein
+Datei `pytorch_model.bin` und eine Datei `config.json` enthalten sollte:
+
+```python
+model.save_pretrained("/path/to/converted/checkpoint/folder")
+```
+
+**7. Implementieren Sie den Vorwärtspass**
+
+Nachdem es Ihnen gelungen ist, die trainierten Gewichte korrekt in die 🤗 Transformers-Implementierung zu laden, sollten Sie nun dafür sorgen
+sicherstellen, dass der Forward Pass korrekt implementiert ist. In [Machen Sie sich mit dem ursprünglichen Repository vertraut](#34-run-a-pretrained-checkpoint-using-the-original-repository) haben Sie bereits ein Skript erstellt, das einen Forward Pass
+Durchlauf des Modells unter Verwendung des Original-Repositorys durchführt. Jetzt sollten Sie ein analoges Skript schreiben, das die 🤗 Transformers
+Implementierung anstelle der Originalimplementierung verwenden. Es sollte wie folgt aussehen:
+
+```python
+model = BrandNewBertModel.from_pretrained("/path/to/converted/checkpoint/folder")
+input_ids = [0, 4, 4, 3, 2, 4, 1, 7, 19]
+output = model(input_ids).last_hidden_states
+```
+
+Es ist sehr wahrscheinlich, dass die 🤗 Transformers-Implementierung und die ursprüngliche Modell-Implementierung nicht genau die gleiche Ausgabe liefern.
+beim ersten Mal nicht die gleiche Ausgabe liefern oder dass der Vorwärtsdurchlauf einen Fehler auslöst. Seien Sie nicht enttäuscht - das ist zu erwarten! Erstens,
+sollten Sie sicherstellen, dass der Vorwärtsdurchlauf keine Fehler auslöst. Es passiert oft, dass die falschen Dimensionen verwendet werden
+verwendet werden, was zu einem *Dimensionality mismatch* Fehler führt oder dass der falsche Datentyp verwendet wird, *z.B.* `torch.long`
+anstelle von `torch.float32`. Zögern Sie nicht, das Hugging Face Team um Hilfe zu bitten, wenn Sie bestimmte Fehler nicht lösen können.
+bestimmte Fehler nicht lösen können.
+
+Um sicherzustellen, dass die Implementierung von 🤗 Transformers korrekt funktioniert, müssen Sie sicherstellen, dass die Ausgaben
+einer Genauigkeit von `1e-3` entsprechen. Zunächst sollten Sie sicherstellen, dass die Ausgabeformen identisch sind, *d.h.*.
+Die Ausgabeform *outputs.shape* sollte für das Skript der 🤗 Transformers-Implementierung und die ursprüngliche
+Implementierung ergeben. Als nächstes sollten Sie sicherstellen, dass auch die Ausgabewerte identisch sind. Dies ist einer der schwierigsten
+Teile des Hinzufügens eines neuen Modells. Häufige Fehler, warum die Ausgaben nicht identisch sind, sind:
+
+- Einige Ebenen wurden nicht hinzugefügt, *d.h.* eine *Aktivierungsebene* wurde nicht hinzugefügt, oder die Restverbindung wurde vergessen
+- Die Worteinbettungsmatrix wurde nicht gebunden
+- Es werden die falschen Positionseinbettungen verwendet, da die ursprüngliche Implementierung einen Offset verwendet
+- Dropout wird während des Vorwärtsdurchlaufs angewendet. Um dies zu beheben, stellen Sie sicher, dass *model.training auf False* steht und dass keine Dropout
+  Schicht während des Vorwärtsdurchlaufs fälschlicherweise aktiviert wird, *d.h.* übergeben Sie *self.training* an [PyTorch's functional dropout](https://pytorch.org/docs/stable/nn.functional.html?highlight=dropout#torch.nn.functional.dropout)
+
+Der beste Weg, das Problem zu beheben, besteht normalerweise darin, sich den Vorwärtsdurchlauf der ursprünglichen Implementierung und die 🤗
+Transformers-Implementierung nebeneinander zu sehen und zu prüfen, ob es Unterschiede gibt. Idealerweise sollten Sie die
+Zwischenergebnisse beider Implementierungen des Vorwärtsdurchlaufs debuggen/ausdrucken, um die genaue Position im Netzwerk zu finden, an der die 🤗
+Transformers-Implementierung eine andere Ausgabe zeigt als die ursprüngliche Implementierung. Stellen Sie zunächst sicher, dass die
+hartcodierten `input_ids` in beiden Skripten identisch sind. Überprüfen Sie dann, ob die Ausgaben der ersten Transformation von
+der `input_ids` (normalerweise die Worteinbettungen) identisch sind. Und dann arbeiten Sie sich bis zur allerletzten Schicht des
+Netzwerks. Irgendwann werden Sie einen Unterschied zwischen den beiden Implementierungen feststellen, der Sie auf den Fehler
+in der Implementierung von 🤗 Transformers hinweist. Unserer Erfahrung nach ist ein einfacher und effizienter Weg, viele Druckanweisungen hinzuzufügen
+sowohl in der Original-Implementierung als auch in der 🤗 Transformers-Implementierung an den gleichen Stellen im Netzwerk
+hinzuzufügen und nacheinander Druckanweisungen zu entfernen, die dieselben Werte für Zwischenpräsentationen anzeigen.
+
+Wenn Sie sicher sind, dass beide Implementierungen die gleiche Ausgabe liefern, überprüfen Sie die Ausgaben mit
+`torch.allclose(original_output, output, atol=1e-3)` überprüfen, haben Sie den schwierigsten Teil hinter sich! Herzlichen Glückwunsch - die
+Arbeit, die noch zu erledigen ist, sollte ein Kinderspiel sein 😊.
+
+**8. Hinzufügen aller notwendigen Modelltests**
+
+An diesem Punkt haben Sie erfolgreich ein neues Modell hinzugefügt. Es ist jedoch sehr gut möglich, dass das Modell noch nicht
+noch nicht vollständig mit dem erforderlichen Design übereinstimmt. Um sicherzustellen, dass die Implementierung vollständig kompatibel mit 🤗 Transformers ist, sollten alle
+gemeinsamen Tests bestehen. Der Cookiecutter sollte automatisch eine Testdatei für Ihr Modell hinzugefügt haben, wahrscheinlich unter
+demselben `tests/models/brand_new_bert/test_modeling_brand_new_bert.py`. Führen Sie diese Testdatei aus, um zu überprüfen, ob alle gängigen
+Tests bestehen:
+
+```bash
+pytest tests/models/brand_new_bert/test_modeling_brand_new_bert.py
+```
+
+Nachdem Sie alle allgemeinen Tests festgelegt haben, müssen Sie nun sicherstellen, dass all die schöne Arbeit, die Sie geleistet haben, gut getestet ist, damit
+
+- a) die Community Ihre Arbeit leicht nachvollziehen kann, indem sie sich spezifische Tests von *brand_new_bert* ansieht
+- b) zukünftige Änderungen an Ihrem Modell keine wichtigen Funktionen des Modells zerstören.
+
+Als erstes sollten Sie Integrationstests hinzufügen. Diese Integrationstests tun im Wesentlichen dasselbe wie die Debugging-Skripte
+die Sie zuvor zur Implementierung des Modells in 🤗 Transformers verwendet haben. Eine Vorlage für diese Modelltests wurde bereits von dem
+Cookiecutter hinzugefügt, die `BrandNewBertModelIntegrationTests` heißt und nur noch von Ihnen ausgefüllt werden muss. Um sicherzustellen, dass diese
+Tests erfolgreich sind, führen Sie
+
+```bash
+RUN_SLOW=1 pytest -sv tests/models/brand_new_bert/test_modeling_brand_new_bert.py::BrandNewBertModelIntegrationTests
+```
+
+<Tip>
+
+Falls Sie Windows verwenden, sollten Sie `RUN_SLOW=1` durch `SET RUN_SLOW=1` ersetzen.
+
+</Tip>
+
+Zweitens sollten alle Funktionen, die speziell für *brand_new_bert* sind, zusätzlich in einem separaten Test getestet werden unter
+`BrandNewBertModelTester`/``BrandNewBertModelTest`. Dieser Teil wird oft vergessen, ist aber in zweierlei Hinsicht äußerst nützlich
+Weise:
+
+- Er hilft dabei, das Wissen, das Sie während der Modellerweiterung erworben haben, an die Community weiterzugeben, indem er zeigt, wie die
+  speziellen Funktionen von *brand_new_bert* funktionieren sollten.
+- Künftige Mitwirkende können Änderungen am Modell schnell testen, indem sie diese speziellen Tests ausführen.
+
+
+**9. Implementieren Sie den Tokenizer**
+
+Als nächstes sollten wir den Tokenizer von *brand_new_bert* hinzufügen. Normalerweise ist der Tokenizer äquivalent oder sehr ähnlich zu einem
+bereits vorhandenen Tokenizer von 🤗 Transformers.
+
+Es ist sehr wichtig, die ursprüngliche Tokenizer-Datei zu finden/extrahieren und es zu schaffen, diese Datei in die 🤗
+Transformers Implementierung des Tokenizers zu laden.
+
+Um sicherzustellen, dass der Tokenizer korrekt funktioniert, empfiehlt es sich, zunächst ein Skript im ursprünglichen Repository zu erstellen
+zu erstellen, das eine Zeichenkette eingibt und die `input_ids` zurückgibt. Es könnte etwa so aussehen (in Pseudocode):
+
+```python
+input_str = "This is a long example input string containing special characters .$?-, numbers 2872 234 12 and words."
+model = BrandNewBertModel.load_pretrained_checkpoint("/path/to/checkpoint/")
+input_ids = model.tokenize(input_str)
+```
+
+Möglicherweise müssen Sie noch einmal einen Blick in das ursprüngliche Repository werfen, um die richtige Tokenizer-Funktion zu finden, oder Sie müssen
+Sie müssen vielleicht sogar Änderungen an Ihrem Klon des Original-Repositorys vornehmen, um nur die `input_ids` auszugeben. Nach dem Schreiben
+ein funktionierendes Tokenisierungsskript geschrieben, das das ursprüngliche Repository verwendet, sollten Sie ein analoges Skript für 🤗 Transformers
+erstellt werden. Es sollte ähnlich wie dieses aussehen:
+
+```python
+from transformers import BrandNewBertTokenizer
+
+input_str = "This is a long example input string containing special characters .$?-, numbers 2872 234 12 and words."
+
+tokenizer = BrandNewBertTokenizer.from_pretrained("/path/to/tokenizer/folder/")
+
+input_ids = tokenizer(input_str).input_ids
+```
+
+Wenn beide `input_ids` die gleichen Werte ergeben, sollte als letzter Schritt auch eine Tokenizer-Testdatei hinzugefügt werden.
+
+Analog zu den Modellierungstestdateien von *brand_new_bert* sollten auch die Tokenisierungs-Testdateien von *brand_new_bert*
+eine Reihe von fest kodierten Integrationstests enthalten.
+
+**10. Führen Sie End-to-End-Integrationstests aus**
+
+Nachdem Sie den Tokenizer hinzugefügt haben, sollten Sie auch ein paar End-to-End-Integrationstests, die sowohl das Modell als auch den
+Tokenizer zu `tests/models/brand_new_bert/test_modeling_brand_new_bert.py` in 🤗 Transformers.
+Ein solcher Test sollte bei einem aussagekräftigen
+Text-zu-Text-Beispiel zeigen, dass die Implementierung von 🤗 Transformers wie erwartet funktioniert. Ein aussagekräftiges Text-zu-Text-Beispiel kann
+z.B. *ein Quell-zu-Ziel-Übersetzungspaar, ein Artikel-zu-Zusammenfassung-Paar, ein Frage-zu-Antwort-Paar, usw... Wenn keiner der
+der portierten Prüfpunkte in einer nachgelagerten Aufgabe feinabgestimmt wurde, genügt es, sich einfach auf die Modelltests zu verlassen. In einem
+letzten Schritt, um sicherzustellen, dass das Modell voll funktionsfähig ist, sollten Sie alle Tests auch auf der GPU durchführen. Es kann
+Es kann vorkommen, dass Sie vergessen haben, einige `.to(self.device)` Anweisungen zu internen Tensoren des Modells hinzuzufügen, was in einem solchen
+Test zu einem Fehler führen würde. Falls Sie keinen Zugang zu einem Grafikprozessor haben, kann das Hugging Face Team diese Tests für Sie durchführen.
+Tests für Sie übernehmen.
+
+**11. Docstring hinzufügen**
+
+Nun sind alle notwendigen Funktionen für *brand_new_bert* hinzugefügt - Sie sind fast fertig! Das Einzige, was Sie noch hinzufügen müssen, ist
+ein schöner Docstring und eine Doku-Seite. Der Cookiecutter sollte eine Vorlagendatei namens
+`docs/source/model_doc/brand_new_bert.md` hinzugefügt haben, die Sie ausfüllen sollten. Die Benutzer Ihres Modells werden in der Regel zuerst einen Blick auf
+diese Seite ansehen, bevor sie Ihr Modell verwenden. Daher muss die Dokumentation verständlich und prägnant sein. Es ist sehr nützlich für
+die Gemeinschaft, einige *Tipps* hinzuzufügen, um zu zeigen, wie das Modell verwendet werden sollte. Zögern Sie nicht, das Hugging Face-Team anzupingen
+bezüglich der Docstrings.
+
+Stellen Sie als nächstes sicher, dass der zu `src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` hinzugefügte docstring
+korrekt ist und alle erforderlichen Eingaben und Ausgaben enthält. Wir haben eine ausführliche Anleitung zum Schreiben von Dokumentationen und unserem Docstring-Format [hier](writing-documentation). Es ist immer gut, sich daran zu erinnern, dass die Dokumentation
+mindestens so sorgfältig behandelt werden sollte wie der Code in 🤗 Transformers, denn die Dokumentation ist in der Regel der erste Kontaktpunkt der
+Berührungspunkt der Community mit dem Modell ist.
+
+**Code refactor**
+
+Großartig, jetzt haben Sie den gesamten erforderlichen Code für *brand_new_bert* hinzugefügt. An diesem Punkt sollten Sie einige mögliche
+falschen Codestil korrigieren, indem Sie ausführen:
+
+```bash
+make style
+```
+
+und überprüfen Sie, ob Ihr Kodierungsstil die Qualitätsprüfung besteht:
+
+```bash
+make quality
+```
+
+Es gibt noch ein paar andere sehr strenge Designtests in 🤗 Transformers, die möglicherweise noch fehlschlagen, was sich in den
+den Tests Ihres Pull Requests. Dies liegt oft an fehlenden Informationen im Docstring oder an einer falschen
+Benennung. Das Hugging Face Team wird Ihnen sicherlich helfen, wenn Sie hier nicht weiterkommen.
+
+Und schließlich ist es immer eine gute Idee, den eigenen Code zu refaktorisieren, nachdem man sichergestellt hat, dass er korrekt funktioniert. Wenn alle
+Tests bestanden haben, ist es nun an der Zeit, den hinzugefügten Code noch einmal durchzugehen und einige Überarbeitungen vorzunehmen.
+
+Sie haben nun den Codierungsteil abgeschlossen, herzlichen Glückwunsch! 🎉 Sie sind großartig! 😎
+
+**12. Laden Sie die Modelle in den Model Hub hoch**
+
+In diesem letzten Teil sollten Sie alle Checkpoints konvertieren und in den Modell-Hub hochladen und eine Modellkarte für jeden
+hochgeladenen Modell-Kontrollpunkt. Sie können sich mit den Hub-Funktionen vertraut machen, indem Sie unsere [Model sharing and uploading Page](model_sharing) lesen. Hier sollten Sie mit dem Hugging Face-Team zusammenarbeiten, um einen passenden Namen für jeden
+Checkpoint festzulegen und die erforderlichen Zugriffsrechte zu erhalten, um das Modell unter der Organisation des Autors *brand_new_bert* hochladen zu können.
+*brand_new_bert*. Die Methode `push_to_hub`, die in allen Modellen in `transformers` vorhanden ist, ist ein schneller und effizienter Weg, Ihren Checkpoint in den Hub zu pushen. Ein kleines Snippet ist unten eingefügt:
+
+```python
+brand_new_bert.push_to_hub("brand_new_bert")
+# Uncomment the following line to push to an organization.
+# brand_new_bert.push_to_hub("<organization>/brand_new_bert")
+```
+
+Es lohnt sich, etwas Zeit darauf zu verwenden, für jeden Kontrollpunkt passende Musterkarten zu erstellen. Die Modellkarten sollten die
+spezifischen Merkmale dieses bestimmten Prüfpunkts hervorheben, * z.B.* auf welchem Datensatz wurde der Prüfpunkt
+vortrainiert/abgestimmt? Für welche nachgelagerte Aufgabe sollte das Modell verwendet werden? Und fügen Sie auch etwas Code bei, wie Sie
+wie das Modell korrekt verwendet wird.
+
+**13. (Optional) Notizbuch hinzufügen**
+
+Es ist sehr hilfreich, ein Notizbuch hinzuzufügen, in dem im Detail gezeigt wird, wie *brand_new_bert* für Schlussfolgerungen verwendet werden kann und/oder
+bei einer nachgelagerten Aufgabe feinabgestimmt wird. Dies ist nicht zwingend erforderlich, um Ihren PR zusammenzuführen, aber sehr nützlich für die Gemeinschaft.
+
+**14. Reichen Sie Ihren fertigen PR ein**
+
+Sie sind jetzt mit der Programmierung fertig und können zum letzten Schritt übergehen, nämlich der Zusammenführung Ihres PR mit main. Normalerweise hat das
+Hugging Face Team Ihnen an diesem Punkt bereits geholfen haben, aber es lohnt sich, sich etwas Zeit zu nehmen, um Ihrem fertigen
+PR eine schöne Beschreibung zu geben und eventuell Kommentare zu Ihrem Code hinzuzufügen, wenn Sie Ihren Gutachter auf bestimmte Designentscheidungen hinweisen wollen.
+Gutachter hinweisen wollen.
+
+### Teilen Sie Ihre Arbeit!!
+
+Jetzt ist es an der Zeit, von der Community Anerkennung für Ihre Arbeit zu bekommen! Die Fertigstellung einer Modellergänzung ist ein wichtiger
+Beitrag zu Transformers und der gesamten NLP-Gemeinschaft. Ihr Code und die portierten vortrainierten Modelle werden sicherlich
+von Hunderten und vielleicht sogar Tausenden von Entwicklern und Forschern genutzt werden. Sie sollten stolz auf Ihre Arbeit sein und Ihre
+Ihre Leistung mit der Gemeinschaft teilen.
+
+**Sie haben ein weiteres Modell erstellt, das für jeden in der Community super einfach zugänglich ist! 🤯**

docs/source/de/add_new_pipeline.md

@@ -0,0 +1,258 @@
+<!--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
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Wie erstellt man eine benutzerdefinierte Pipeline?
+
+In dieser Anleitung sehen wir uns an, wie Sie eine benutzerdefinierte Pipeline erstellen und sie auf dem [Hub](hf.co/models) freigeben oder sie der
+🤗 Transformers-Bibliothek hinzufügen.
+
+Zuallererst müssen Sie entscheiden, welche Roheingaben die Pipeline verarbeiten kann. Es kann sich um Strings, rohe Bytes,
+Dictionaries oder was auch immer die wahrscheinlichste gewünschte Eingabe ist. Versuchen Sie, diese Eingaben so rein wie möglich in Python zu halten
+denn das macht die Kompatibilität einfacher (auch mit anderen Sprachen über JSON). Dies werden die Eingaben der
+Pipeline (`Vorverarbeitung`).
+
+Definieren Sie dann die `Outputs`. Dieselbe Richtlinie wie für die Eingänge. Je einfacher, desto besser. Dies werden die Ausgaben der
+Methode `Postprocess`.
+
+Beginnen Sie damit, die Basisklasse `Pipeline` mit den 4 Methoden zu erben, die für die Implementierung von `preprocess` benötigt werden,
+Weiterleiten", "Nachbearbeitung" und "Parameter säubern".
+
+
+```python
+from transformers import Pipeline
+
+
+class MyPipeline(Pipeline):
+    def _sanitize_parameters(self, **kwargs):
+        preprocess_kwargs = {}
+        if "maybe_arg" in kwargs:
+            preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]
+        return preprocess_kwargs, {}, {}
+
+    def preprocess(self, inputs, maybe_arg=2):
+        model_input = Tensor(inputs["input_ids"])
+        return {"model_input": model_input}
+
+    def _forward(self, model_inputs):
+        # model_inputs == {"model_input": model_input}
+        outputs = self.model(**model_inputs)
+        # Maybe {"logits": Tensor(...)}
+        return outputs
+
+    def postprocess(self, model_outputs):
+        best_class = model_outputs["logits"].softmax(-1)
+        return best_class
+```
+
+Die Struktur dieser Aufteilung soll eine relativ nahtlose Unterstützung für CPU/GPU ermöglichen und gleichzeitig die Durchführung von
+Vor-/Nachbearbeitung auf der CPU in verschiedenen Threads
+
+Preprocess" nimmt die ursprünglich definierten Eingaben und wandelt sie in etwas um, das in das Modell eingespeist werden kann. Es kann
+mehr Informationen enthalten und ist normalerweise ein `Dict`.
+
+`_forward` ist das Implementierungsdetail und ist nicht dafür gedacht, direkt aufgerufen zu werden. Weiterleiten" ist die bevorzugte
+aufgerufene Methode, da sie Sicherheitsvorkehrungen enthält, die sicherstellen, dass alles auf dem erwarteten Gerät funktioniert. Wenn etwas
+mit einem realen Modell verknüpft ist, gehört es in die Methode `_forward`, alles andere gehört in die Methoden preprocess/postprocess.
+
+Die Methode `Postprocess` nimmt die Ausgabe von `_forward` und verwandelt sie in die endgültige Ausgabe, die zuvor festgelegt wurde.
+zuvor entschieden wurde.
+
+Die Methode `_sanitize_parameters` ermöglicht es dem Benutzer, beliebige Parameter zu übergeben, wann immer er möchte, sei es bei der Initialisierung
+Zeit `pipeline(...., maybe_arg=4)` oder zur Aufrufzeit `pipe = pipeline(...); output = pipe(...., maybe_arg=4)`.
+
+Die Rückgabe von `_sanitize_parameters` sind die 3 Dicts von kwargs, die direkt an `preprocess` übergeben werden,
+`_forward` und `postprocess` übergeben werden. Füllen Sie nichts aus, wenn der Aufrufer keinen zusätzlichen Parameter angegeben hat. Das
+erlaubt es, die Standardargumente in der Funktionsdefinition beizubehalten, was immer "natürlicher" ist.
+
+Ein klassisches Beispiel wäre das Argument `top_k` in der Nachbearbeitung bei Klassifizierungsaufgaben.
+
+```python
+>>> pipe = pipeline("my-new-task")
+>>> pipe("This is a test")
+[{"label": "1-star", "score": 0.8}, {"label": "2-star", "score": 0.1}, {"label": "3-star", "score": 0.05}
+{"label": "4-star", "score": 0.025}, {"label": "5-star", "score": 0.025}]
+
+>>> pipe("This is a test", top_k=2)
+[{"label": "1-star", "score": 0.8}, {"label": "2-star", "score": 0.1}]
+```
+
+In order to achieve that, we'll update our `postprocess` method with a default parameter to `5`. and edit
+`_sanitize_parameters` to allow this new parameter.
+
+
+```python
+def postprocess(self, model_outputs, top_k=5):
+    best_class = model_outputs["logits"].softmax(-1)
+    # Add logic to handle top_k
+    return best_class
+
+
+def _sanitize_parameters(self, **kwargs):
+    preprocess_kwargs = {}
+    if "maybe_arg" in kwargs:
+        preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]
+
+    postprocess_kwargs = {}
+    if "top_k" in kwargs:
+        postprocess_kwargs["top_k"] = kwargs["top_k"]
+    return preprocess_kwargs, {}, postprocess_kwargs
+```
+
+Versuchen Sie, die Eingaben/Ausgaben sehr einfach und idealerweise JSON-serialisierbar zu halten, da dies die Verwendung der Pipeline sehr einfach macht
+ohne dass die Benutzer neue Arten von Objekten verstehen müssen. Es ist auch relativ üblich, viele verschiedene Arten von Argumenten zu unterstützen
+von Argumenten zu unterstützen (Audiodateien, die Dateinamen, URLs oder reine Bytes sein können).
+
+
+
+## Hinzufügen zur Liste der unterstützten Aufgaben
+
+Um Ihre `neue Aufgabe` in die Liste der unterstützten Aufgaben aufzunehmen, müssen Sie sie zur `PIPELINE_REGISTRY` hinzufügen:
+
+```python
+from transformers.pipelines import PIPELINE_REGISTRY
+
+PIPELINE_REGISTRY.register_pipeline(
+    "new-task",
+    pipeline_class=MyPipeline,
+    pt_model=AutoModelForSequenceClassification,
+)
+```
+
+Wenn Sie möchten, können Sie ein Standardmodell angeben. In diesem Fall sollte es mit einer bestimmten Revision (die der Name einer Verzweigung oder ein Commit-Hash sein kann, hier haben wir `"abcdef"` genommen) sowie mit dem Typ versehen sein:
+
+```python
+PIPELINE_REGISTRY.register_pipeline(
+    "new-task",
+    pipeline_class=MyPipeline,
+    pt_model=AutoModelForSequenceClassification,
+    default={"pt": ("user/awesome_model", "abcdef")},
+    type="text",  # current support type: text, audio, image, multimodal
+)
+```
+
+## Teilen Sie Ihre Pipeline auf dem Hub
+
+Um Ihre benutzerdefinierte Pipeline auf dem Hub freizugeben, müssen Sie lediglich den benutzerdefinierten Code Ihrer `Pipeline`-Unterklasse in einer
+Python-Datei speichern. Nehmen wir zum Beispiel an, Sie möchten eine benutzerdefinierte Pipeline für die Klassifizierung von Satzpaaren wie folgt verwenden:
+
+```py
+import numpy as np
+
+from transformers import Pipeline
+
+
+def softmax(outputs):
+    maxes = np.max(outputs, axis=-1, keepdims=True)
+    shifted_exp = np.exp(outputs - maxes)
+    return shifted_exp / shifted_exp.sum(axis=-1, keepdims=True)
+
+
+class PairClassificationPipeline(Pipeline):
+    def _sanitize_parameters(self, **kwargs):
+        preprocess_kwargs = {}
+        if "second_text" in kwargs:
+            preprocess_kwargs["second_text"] = kwargs["second_text"]
+        return preprocess_kwargs, {}, {}
+
+    def preprocess(self, text, second_text=None):
+        return self.tokenizer(text, text_pair=second_text, return_tensors=self.framework)
+
+    def _forward(self, model_inputs):
+        return self.model(**model_inputs)
+
+    def postprocess(self, model_outputs):
+        logits = model_outputs.logits[0].numpy()
+        probabilities = softmax(logits)
+
+        best_class = np.argmax(probabilities)
+        label = self.model.config.id2label[best_class]
+        score = probabilities[best_class].item()
+        logits = logits.tolist()
+        return {"label": label, "score": score, "logits": logits}
+```
+
+Die Implementierung ist Framework-unabhängig und funktioniert für PyTorch- und TensorFlow-Modelle. Wenn wir dies in einer Datei
+einer Datei namens `pair_classification.py` gespeichert haben, können wir sie importieren und wie folgt registrieren:
+
+```py
+from pair_classification import PairClassificationPipeline
+from transformers.pipelines import PIPELINE_REGISTRY
+from transformers import AutoModelForSequenceClassification, TFAutoModelForSequenceClassification
+
+PIPELINE_REGISTRY.register_pipeline(
+    "pair-classification",
+    pipeline_class=PairClassificationPipeline,
+    pt_model=AutoModelForSequenceClassification,
+    tf_model=TFAutoModelForSequenceClassification,
+)
+```
+
+Sobald dies geschehen ist, können wir es mit einem vortrainierten Modell verwenden. Zum Beispiel wurde `sgugger/finetuned-bert-mrpc` auf den
+auf den MRPC-Datensatz abgestimmt, der Satzpaare als Paraphrasen oder nicht klassifiziert.
+
+```py
+from transformers import pipeline
+
+classifier = pipeline("pair-classification", model="sgugger/finetuned-bert-mrpc")
+```
+
+Dann können wir sie auf dem Hub mit der Methode `save_pretrained` in einem `Repository` freigeben:
+
+```py
+from huggingface_hub import Repository
+
+repo = Repository("test-dynamic-pipeline", clone_from="{your_username}/test-dynamic-pipeline")
+classifier.save_pretrained("test-dynamic-pipeline")
+repo.push_to_hub()
+```
+
+Dadurch wird die Datei, in der Sie `PairClassificationPipeline` definiert haben, in den Ordner `"test-dynamic-pipeline"` kopiert,
+und speichert das Modell und den Tokenizer der Pipeline, bevor Sie alles in das Repository verschieben
+`{Ihr_Benutzername}/test-dynamic-pipeline`. Danach kann jeder die Pipeline verwenden, solange er die Option
+`trust_remote_code=True` angeben:
+
+```py
+from transformers import pipeline
+
+classifier = pipeline(model="{your_username}/test-dynamic-pipeline", trust_remote_code=True)
+```
+
+## Hinzufügen der Pipeline zu 🤗 Transformers
+
+Wenn Sie Ihre Pipeline zu 🤗 Transformers beitragen möchten, müssen Sie ein neues Modul im Untermodul `pipelines` hinzufügen
+mit dem Code Ihrer Pipeline hinzufügen. Fügen Sie es dann der Liste der in `pipelines/__init__.py` definierten Aufgaben hinzu.
+
+Dann müssen Sie noch Tests hinzufügen. Erstellen Sie eine neue Datei `tests/test_pipelines_MY_PIPELINE.py` mit Beispielen für die anderen Tests.
+
+Die Funktion `run_pipeline_test` ist sehr allgemein gehalten und läuft auf kleinen Zufallsmodellen auf jeder möglichen
+Architektur, wie durch `model_mapping` und `tf_model_mapping` definiert.
+
+Dies ist sehr wichtig, um die zukünftige Kompatibilität zu testen, d.h. wenn jemand ein neues Modell für
+`XXXForQuestionAnswering` hinzufügt, wird der Pipeline-Test versuchen, mit diesem Modell zu arbeiten. Da die Modelle zufällig sind, ist es
+ist es unmöglich, die tatsächlichen Werte zu überprüfen. Deshalb gibt es eine Hilfsfunktion `ANY`, die einfach versucht, die
+Ausgabe der Pipeline TYPE.
+
+Außerdem *müssen* Sie 2 (idealerweise 4) Tests implementieren.
+
+- test_small_model_pt` : Definieren Sie 1 kleines Modell für diese Pipeline (es spielt keine Rolle, ob die Ergebnisse keinen Sinn ergeben)
+  und testen Sie die Ausgaben der Pipeline. Die Ergebnisse sollten die gleichen sein wie bei `test_small_model_tf`.
+- test_small_model_tf : Definieren Sie 1 kleines Modell für diese Pipeline (es spielt keine Rolle, ob die Ergebnisse keinen Sinn ergeben)
+  und testen Sie die Ausgaben der Pipeline. Die Ergebnisse sollten die gleichen sein wie bei `test_small_model_pt`.
+- test_large_model_pt` (`optional`): Testet die Pipeline an einer echten Pipeline, bei der die Ergebnisse
+  Sinn machen. Diese Tests sind langsam und sollten als solche gekennzeichnet werden. Hier geht es darum, die Pipeline zu präsentieren und sicherzustellen
+  sicherzustellen, dass es in zukünftigen Versionen keine Abweichungen gibt.
+- test_large_model_tf` (`optional`): Testet die Pipeline an einer echten Pipeline, bei der die Ergebnisse
+  Sinn machen. Diese Tests sind langsam und sollten als solche gekennzeichnet werden. Hier geht es darum, die Pipeline zu präsentieren und sicherzustellen
+  sicherzustellen, dass es in zukünftigen Versionen keine Abweichungen gibt.

docs/source/de/add_tensorflow_model.md

@@ -0,0 +1,356 @@
+<!--Copyright 2022 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Wie konvertiert man ein 🤗 Transformers-Modell in TensorFlow?
+
+Die Tatsache, dass mehrere Frameworks für die Verwendung mit 🤗 Transformers zur Verfügung stehen, gibt Ihnen die Flexibilität, deren Stärken beim Entwurf Ihrer Anwendung auszuspielen.
+Ihre Anwendung zu entwerfen, aber das bedeutet auch, dass die Kompatibilität für jedes Modell einzeln hinzugefügt werden muss. Die gute Nachricht ist, dass
+das Hinzufügen von TensorFlow-Kompatibilität zu einem bestehenden Modell einfacher ist als [das Hinzufügen eines neuen Modells von Grund auf](add_new_model)!
+Ob Sie ein tieferes Verständnis für große TensorFlow-Modelle haben möchten, einen wichtigen Open-Source-Beitrag leisten oder
+TensorFlow für das Modell Ihrer Wahl aktivieren wollen, dieser Leitfaden ist für Sie.
+
+Dieser Leitfaden befähigt Sie, ein Mitglied unserer Gemeinschaft, TensorFlow-Modellgewichte und/oder
+Architekturen beizusteuern, die in 🤗 Transformers verwendet werden sollen, und zwar mit minimaler Betreuung durch das Hugging Face Team. Das Schreiben eines neuen Modells
+ist keine Kleinigkeit, aber ich hoffe, dass dieser Leitfaden dazu beiträgt, dass es weniger eine Achterbahnfahrt 🎢 und mehr ein Spaziergang im Park 🚶 ist.
+Die Nutzung unserer kollektiven Erfahrungen ist absolut entscheidend, um diesen Prozess immer einfacher zu machen, und deshalb möchten wir
+ermutigen Sie daher, Verbesserungsvorschläge für diesen Leitfaden zu machen!
+
+Bevor Sie tiefer eintauchen, empfehlen wir Ihnen, die folgenden Ressourcen zu lesen, wenn Sie neu in 🤗 Transformers sind:
+- [Allgemeiner Überblick über 🤗 Transformers](add_new_model#general-overview-of-transformers)
+- [Die TensorFlow-Philosophie von Hugging Face](https://huggingface.co/blog/tensorflow-philosophy)
+
+Im Rest dieses Leitfadens werden Sie lernen, was nötig ist, um eine neue TensorFlow Modellarchitektur hinzuzufügen, die
+Verfahren zur Konvertierung von PyTorch in TensorFlow-Modellgewichte und wie Sie Unstimmigkeiten zwischen ML
+Frameworks. Legen Sie los!
+
+<Tip>
+
+Sind Sie unsicher, ob das Modell, das Sie verwenden möchten, bereits eine entsprechende TensorFlow-Architektur hat?
+
+&nbsp;
+
+Überprüfen Sie das Feld `model_type` in der `config.json` des Modells Ihrer Wahl
+([Beispiel](https://huggingface.co/bert-base-uncased/blob/main/config.json#L14)). Wenn der entsprechende Modellordner in
+🤗 Transformers eine Datei hat, deren Name mit "modeling_tf" beginnt, bedeutet dies, dass es eine entsprechende TensorFlow
+Architektur hat ([Beispiel](https://github.com/huggingface/transformers/tree/main/src/transformers/models/bert)).
+
+</Tip>
+
+
+## Schritt-für-Schritt-Anleitung zum Hinzufügen von TensorFlow-Modellarchitektur-Code
+
+Es gibt viele Möglichkeiten, eine große Modellarchitektur zu entwerfen, und viele Möglichkeiten, diesen Entwurf zu implementieren. Wie auch immer,
+Sie erinnern sich vielleicht an unseren [allgemeinen Überblick über 🤗 Transformers](add_new_model#general-overview-of-transformers)
+wissen, dass wir ein meinungsfreudiger Haufen sind - die Benutzerfreundlichkeit von 🤗 Transformers hängt von konsistenten Designentscheidungen ab. Aus
+Erfahrung können wir Ihnen ein paar wichtige Dinge über das Hinzufügen von TensorFlow-Modellen sagen:
+
+- Erfinden Sie das Rad nicht neu! In den meisten Fällen gibt es mindestens zwei Referenzimplementierungen, die Sie überprüfen sollten: das
+PyTorch-Äquivalent des Modells, das Sie implementieren, und andere TensorFlow-Modelle für dieselbe Klasse von Problemen.
+- Gute Modellimplementierungen überleben den Test der Zeit. Dies geschieht nicht, weil der Code hübsch ist, sondern eher
+sondern weil der Code klar, einfach zu debuggen und darauf aufzubauen ist. Wenn Sie den Maintainern das Leben mit Ihrer
+TensorFlow-Implementierung leicht machen, indem Sie die gleichen Muster wie in anderen TensorFlow-Modellen nachbilden und die Abweichung
+zur PyTorch-Implementierung minimieren, stellen Sie sicher, dass Ihr Beitrag lange Bestand haben wird.
+- Bitten Sie um Hilfe, wenn Sie nicht weiterkommen! Das 🤗 Transformers-Team ist da, um zu helfen, und wir haben wahrscheinlich Lösungen für die gleichen
+Probleme gefunden, vor denen Sie stehen.
+
+Hier finden Sie einen Überblick über die Schritte, die zum Hinzufügen einer TensorFlow-Modellarchitektur erforderlich sind:
+1. Wählen Sie das Modell, das Sie konvertieren möchten
+2. Bereiten Sie die Transformers-Entwicklungsumgebung vor.
+3. (Optional) Verstehen Sie die theoretischen Aspekte und die bestehende Implementierung
+4. Implementieren Sie die Modellarchitektur
+5. Implementieren Sie Modelltests
+6. Reichen Sie den Pull-Antrag ein
+7. (Optional) Erstellen Sie Demos und teilen Sie diese mit der Welt
+
+### 1.-3. Bereiten Sie Ihren Modellbeitrag vor
+
+**1. Wählen Sie das Modell, das Sie konvertieren möchten**
+
+Beginnen wir mit den Grundlagen: Als erstes müssen Sie die Architektur kennen, die Sie konvertieren möchten. Wenn Sie
+Sie sich nicht auf eine bestimmte Architektur festgelegt haben, ist es eine gute Möglichkeit, das 🤗 Transformers-Team um Vorschläge zu bitten.
+Wir werden Sie zu den wichtigsten Architekturen führen, die auf der TensorFlow-Seite noch fehlen.
+Seite fehlen. Wenn das spezifische Modell, das Sie mit TensorFlow verwenden möchten, bereits eine Implementierung der TensorFlow-Architektur in
+🤗 Transformers, aber es fehlen Gewichte, können Sie direkt in den
+Abschnitt [Gewichtskonvertierung](#adding-tensorflow-weights-to-hub)
+auf dieser Seite.
+
+Der Einfachheit halber wird im Rest dieser Anleitung davon ausgegangen, dass Sie sich entschieden haben, mit der TensorFlow-Version von
+*BrandNewBert* (dasselbe Beispiel wie in der [Anleitung](add_new_model), um ein neues Modell von Grund auf hinzuzufügen).
+
+<Tip>
+
+Bevor Sie mit der Arbeit an einer TensorFlow-Modellarchitektur beginnen, sollten Sie sich vergewissern, dass es keine laufenden Bemühungen in dieser Richtung gibt.
+Sie können nach `BrandNewBert` auf der
+[pull request GitHub page](https://github.com/huggingface/transformers/pulls?q=is%3Apr), um zu bestätigen, dass es keine
+TensorFlow-bezogene Pull-Anfrage gibt.
+
+</Tip>
+
+
+**2. Transformers-Entwicklungsumgebung vorbereiten**
+
+Nachdem Sie die Modellarchitektur ausgewählt haben, öffnen Sie einen PR-Entwurf, um Ihre Absicht zu signalisieren, daran zu arbeiten. Folgen Sie den
+Anweisungen, um Ihre Umgebung einzurichten und einen PR-Entwurf zu öffnen.
+
+1. Forken Sie das [repository](https://github.com/huggingface/transformers), indem Sie auf der Seite des Repositorys auf die Schaltfläche 'Fork' klicken.
+   Seite des Repositorys klicken. Dadurch wird eine Kopie des Codes unter Ihrem GitHub-Benutzerkonto erstellt.
+
+2. Klonen Sie Ihren `transformers` Fork auf Ihre lokale Festplatte und fügen Sie das Basis-Repository als Remote hinzu:
+
+```bash
+git clone https://github.com/[your Github handle]/transformers.git
+cd transformers
+git remote add upstream https://github.com/huggingface/transformers.git
+```
+
+3. Richten Sie eine Entwicklungsumgebung ein, indem Sie z.B. den folgenden Befehl ausführen:
+
+```bash
+python -m venv .env
+source .env/bin/activate
+pip install -e ".[dev]"
+```
+
+Abhängig von Ihrem Betriebssystem und da die Anzahl der optionalen Abhängigkeiten von Transformers wächst, kann es sein, dass Sie bei diesem Befehl einen
+Fehler mit diesem Befehl erhalten. Wenn das der Fall ist, stellen Sie sicher, dass Sie TensorFlow installieren und dann ausführen:
+
+```bash
+pip install -e ".[quality]"
+```
+
+**Hinweis:** Sie müssen CUDA nicht installiert haben. Es reicht aus, das neue Modell auf der CPU laufen zu lassen.
+
+4. Erstellen Sie eine Verzweigung mit einem beschreibenden Namen von Ihrer Hauptverzweigung
+
+```bash
+git checkout -b add_tf_brand_new_bert
+```
+
+5. Abrufen und zurücksetzen auf die aktuelle Hauptversion
+
+```bash
+git fetch upstream
+git rebase upstream/main
+```
+
+6. Fügen Sie eine leere `.py` Datei in `transformers/src/models/brandnewbert/` mit dem Namen `modeling_tf_brandnewbert.py` hinzu. Dies wird
+Ihre TensorFlow-Modelldatei sein.
+
+7. Übertragen Sie die Änderungen auf Ihr Konto mit:
+
+```bash
+git add .
+git commit -m "initial commit"
+git push -u origin add_tf_brand_new_bert
+```
+
+8. Wenn Sie zufrieden sind, gehen Sie auf die Webseite Ihrer Abspaltung auf GitHub. Klicken Sie auf "Pull request". Stellen Sie sicher, dass Sie das
+   GitHub-Handle einiger Mitglieder des Hugging Face-Teams als Reviewer hinzuzufügen, damit das Hugging Face-Team über zukünftige Änderungen informiert wird.
+   zukünftige Änderungen benachrichtigt wird.
+
+9. Ändern Sie den PR in einen Entwurf, indem Sie auf der rechten Seite der GitHub-Pull-Request-Webseite auf "In Entwurf umwandeln" klicken.
+
+
+Jetzt haben Sie eine Entwicklungsumgebung eingerichtet, um *BrandNewBert* nach TensorFlow in 🤗 Transformers zu portieren.
+
+
+**3. (Optional) Verstehen Sie die theoretischen Aspekte und die bestehende Implementierung**
+
+Sie sollten sich etwas Zeit nehmen, um die Arbeit von *BrandNewBert* zu lesen, falls eine solche Beschreibung existiert. Möglicherweise gibt es große
+Abschnitte des Papiers, die schwer zu verstehen sind. Wenn das der Fall ist, ist das in Ordnung - machen Sie sich keine Sorgen! Das Ziel ist
+ist es nicht, ein tiefes theoretisches Verständnis des Papiers zu erlangen, sondern die notwendigen Informationen zu extrahieren, um
+das Modell mit Hilfe von TensorFlow effektiv in 🤗 Transformers neu zu implementieren. Das heißt, Sie müssen nicht zu viel Zeit auf die
+viel Zeit auf die theoretischen Aspekte verwenden, sondern sich lieber auf die praktischen Aspekte konzentrieren, nämlich auf die bestehende Modelldokumentation
+Seite (z.B. [model docs for BERT](model_doc/bert)).
+
+Nachdem Sie die Grundlagen der Modelle, die Sie implementieren wollen, verstanden haben, ist es wichtig, die bestehende
+Implementierung zu verstehen. Dies ist eine gute Gelegenheit, sich zu vergewissern, dass eine funktionierende Implementierung mit Ihren Erwartungen an das
+Modell entspricht, und um technische Herausforderungen auf der TensorFlow-Seite vorauszusehen.
+
+Es ist ganz natürlich, dass Sie sich von der Menge an Informationen, die Sie gerade aufgesogen haben, überwältigt fühlen. Es ist
+Es ist definitiv nicht erforderlich, dass Sie in dieser Phase alle Facetten des Modells verstehen. Dennoch empfehlen wir Ihnen dringend
+ermutigen wir Sie, alle dringenden Fragen in unserem [Forum](https://discuss.huggingface.co/) zu klären.
+
+
+### 4. Implementierung des Modells
+
+Jetzt ist es an der Zeit, endlich mit dem Programmieren zu beginnen. Als Ausgangspunkt empfehlen wir die PyTorch-Datei selbst: Kopieren Sie den Inhalt von
+modeling_brand_new_bert.py` in `src/transformers/models/brand_new_bert/` nach
+modeling_tf_brand_new_bert.py`. Das Ziel dieses Abschnitts ist es, die Datei zu ändern und die Importstruktur von
+🤗 Transformers zu aktualisieren, so dass Sie `TFBrandNewBert` und
+`TFBrandNewBert.from_pretrained(model_repo, from_pt=True)` erfolgreich ein funktionierendes TensorFlow *BrandNewBert* Modell lädt.
+
+Leider gibt es kein Rezept, um ein PyTorch-Modell in TensorFlow zu konvertieren. Sie können jedoch unsere Auswahl an
+Tipps befolgen, um den Prozess so reibungslos wie möglich zu gestalten:
+- Stellen Sie `TF` dem Namen aller Klassen voran (z.B. wird `BrandNewBert` zu `TFBrandNewBert`).
+- Die meisten PyTorch-Operationen haben einen direkten TensorFlow-Ersatz. Zum Beispiel entspricht `torch.nn.Linear` der Klasse
+  `tf.keras.layers.Dense`, `torch.nn.Dropout` entspricht `tf.keras.layers.Dropout`, usw. Wenn Sie sich nicht sicher sind
+  über eine bestimmte Operation nicht sicher sind, können Sie die [TensorFlow-Dokumentation](https://www.tensorflow.org/api_docs/python/tf)
+  oder die [PyTorch-Dokumentation](https://pytorch.org/docs/stable/).
+- Suchen Sie nach Mustern in der Codebasis von 🤗 Transformers. Wenn Sie auf eine bestimmte Operation stoßen, für die es keinen direkten Ersatz gibt
+   Ersatz hat, stehen die Chancen gut, dass jemand anderes bereits das gleiche Problem hatte.
+- Behalten Sie standardmäßig die gleichen Variablennamen und die gleiche Struktur wie in PyTorch bei. Dies erleichtert die Fehlersuche, die Verfolgung von
+   Probleme zu verfolgen und spätere Korrekturen vorzunehmen.
+- Einige Ebenen haben in jedem Framework unterschiedliche Standardwerte. Ein bemerkenswertes Beispiel ist die Schicht für die Batch-Normalisierung
+   epsilon (`1e-5` in [PyTorch](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html#torch.nn.BatchNorm2d)
+   und `1e-3` in [TensorFlow](https://www.tensorflow.org/api_docs/python/tf/keras/layers/BatchNormalization)).
+   Prüfen Sie die Dokumentation genau!
+- Die Variablen `nn.Parameter` von PyTorch müssen in der Regel innerhalb von TF Layer's `build()` initialisiert werden. Siehe das folgende
+   Beispiel: [PyTorch](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_vit_mae.py#L212) /
+   [TensorFlow](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_tf_vit_mae.py#L220)
+- Wenn das PyTorch-Modell ein `#copied from ...` am Anfang einer Funktion hat, stehen die Chancen gut, dass Ihr TensorFlow-Modell diese Funktion auch
+   diese Funktion von der Architektur ausleihen kann, von der sie kopiert wurde, vorausgesetzt, es hat eine TensorFlow-Architektur.
+- Die korrekte Zuweisung des Attributs `name` in TensorFlow-Funktionen ist entscheidend, um das `from_pt=True` Gewicht zu erreichen
+   Cross-Loading. Name" ist fast immer der Name der entsprechenden Variablen im PyTorch-Code. Wenn `name` nicht
+   nicht richtig gesetzt ist, sehen Sie dies in der Fehlermeldung beim Laden der Modellgewichte.
+- Die Logik der Basismodellklasse, `BrandNewBertModel`, befindet sich in `TFBrandNewBertMainLayer`, einer Keras
+   Schicht-Unterklasse ([Beispiel](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L719)).
+   TFBrandNewBertModel" ist lediglich ein Wrapper für diese Schicht.
+- Keras-Modelle müssen erstellt werden, um die vorher trainierten Gewichte zu laden. Aus diesem Grund muss `TFBrandNewBertPreTrainedModel`
+   ein Beispiel für die Eingaben in das Modell enthalten, die `dummy_inputs`
+   ([Beispiel](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L916)).
+- Wenn Sie nicht weiterkommen, fragen Sie nach Hilfe - wir sind für Sie da! 🤗
+
+Neben der Modelldatei selbst müssen Sie auch die Verweise auf die Modellklassen und die zugehörigen
+Dokumentationsseiten hinzufügen. Sie können diesen Teil ganz nach den Mustern in anderen PRs erledigen
+([Beispiel](https://github.com/huggingface/transformers/pull/18020/files)). Hier ist eine Liste der erforderlichen manuellen
+Änderungen:
+- Fügen Sie alle öffentlichen Klassen von *BrandNewBert* in `src/transformers/__init__.py` ein.
+- Fügen Sie *BrandNewBert* Klassen zu den entsprechenden Auto Klassen in `src/transformers/models/auto/modeling_tf_auto.py` hinzu.
+- Fügen Sie die *BrandNewBert* zugehörigen Klassen für träges Laden in `src/transformers/utils/dummy_tf_objects.py` hinzu.
+- Aktualisieren Sie die Importstrukturen für die öffentlichen Klassen in `src/transformers/models/brand_new_bert/__init__.py`.
+- Fügen Sie die Dokumentationszeiger auf die öffentlichen Methoden von *BrandNewBert* in `docs/source/de/model_doc/brand_new_bert.md` hinzu.
+- Fügen Sie sich selbst zur Liste der Mitwirkenden an *BrandNewBert* in `docs/source/de/model_doc/brand_new_bert.md` hinzu.
+- Fügen Sie schließlich ein grünes Häkchen ✅ in der TensorFlow-Spalte von *BrandNewBert* in `docs/source/de/index.md` hinzu.
+
+Wenn Sie mit Ihrer Implementierung zufrieden sind, führen Sie die folgende Checkliste aus, um zu bestätigen, dass Ihre Modellarchitektur
+fertig ist:
+1. Alle Schichten, die sich zur Trainingszeit anders verhalten (z.B. Dropout), werden mit einem `Training` Argument aufgerufen, das
+von den Top-Level-Klassen weitergegeben wird
+2. Sie haben `#copied from ...` verwendet, wann immer es möglich war.
+3. Die Funktion `TFBrandNewBertMainLayer` und alle Klassen, die sie verwenden, haben ihre Funktion `call` mit `@unpack_inputs` dekoriert
+4. TFBrandNewBertMainLayer` ist mit `@keras_serializable` dekoriert
+5. Ein TensorFlow-Modell kann aus PyTorch-Gewichten mit `TFBrandNewBert.from_pretrained(model_repo, from_pt=True)` geladen werden.
+6. Sie können das TensorFlow Modell mit dem erwarteten Eingabeformat aufrufen
+
+
+### 5. Modell-Tests hinzufügen
+
+Hurra, Sie haben ein TensorFlow-Modell implementiert! Jetzt ist es an der Zeit, Tests hinzuzufügen, um sicherzustellen, dass sich Ihr Modell wie erwartet verhält.
+erwartet. Wie im vorigen Abschnitt schlagen wir vor, dass Sie zunächst die Datei `test_modeling_brand_new_bert.py` in
+`tests/models/brand_new_bert/` in die Datei `test_modeling_tf_brand_new_bert.py` zu kopieren und dann die notwendigen
+TensorFlow-Ersetzungen vornehmen. Für den Moment sollten Sie in allen Aufrufen von `.from_pretrained()` das Flag `from_pt=True` verwenden, um die
+die vorhandenen PyTorch-Gewichte zu laden.
+
+Wenn Sie damit fertig sind, kommt der Moment der Wahrheit: Führen Sie die Tests durch! 😬
+
+```bash
+NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
+py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
+```
+
+Das wahrscheinlichste Ergebnis ist, dass Sie eine Reihe von Fehlern sehen werden. Machen Sie sich keine Sorgen, das ist zu erwarten! Das Debuggen von ML-Modellen ist
+notorisch schwierig, und der Schlüssel zum Erfolg ist Geduld (und `breakpoint()`). Nach unserer Erfahrung sind die schwierigsten
+Probleme aus subtilen Unstimmigkeiten zwischen ML-Frameworks, zu denen wir am Ende dieses Leitfadens ein paar Hinweise geben.
+In anderen Fällen kann es sein, dass ein allgemeiner Test nicht direkt auf Ihr Modell anwendbar ist; in diesem Fall empfehlen wir eine Überschreibung
+auf der Ebene der Modelltestklasse. Zögern Sie nicht, in Ihrem Entwurf einer Pull-Anfrage um Hilfe zu bitten, wenn
+Sie nicht weiterkommen.
+
+Wenn alle Tests erfolgreich waren, können Sie Ihr Modell in die 🤗 Transformers-Bibliothek aufnehmen! 🎉
+
+### 6.-7. Stellen Sie sicher, dass jeder Ihr Modell verwenden kann
+
+**6. Reichen Sie den Pull Request ein**
+
+Sobald Sie mit der Implementierung und den Tests fertig sind, ist es an der Zeit, eine Pull-Anfrage einzureichen. Bevor Sie Ihren Code einreichen,
+führen Sie unser Dienstprogramm zur Codeformatierung, `make fixup` 🪄, aus. Damit werden automatisch alle Formatierungsfehler behoben, die dazu führen würden, dass
+unsere automatischen Prüfungen fehlschlagen würden.
+
+Nun ist es an der Zeit, Ihren Entwurf einer Pull-Anfrage in eine echte Pull-Anfrage umzuwandeln. Klicken Sie dazu auf die Schaltfläche "Bereit für
+Review" und fügen Sie Joao (`@gante`) und Matt (`@Rocketknight1`) als Reviewer hinzu. Eine Modell-Pull-Anfrage benötigt
+mindestens 3 Reviewer, aber sie werden sich darum kümmern, geeignete zusätzliche Reviewer für Ihr Modell zu finden.
+
+Nachdem alle Gutachter mit dem Stand Ihres PR zufrieden sind, entfernen Sie als letzten Aktionspunkt das Flag `from_pt=True` in
+.from_pretrained()-Aufrufen zu entfernen. Da es keine TensorFlow-Gewichte gibt, müssen Sie sie hinzufügen! Lesen Sie den Abschnitt
+unten, um zu erfahren, wie Sie dies tun können.
+
+Wenn schließlich die TensorFlow-Gewichte zusammengeführt werden, Sie mindestens 3 Genehmigungen von Prüfern haben und alle CI-Checks grün sind
+grün sind, überprüfen Sie die Tests ein letztes Mal lokal
+
+```bash
+NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
+py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
+```
+
+und wir werden Ihren PR zusammenführen! Herzlichen Glückwunsch zu dem Meilenstein 🎉.
+
+**7. (Optional) Erstellen Sie Demos und teilen Sie sie mit der Welt**
+
+Eine der schwierigsten Aufgaben bei Open-Source ist die Entdeckung. Wie können die anderen Benutzer von der Existenz Ihres
+fabelhaften TensorFlow-Beitrags erfahren? Mit der richtigen Kommunikation, natürlich! 📣
+
+Es gibt vor allem zwei Möglichkeiten, Ihr Modell mit der Community zu teilen:
+- Erstellen Sie Demos. Dazu gehören Gradio-Demos, Notebooks und andere unterhaltsame Möglichkeiten, Ihr Modell vorzuführen. Wir raten Ihnen
+   ermutigen Sie, ein Notizbuch zu unseren [community-driven demos](https://huggingface.co/docs/transformers/community) hinzuzufügen.
+- Teilen Sie Geschichten in sozialen Medien wie Twitter und LinkedIn. Sie sollten stolz auf Ihre Arbeit sein und sie mit der
+   Ihre Leistung mit der Community teilen - Ihr Modell kann nun von Tausenden von Ingenieuren und Forschern auf der ganzen Welt genutzt werden
+   der Welt genutzt werden 🌍! Wir werden Ihre Beiträge gerne retweeten und Ihnen helfen, Ihre Arbeit mit der Community zu teilen.
+
+
+## Hinzufügen von TensorFlow-Gewichten zum 🤗 Hub
+
+Unter der Annahme, dass die TensorFlow-Modellarchitektur in 🤗 Transformers verfügbar ist, ist die Umwandlung von PyTorch-Gewichten in
+TensorFlow-Gewichte ist ein Kinderspiel!
+
+Hier sehen Sie, wie es geht:
+1. Stellen Sie sicher, dass Sie in Ihrem Terminal bei Ihrem Hugging Face Konto angemeldet sind. Sie können sich mit dem folgenden Befehl anmelden
+   `huggingface-cli login` (Ihre Zugangstoken finden Sie [hier](https://huggingface.co/settings/tokens))
+2. Führen Sie `transformers-cli pt-to-tf --model-name foo/bar` aus, wobei `foo/bar` der Name des Modell-Repositorys ist
+   ist, das die PyTorch-Gewichte enthält, die Sie konvertieren möchten.
+3. Markieren Sie `@joaogante` und `@Rocketknight1` in dem 🤗 Hub PR, den der obige Befehl gerade erstellt hat
+
+Das war's! 🎉
+
+
+## Fehlersuche in verschiedenen ML-Frameworks 🐛
+
+Irgendwann, wenn Sie eine neue Architektur hinzufügen oder TensorFlow-Gewichte für eine bestehende Architektur erstellen, werden Sie
+stoßen Sie vielleicht auf Fehler, die sich über Unstimmigkeiten zwischen PyTorch und TensorFlow beschweren. Sie könnten sich sogar dazu entschließen, den
+Modellarchitektur-Code für die beiden Frameworks zu öffnen, und stellen fest, dass sie identisch aussehen. Was ist denn da los? 🤔
+
+Lassen Sie uns zunächst darüber sprechen, warum es wichtig ist, diese Diskrepanzen zu verstehen. Viele Community-Mitglieder werden 🤗
+Transformers-Modelle und vertrauen darauf, dass sich unsere Modelle wie erwartet verhalten. Wenn es eine große Diskrepanz gibt
+zwischen den beiden Frameworks auftritt, bedeutet dies, dass das Modell nicht der Referenzimplementierung für mindestens eines der Frameworks folgt.
+der Frameworks folgt. Dies kann zu stillen Fehlern führen, bei denen das Modell zwar läuft, aber eine schlechte Leistung aufweist. Dies ist
+wohl schlimmer als ein Modell, das überhaupt nicht läuft! Aus diesem Grund streben wir an, dass die Abweichung zwischen den Frameworks kleiner als
+1e-5" in allen Phasen des Modells.
+
+Wie bei anderen numerischen Problemen auch, steckt der Teufel im Detail. Und wie bei jedem detailorientierten Handwerk ist die geheime
+Zutat hier Geduld. Hier ist unser Vorschlag für den Arbeitsablauf, wenn Sie auf diese Art von Problemen stoßen:
+1. Lokalisieren Sie die Quelle der Abweichungen. Das Modell, das Sie konvertieren, hat wahrscheinlich bis zu einem gewissen Punkt nahezu identische innere Variablen.
+   bestimmten Punkt. Platzieren Sie `Breakpoint()`-Anweisungen in den Architekturen der beiden Frameworks und vergleichen Sie die Werte der
+   numerischen Variablen von oben nach unten, bis Sie die Quelle der Probleme gefunden haben.
+2. Nachdem Sie nun die Ursache des Problems gefunden haben, setzen Sie sich mit dem 🤗 Transformers-Team in Verbindung. Es ist möglich
+   dass wir ein ähnliches Problem schon einmal gesehen haben und umgehend eine Lösung anbieten können. Als Ausweichmöglichkeit können Sie beliebte Seiten
+   wie StackOverflow und GitHub-Probleme.
+3. Wenn keine Lösung in Sicht ist, bedeutet das, dass Sie tiefer gehen müssen. Die gute Nachricht ist, dass Sie das Problem gefunden haben.
+   Problem ausfindig gemacht haben, so dass Sie sich auf die problematische Anweisung konzentrieren und den Rest des Modells ausblenden können! Die schlechte Nachricht ist
+   dass Sie sich in die Quellimplementierung der besagten Anweisung einarbeiten müssen. In manchen Fällen finden Sie vielleicht ein
+   Problem mit einer Referenzimplementierung - verzichten Sie nicht darauf, ein Problem im Upstream-Repository zu öffnen.
+
+In einigen Fällen können wir nach Rücksprache mit dem 🤗 Transformers-Team zu dem Schluss kommen, dass die Behebung der Abweichung nicht machbar ist.
+Wenn die Abweichung in den Ausgabeschichten des Modells sehr klein ist (aber möglicherweise groß in den versteckten Zuständen), können wir
+könnten wir beschließen, sie zu ignorieren und das Modell zu verteilen. Die oben erwähnte CLI `pt-to-tf` hat ein `--max-error`
+Flag, um die Fehlermeldung bei der Gewichtskonvertierung zu überschreiben.

docs/source/de/llm_tutorial.md

@@ -0,0 +1,221 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+
+# Generation with LLMs
+
+[[open-in-colab]]
+
+LLMs (Large Language Models) sind die Schlüsselkomponente bei der Texterstellung. Kurz gesagt, bestehen sie aus großen, vortrainierten Transformationsmodellen, die darauf trainiert sind, das nächste Wort (oder genauer gesagt Token) aus einem Eingabetext vorherzusagen. Da sie jeweils ein Token vorhersagen, müssen Sie etwas Aufwändigeres tun, um neue Sätze zu generieren, als nur das Modell aufzurufen - Sie müssen eine autoregressive Generierung durchführen.
+
+Die autoregressive Generierung ist ein Verfahren zur Inferenzzeit, bei dem ein Modell mit seinen eigenen generierten Ausgaben iterativ aufgerufen wird, wenn einige anfängliche Eingaben vorliegen. In 🤗 Transformers wird dies von der Methode [`~generation.GenerationMixin.generate`] übernommen, die allen Modellen mit generativen Fähigkeiten zur Verfügung steht.
+
+Dieses Tutorial zeigt Ihnen, wie Sie:
+
+* Text mit einem LLM generieren
+* Vermeiden Sie häufige Fallstricke
+* Nächste Schritte, damit Sie das Beste aus Ihrem LLM herausholen können
+
+Bevor Sie beginnen, stellen Sie sicher, dass Sie alle erforderlichen Bibliotheken installiert haben:
+
+```bash
+pip install transformers bitsandbytes>=0.39.0 -q
+```
+
+
+## Text generieren
+
+Ein Sprachmodell, das für [causal language modeling](tasks/language_modeling) trainiert wurde, nimmt eine Folge von Text-Token als Eingabe und gibt die Wahrscheinlichkeitsverteilung für das nächste Token zurück.
+
+<!-- [GIF 1 -- FWD PASS] -->
+<figure class="image table text-center m-0 w-full">
+    <video
+        style="max-width: 90%; margin: auto;"
+        autoplay loop muted playsinline
+        src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/assisted-generation/gif_1_1080p.mov"
+    ></video>
+    <figcaption>"Forward pass of an LLM"</figcaption>
+</figure>
+
+Ein wichtiger Aspekt der autoregressiven Generierung mit LLMs ist die Auswahl des nächsten Tokens aus dieser Wahrscheinlichkeitsverteilung. In diesem Schritt ist alles möglich, solange Sie am Ende ein Token für die nächste Iteration haben. Das heißt, es kann so einfach sein wie die Auswahl des wahrscheinlichsten Tokens aus der Wahrscheinlichkeitsverteilung oder so komplex wie die Anwendung von einem Dutzend Transformationen vor der Stichprobenziehung aus der resultierenden Verteilung.
+
+<!-- [GIF 2 -- TEXT GENERATION] -->
+<figure class="image table text-center m-0 w-full">
+    <video
+        style="max-width: 90%; margin: auto;"
+        autoplay loop muted playsinline
+        src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/assisted-generation/gif_2_1080p.mov"
+    ></video>
+    <figcaption>"Die autoregressive Generierung wählt iterativ das nächste Token aus einer Wahrscheinlichkeitsverteilung aus, um Text zu erzeugen"</figcaption>
+</figure>
+
+Der oben dargestellte Prozess wird iterativ wiederholt, bis eine bestimmte Abbruchbedingung erreicht ist. Im Idealfall wird die Abbruchbedingung vom Modell vorgegeben, das lernen sollte, wann es ein Ende-der-Sequenz-Token (EOS) ausgeben muss. Ist dies nicht der Fall, stoppt die Generierung, wenn eine vordefinierte Maximallänge erreicht ist.
+
+Damit sich Ihr Modell so verhält, wie Sie es für Ihre Aufgabe erwarten, müssen Sie den Schritt der Token-Auswahl und die Abbruchbedingung richtig einstellen. Aus diesem Grund haben wir zu jedem Modell eine [`~generation.GenerationConfig`]-Datei, die eine gute generative Standardparametrisierung enthält und zusammen mit Ihrem Modell geladen wird.
+
+Lassen Sie uns über Code sprechen!
+
+<Tip>
+
+Wenn Sie an der grundlegenden Verwendung von LLMs interessiert sind, ist unsere High-Level-Schnittstelle [`Pipeline`](pipeline_tutorial) ein guter Ausgangspunkt. LLMs erfordern jedoch oft fortgeschrittene Funktionen wie Quantisierung und Feinsteuerung des Token-Auswahlschritts, was am besten über [`~generation.GenerationMixin.generate`] erfolgt. Die autoregressive Generierung mit LLMs ist ebenfalls ressourcenintensiv und sollte für einen angemessenen Durchsatz auf einer GPU ausgeführt werden.
+
+</Tip>
+
+<!-- TODO: update example to llama 2 (or a newer popular baseline) when it becomes ungated -->
+Zunächst müssen Sie das Modell laden.
+
+```py
+>>> from transformers import AutoModelForCausalLM
+
+>>> model = AutoModelForCausalLM.from_pretrained(
+...     "openlm-research/open_llama_7b", device_map="auto", load_in_4bit=True
+... )
+```
+
+Sie werden zwei Flags in dem Aufruf `from_pretrained` bemerken:
+
+ - `device_map` stellt sicher, dass das Modell auf Ihre GPU(s) übertragen wird
+ - `load_in_4bit` wendet [dynamische 4-Bit-Quantisierung](main_classes/quantization) an, um die Ressourcenanforderungen massiv zu reduzieren
+
+Es gibt noch andere Möglichkeiten, ein Modell zu initialisieren, aber dies ist eine gute Grundlage, um mit einem LLM zu beginnen.
+
+Als nächstes müssen Sie Ihre Texteingabe mit einem [tokenizer](tokenizer_summary) vorverarbeiten.
+
+```py
+>>> from transformers import AutoTokenizer
+
+>>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
+>>> model_inputs = tokenizer(["A list of colors: red, blue"], return_tensors="pt").to("cuda")
+```
+
+Die Variable `model_inputs` enthält die tokenisierte Texteingabe sowie die Aufmerksamkeitsmaske. Obwohl [`~generation.GenerationMixin.generate`] sein Bestes tut, um die Aufmerksamkeitsmaske abzuleiten, wenn sie nicht übergeben wird, empfehlen wir, sie für optimale Ergebnisse wann immer möglich zu übergeben.
+
+Rufen Sie schließlich die Methode [~generation.GenerationMixin.generate] auf, um die generierten Token zurückzugeben, die vor dem Drucken in Text umgewandelt werden sollten.
+
+```py
+>>> generated_ids = model.generate(**model_inputs)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'A list of colors: red, blue, green, yellow, black, white, and brown'
+```
+
+Und das war's! Mit ein paar Zeilen Code können Sie sich die Macht eines LLM zunutze machen.
+
+
+## Häufige Fallstricke
+
+Es gibt viele [Generierungsstrategien](generation_strategies), und manchmal sind die Standardwerte für Ihren Anwendungsfall vielleicht nicht geeignet. Wenn Ihre Ausgaben nicht mit dem übereinstimmen, was Sie erwarten, haben wir eine Liste der häufigsten Fallstricke erstellt und wie Sie diese vermeiden können.
+
+```py
+>>> from transformers import AutoModelForCausalLM, AutoTokenizer
+
+>>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
+>>> tokenizer.pad_token = tokenizer.eos_token  # Llama has no pad token by default
+>>> model = AutoModelForCausalLM.from_pretrained(
+...     "openlm-research/open_llama_7b", device_map="auto", load_in_4bit=True
+... )
+```
+
+### Generierte Ausgabe ist zu kurz/lang
+
+Wenn in der Datei [~generation.GenerationConfig`] nichts angegeben ist, gibt `generate` standardmäßig bis zu 20 Token zurück. Wir empfehlen dringend, `max_new_tokens` in Ihrem `generate`-Aufruf manuell zu setzen, um die maximale Anzahl neuer Token zu kontrollieren, die zurückgegeben werden können. Beachten Sie, dass LLMs (genauer gesagt, [decoder-only models](https://huggingface.co/learn/nlp-course/chapter1/6?fw=pt)) auch die Eingabeaufforderung als Teil der Ausgabe zurückgeben.
+
+
+```py
+>>> model_inputs = tokenizer(["A sequence of numbers: 1, 2"], return_tensors="pt").to("cuda")
+
+>>> # By default, the output will contain up to 20 tokens
+>>> generated_ids = model.generate(**model_inputs)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'A sequence of numbers: 1, 2, 3, 4, 5'
+
+>>> # Setting `max_new_tokens` allows you to control the maximum length
+>>> generated_ids = model.generate(**model_inputs, max_new_tokens=50)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'A sequence of numbers: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,'
+```
+
+### Falscher Generierungsmodus
+
+Standardmäßig und sofern nicht in der Datei [~generation.GenerationConfig`] angegeben, wählt `generate` bei jeder Iteration das wahrscheinlichste Token aus (gierige Dekodierung). Je nach Aufgabe kann dies unerwünscht sein; kreative Aufgaben wie Chatbots oder das Schreiben eines Aufsatzes profitieren vom Sampling. Andererseits profitieren Aufgaben, bei denen es auf die Eingabe ankommt, wie z.B. Audiotranskription oder Übersetzung, von der gierigen Dekodierung. Aktivieren Sie das Sampling mit `do_sample=True`. Mehr zu diesem Thema erfahren Sie in diesem [Blogbeitrag] (https://huggingface.co/blog/how-to-generate).
+
+```py
+>>> # Set seed or reproducibility -- you don't need this unless you want full reproducibility
+>>> from transformers import set_seed
+>>> set_seed(0)
+
+>>> model_inputs = tokenizer(["I am a cat."], return_tensors="pt").to("cuda")
+
+>>> # LLM + greedy decoding = repetitive, boring output
+>>> generated_ids = model.generate(**model_inputs)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'I am a cat. I am a cat. I am a cat. I am a cat'
+
+>>> # With sampling, the output becomes more creative!
+>>> generated_ids = model.generate(**model_inputs, do_sample=True)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'I am a cat.\nI just need to be. I am always.\nEvery time'
+```
+
+### Falsche Auffüllseite
+
+LLMs sind [decoder-only](https://huggingface.co/learn/nlp-course/chapter1/6?fw=pt)-Architekturen, d.h. sie iterieren weiter über Ihre Eingabeaufforderung. Wenn Ihre Eingaben nicht die gleiche Länge haben, müssen sie aufgefüllt werden. Da LLMs nicht darauf trainiert sind, mit aufgefüllten Token fortzufahren, muss Ihre Eingabe links aufgefüllt werden. Vergessen Sie auch nicht, die Aufmerksamkeitsmaske an generate zu übergeben!
+
+```py
+>>> # The tokenizer initialized above has right-padding active by default: the 1st sequence,
+>>> # which is shorter, has padding on the right side. Generation fails.
+>>> model_inputs = tokenizer(
+...     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
+... ).to("cuda")
+>>> generated_ids = model.generate(**model_inputs)
+>>> tokenizer.batch_decode(generated_ids[0], skip_special_tokens=True)[0]
+''
+
+>>> # With left-padding, it works as expected!
+>>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b", padding_side="left")
+>>> tokenizer.pad_token = tokenizer.eos_token  # Llama has no pad token by default
+>>> model_inputs = tokenizer(
+...     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
+... ).to("cuda")
+>>> generated_ids = model.generate(**model_inputs)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'1, 2, 3, 4, 5, 6,'
+```
+
+<!-- TODO: when the prompting guide is ready, mention the importance of setting the right prompt in this section -->
+
+## Weitere Ressourcen
+
+Während der Prozess der autoregressiven Generierung relativ einfach ist, kann die optimale Nutzung Ihres LLM ein schwieriges Unterfangen sein, da es viele bewegliche Teile gibt. Für Ihre nächsten Schritte, die Ihnen helfen, tiefer in die LLM-Nutzung und das Verständnis einzutauchen:
+
+<!-- TODO: mit neuen Anleitungen vervollständigen -->
+### Fortgeschrittene Nutzung generieren
+
+1. [Leitfaden](generation_strategies) zur Steuerung verschiedener Generierungsmethoden, zur Einrichtung der Generierungskonfigurationsdatei und zum Streaming der Ausgabe;
+2. API-Referenz zu [`~generation.GenerationConfig`], [`~generation.GenerationMixin.generate`] und [generate-bezogene Klassen](internal/generation_utils).
+
+### LLM-Ranglisten
+
+1. [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard), das sich auf die Qualität der Open-Source-Modelle konzentriert;
+2. [Open LLM-Perf Leaderboard](https://huggingface.co/spaces/optimum/llm-perf-leaderboard), das sich auf den LLM-Durchsatz konzentriert.
+
+### Latenz und Durchsatz
+
+1. [Leitfaden](main_classes/quantization) zur dynamischen Quantisierung, der Ihnen zeigt, wie Sie Ihren Speicherbedarf drastisch reduzieren können.
+
+### Verwandte Bibliotheken
+
+1. [text-generation-inference](https://github.com/huggingface/text-generation-inference), ein produktionsreifer Server für LLMs;
+2. [`optimum`](https://github.com/huggingface/optimum), eine Erweiterung von 🤗 Transformers, die für bestimmte Hardware-Geräte optimiert.

docs/source/de/peft.md

@@ -0,0 +1,216 @@
+<!--Copyright 2023 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.
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+-->
+
+# Adapter mit 🤗 PEFT laden
+
+[[open-in-colab]]
+
+Die [Parameter-Efficient Fine Tuning (PEFT)](https://huggingface.co/blog/peft) Methoden frieren die vorab trainierten Modellparameter während der Feinabstimmung ein und fügen eine kleine Anzahl trainierbarer Parameter (die Adapter) hinzu. Die Adapter werden trainiert, um aufgabenspezifische Informationen zu lernen. Es hat sich gezeigt, dass dieser Ansatz sehr speichereffizient ist und weniger Rechenleistung beansprucht, während die Ergebnisse mit denen eines vollständig feinabgestimmten Modells vergleichbar sind. 
+
+Adapter, die mit PEFT trainiert wurden, sind in der Regel um eine Größenordnung kleiner als das vollständige Modell, so dass sie bequem gemeinsam genutzt, gespeichert und geladen werden können.
+
+<div class="flex flex-col justify-center">
+  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/peft/PEFT-hub-screenshot.png"/>
+  <figcaption class="text-center">Die Adaptergewichte für ein OPTForCausalLM-Modell, die auf dem Hub gespeichert sind, sind nur ~6MB groß, verglichen mit der vollen Größe der Modellgewichte, die ~700MB betragen können.</figcaption>
+</div>
+
+Wenn Sie mehr über die 🤗 PEFT-Bibliothek erfahren möchten, sehen Sie sich die [Dokumentation](https://huggingface.co/docs/peft/index) an.
+
+## Setup
+
+Starten Sie mit der Installation von 🤗 PEFT:
+
+```bash
+pip install peft
+```
+
+Wenn Sie die brandneuen Funktionen ausprobieren möchten, sollten Sie die Bibliothek aus dem Quellcode installieren:
+
+```bash
+pip install git+https://github.com/huggingface/peft.git
+```
+
+## Unterstützte PEFT-Modelle
+
+Transformers unterstützt nativ einige PEFT-Methoden, d.h. Sie können lokal oder auf dem Hub gespeicherte Adaptergewichte laden und sie mit wenigen Zeilen Code einfach ausführen oder trainieren. Die folgenden Methoden werden unterstützt:
+
+- [Low Rank Adapters](https://huggingface.co/docs/peft/conceptual_guides/lora)
+- [IA3](https://huggingface.co/docs/peft/conceptual_guides/ia3)
+- [AdaLoRA](https://arxiv.org/abs/2303.10512)
+
+Wenn Sie andere PEFT-Methoden, wie z.B. Prompt Learning oder Prompt Tuning, verwenden möchten, oder über die 🤗 PEFT-Bibliothek im Allgemeinen, lesen Sie bitte die [Dokumentation](https://huggingface.co/docs/peft/index).
+
+
+## Laden Sie einen PEFT-Adapter
+
+Um ein PEFT-Adaptermodell von 🤗 Transformers zu laden und zu verwenden, stellen Sie sicher, dass das Hub-Repository oder das lokale Verzeichnis eine `adapter_config.json`-Datei und die Adaptergewichte enthält, wie im obigen Beispielbild gezeigt. Dann können Sie das PEFT-Adaptermodell mit der Klasse `AutoModelFor` laden. Um zum Beispiel ein PEFT-Adaptermodell für die kausale Sprachmodellierung zu laden:
+
+1. Geben Sie die PEFT-Modell-ID an.
+2. übergeben Sie es an die Klasse [`AutoModelForCausalLM`].
+
+```py
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+peft_model_id = "ybelkada/opt-350m-lora"
+model = AutoModelForCausalLM.from_pretrained(peft_model_id)
+```
+
+<Tip>
+
+Sie können einen PEFT-Adapter entweder mit einer `AutoModelFor`-Klasse oder der Basismodellklasse wie `OPTForCausalLM` oder `LlamaForCausalLM` laden.
+
+</Tip>
+
+Sie können einen PEFT-Adapter auch laden, indem Sie die Methode `load_adapter` aufrufen:
+
+```py
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_id = "facebook/opt-350m"
+peft_model_id = "ybelkada/opt-350m-lora"
+
+model = AutoModelForCausalLM.from_pretrained(model_id)
+model.load_adapter(peft_model_id)
+```
+
+## Laden in 8bit oder 4bit
+
+Die `bitsandbytes`-Integration unterstützt Datentypen mit 8bit und 4bit Genauigkeit, was für das Laden großer Modelle nützlich ist, weil es Speicher spart (lesen Sie den `bitsandbytes`-Integrations [guide](./quantization#bitsandbytes-integration), um mehr zu erfahren). Fügen Sie die Parameter `load_in_8bit` oder `load_in_4bit` zu [`~PreTrainedModel.from_pretrained`] hinzu und setzen Sie `device_map="auto"`, um das Modell effektiv auf Ihre Hardware zu verteilen:
+
+```py
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+peft_model_id = "ybelkada/opt-350m-lora"
+model = AutoModelForCausalLM.from_pretrained(peft_model_id, device_map="auto", load_in_8bit=True)
+```
+
+## Einen neuen Adapter hinzufügen
+
+Sie können [`~peft.PeftModel.add_adapter`] verwenden, um einen neuen Adapter zu einem Modell mit einem bestehenden Adapter hinzuzufügen, solange der neue Adapter vom gleichen Typ ist wie der aktuelle Adapter. Wenn Sie zum Beispiel einen bestehenden LoRA-Adapter an ein Modell angehängt haben:
+
+```py
+from transformers import AutoModelForCausalLM, OPTForCausalLM, AutoTokenizer
+from peft import PeftConfig
+
+model_id = "facebook/opt-350m"
+model = AutoModelForCausalLM.from_pretrained(model_id)
+
+lora_config = LoraConfig(
+    target_modules=["q_proj", "k_proj"],
+    init_lora_weights=False
+)
+
+model.add_adapter(lora_config, adapter_name="adapter_1")
+```
+
+Um einen neuen Adapter hinzuzufügen:
+
+```py
+# attach new adapter with same config
+model.add_adapter(lora_config, adapter_name="adapter_2")
+```
+
+Jetzt können Sie mit [`~peft.PeftModel.set_adapter`] festlegen, welcher Adapter verwendet werden soll:
+
+```py
+# use adapter_1
+model.set_adapter("adapter_1")
+output = model.generate(**inputs)
+print(tokenizer.decode(output_disabled[0], skip_special_tokens=True))
+
+# use adapter_2
+model.set_adapter("adapter_2")
+output_enabled = model.generate(**inputs)
+print(tokenizer.decode(output_enabled[0], skip_special_tokens=True))
+```
+
+## Aktivieren und Deaktivieren von Adaptern
+
+Sobald Sie einen Adapter zu einem Modell hinzugefügt haben, können Sie das Adaptermodul aktivieren oder deaktivieren. So aktivieren Sie das Adaptermodul:
+
+```py
+from transformers import AutoModelForCausalLM, OPTForCausalLM, AutoTokenizer
+from peft import PeftConfig
+
+model_id = "facebook/opt-350m"
+adapter_model_id = "ybelkada/opt-350m-lora"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+text = "Hello"
+inputs = tokenizer(text, return_tensors="pt")
+
+model = AutoModelForCausalLM.from_pretrained(model_id)
+peft_config = PeftConfig.from_pretrained(adapter_model_id)
+
+# to initiate with random weights
+peft_config.init_lora_weights = False
+
+model.add_adapter(peft_config)
+model.enable_adapters()
+output = model.generate(**inputs)
+```
+
+So deaktivieren Sie das Adaptermodul:
+
+```py
+model.disable_adapters()
+output = model.generate(**inputs)
+```
+
+## PEFT-Adapter trainieren
+
+PEFT-Adapter werden von der Klasse [`Trainer`] unterstützt, so dass Sie einen Adapter für Ihren speziellen Anwendungsfall trainieren können. Dazu müssen Sie nur ein paar weitere Codezeilen hinzufügen. Zum Beispiel, um einen LoRA-Adapter zu trainieren:
+
+<Tip>
+
+Wenn Sie mit der Feinabstimmung eines Modells mit [`Trainer`] noch nicht vertraut sind, werfen Sie einen Blick auf das Tutorial [Feinabstimmung eines vortrainierten Modells](Training).
+
+</Tip>
+
+1. Definieren Sie Ihre Adapterkonfiguration mit dem Aufgabentyp und den Hyperparametern (siehe [`~peft.LoraConfig`] für weitere Details darüber, was die Hyperparameter tun).
+
+```py
+from peft import LoraConfig
+
+peft_config = LoraConfig(
+    lora_alpha=16,
+    lora_dropout=0.1,
+    r=64,
+    bias="none",
+    task_type="CAUSAL_LM",
+)
+```
+
+2. Fügen Sie dem Modell einen Adapter hinzu.
+
+```py
+model.add_adapter(peft_config)
+```
+
+3. Jetzt können Sie das Modell an [`Trainer`] übergeben!
+
+```py
+trainer = Trainer(model=model, ...)
+trainer.train()
+```
+
+So speichern Sie Ihren trainierten Adapter und laden ihn wieder:
+
+```py
+model.save_pretrained(save_dir)
+model = AutoModelForCausalLM.from_pretrained(save_dir)
+```
+
+<!--
+TODO: (@younesbelkada @stevhliu)
+-   Link to PEFT docs for further details
+-   Trainer  
+-   8-bit / 4-bit examples ?
+-->

docs/source/de/pr_checks.md

@@ -0,0 +1,199 @@
+<!---
+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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Überprüfungen bei einer Pull-Anfrage
+
+Wenn Sie eine Pull-Anfrage für 🤗 Transformers öffnen, wird eine ganze Reihe von Prüfungen durchgeführt, um sicherzustellen, dass der Patch, den Sie hinzufügen, nichts Bestehendes zerstört. Es gibt vier Arten von Prüfungen:
+- reguläre Tests
+- Erstellung der Dokumentation
+- Stil von Code und Dokumentation
+- allgemeine Konsistenz des Repository
+
+In diesem Dokument werden wir versuchen zu erklären, worum es sich bei diesen verschiedenen Prüfungen handelt und wie Sie sie lokal debuggen können, wenn eine der Prüfungen in Ihrer PR fehlschlägt.
+
+Beachten Sie, dass Sie im Idealfall eine Dev-Installation benötigen:
+
+```bash
+pip install transformers[dev]
+```
+
+oder für eine bearbeitbare Installation:
+
+```bash
+pip install -e .[dev]
+```
+
+innerhalb des Transformers Repo. Da die Anzahl der optionalen Abhängigkeiten von Transformers stark zugenommen hat, ist es möglich, dass Sie nicht alle davon bekommen können. Wenn die Dev-Installation fehlschlägt, stellen Sie sicher, dass Sie das Deep Learning-Framework, mit dem Sie arbeiten, installieren (PyTorch, TensorFlow und/oder Flax).
+
+```bash
+pip install transformers[quality]
+```
+
+oder für eine bearbeitbare Installation:
+
+```bash
+pip install -e .[quality]
+```
+
+
+## Tests
+
+Alle Jobs, die mit `ci/circleci: run_tests_` beginnen, führen Teile der Transformers-Testsuite aus. Jeder dieser Jobs konzentriert sich auf einen Teil der Bibliothek in einer bestimmten Umgebung: `ci/circleci: run_tests_pipelines_tf` zum Beispiel führt den Pipelines-Test in einer Umgebung aus, in der nur TensorFlow installiert ist.
+
+Beachten Sie, dass nur ein Teil der Testsuite jedes Mal ausgeführt wird, um zu vermeiden, dass Tests ausgeführt werden, wenn es keine wirkliche Änderung in den Modulen gibt, die sie testen: ein Dienstprogramm wird ausgeführt, um die Unterschiede in der Bibliothek zwischen vor und nach dem PR zu ermitteln (was GitHub Ihnen auf der Registerkarte "Files changes" anzeigt) und die Tests auszuwählen, die von diesem Unterschied betroffen sind. Dieses Dienstprogramm kann lokal mit ausgeführt werden:
+
+```bash
+python utils/tests_fetcher.py
+```
+
+aus dem Stammverzeichnis des Transformers-Repositoriums. Es wird:
+
+1. Überprüfen Sie für jede Datei im Diff, ob die Änderungen im Code oder nur in Kommentaren oder Docstrings enthalten sind. Nur die Dateien mit echten Codeänderungen werden beibehalten.
+2. Erstellen Sie eine interne Map, die für jede Datei des Quellcodes der Bibliothek alle Dateien angibt, auf die sie rekursiv Einfluss nimmt. Von Modul A wird gesagt, dass es sich auf Modul B auswirkt, wenn Modul B Modul A importiert. Für die rekursive Auswirkung benötigen wir eine Kette von Modulen, die von Modul A zu Modul B führt und in der jedes Modul das vorherige importiert.
+3. Wenden Sie diese Zuordnung auf die in Schritt 1 gesammelten Dateien an. So erhalten wir die Liste der Modelldateien, die von der PR betroffen sind.
+4. Ordnen Sie jede dieser Dateien der/den entsprechenden Testdatei(en) zu und erhalten Sie die Liste der auszuführenden Tests.
+
+Wenn Sie das Skript lokal ausführen, sollten Sie die Ergebnisse von Schritt 1, 3 und 4 ausgegeben bekommen und somit wissen, welche Tests ausgeführt werden. Das Skript erstellt außerdem eine Datei namens `test_list.txt`, die die Liste der auszuführenden Tests enthält, die Sie mit dem folgenden Befehl lokal ausführen können:
+
+```bash
+python -m pytest -n 8 --dist=loadfile -rA -s $(cat test_list.txt)
+```
+
+Für den Fall, dass Ihnen etwas entgangen ist, wird die komplette Testreihe ebenfalls täglich ausgeführt.
+
+## Dokumentation erstellen
+
+Der Job `build_pr_documentation` erstellt und generiert eine Vorschau der Dokumentation, um sicherzustellen, dass alles in Ordnung ist, wenn Ihr PR zusammengeführt wird. Ein Bot fügt einen Link zur Vorschau der Dokumentation zu Ihrem PR hinzu. Alle Änderungen, die Sie an dem PR vornehmen, werden automatisch in der Vorschau aktualisiert. Wenn die Dokumentation nicht erstellt werden kann, klicken Sie auf **Details** neben dem fehlgeschlagenen Auftrag, um zu sehen, wo der Fehler liegt. Oft ist der Fehler so einfach wie eine fehlende Datei im `toctree`.
+
+Wenn Sie daran interessiert sind, die Dokumentation lokal zu erstellen oder in der Vorschau anzusehen, werfen Sie einen Blick in die [`README.md`](https://github.com/huggingface/transformers/tree/main/docs) im Ordner docs.
+
+## Code und Dokumentationsstil
+
+Die Formatierung des Codes erfolgt für alle Quelldateien, die Beispiele und die Tests mit `black` und `ruff`. Wir haben auch ein benutzerdefiniertes Tool, das sich um die Formatierung von docstrings und `rst`-Dateien kümmert (`utils/style_doc.py`), sowie um die Reihenfolge der Lazy-Importe, die in den Transformers `__init__.py`-Dateien durchgeführt werden (`utils/custom_init_isort.py`). All dies können Sie starten, indem Sie Folgendes ausführen
+
+```bash
+make style
+```
+
+Das CI prüft, ob diese innerhalb der Prüfung `ci/circleci: check_code_quality` angewendet wurden. Es führt auch `ruff` aus, das einen grundlegenden Blick auf Ihren Code wirft und sich beschwert, wenn es eine undefinierte Variable findet oder eine, die nicht verwendet wird. Um diese Prüfung lokal auszuführen, verwenden Sie
+
+```bash
+make quality
+```
+
+Dies kann sehr viel Zeit in Anspruch nehmen. Um dasselbe nur für die Dateien zu tun, die Sie im aktuellen Zweig geändert haben, führen Sie
+
+```bash
+make fixup
+```
+
+Dieser letzte Befehl führt auch alle zusätzlichen Prüfungen für die Konsistenz des Repositorys durch. Schauen wir uns diese an.
+
+## Repository-Konsistenz
+
+Dies fasst alle Tests zusammen, die sicherstellen, dass Ihr PR das Repository in einem guten Zustand verlässt. Sie können diese Prüfung lokal durchführen, indem Sie Folgendes ausführen:
+
+```bash
+make repo-consistency
+```
+
+Dies überprüft, ob:
+
+- Alle zum Init hinzugefügten Objekte sind dokumentiert (ausgeführt von `utils/check_repo.py`)
+- Alle `__init__.py`-Dateien haben in ihren beiden Abschnitten den gleichen Inhalt (ausgeführt von `utils/check_inits.py`)
+- Der gesamte Code, der als Kopie eines anderen Moduls identifiziert wurde, stimmt mit dem Original überein (ausgeführt von `utils/check_copies.py`)
+- Alle Konfigurationsklassen haben mindestens einen gültigen Prüfpunkt, der in ihren Dokumentationen erwähnt wird (ausgeführt von `utils/check_config_docstrings.py`)
+- Alle Konfigurationsklassen enthalten nur Attribute, die in den entsprechenden Modellierungsdateien verwendet werden (ausgeführt von `utils/check_config_attributes.py`)
+- Die Übersetzungen der READMEs und der Index des Dokuments haben die gleiche Modellliste wie die Haupt-README (durchgeführt von `utils/check_copies.py`)
+- Die automatisch generierten Tabellen in der Dokumentation sind auf dem neuesten Stand (ausgeführt von `utils/check_table.py`)
+- Die Bibliothek verfügt über alle Objekte, auch wenn nicht alle optionalen Abhängigkeiten installiert sind (ausgeführt von `utils/check_dummies.py`)
+
+Sollte diese Prüfung fehlschlagen, müssen die ersten beiden Punkte manuell korrigiert werden, die letzten vier können automatisch für Sie korrigiert werden, indem Sie den Befehl
+
+```bash
+make fix-copies
+```
+
+Zusätzliche Prüfungen betreffen PRs, die neue Modelle hinzufügen, vor allem, dass:
+
+- Alle hinzugefügten Modelle befinden sich in einer Auto-Zuordnung (durchgeführt von `utils/check_repo.py`)
+<!-- TODO Sylvain, add a check that makes sure the common tests are implemented.-->
+- Alle Modelle werden ordnungsgemäß getestet (ausgeführt von `utils/check_repo.py`)
+
+<!-- TODO Sylvain, add the following
+- All models are added to the main README, inside the main doc
+- All checkpoints used actually exist on the Hub
+
+-->
+
+### Kopien prüfen
+
+Da die Transformers-Bibliothek in Bezug auf den Modellcode sehr eigenwillig ist und jedes Modell vollständig in einer einzigen Datei implementiert sein sollte, ohne sich auf andere Modelle zu stützen, haben wir einen Mechanismus hinzugefügt, der überprüft, ob eine Kopie des Codes einer Ebene eines bestimmten Modells mit dem Original übereinstimmt. Auf diese Weise können wir bei einer Fehlerbehebung alle anderen betroffenen Modelle sehen und entscheiden, ob wir die Änderung weitergeben oder die Kopie zerstören.
+
+<Tip>
+
+Wenn eine Datei eine vollständige Kopie einer anderen Datei ist, sollten Sie sie in der Konstante `FULL_COPIES` von `utils/check_copies.py` registrieren.
+
+</Tip>
+
+Dieser Mechanismus stützt sich auf Kommentare der Form `# Kopiert von xxx`. Das `xxx` sollte den gesamten Pfad zu der Klasse der Funktion enthalten, die darunter kopiert wird. Zum Beispiel ist `RobertaSelfOutput` eine direkte Kopie der Klasse `BertSelfOutput`. Sie können also [hier](https://github.com/huggingface/transformers/blob/2bd7a27a671fd1d98059124024f580f8f5c0f3b5/src/transformers/models/roberta/modeling_roberta.py#L289) sehen, dass sie einen Kommentar hat:
+
+```py
+# Copied from transformers.models.bert.modeling_bert.BertSelfOutput
+```
+
+Beachten Sie, dass Sie dies nicht auf eine ganze Klasse anwenden, sondern auf die entsprechenden Methoden, von denen kopiert wird. Zum Beispiel [hier](https://github.com/huggingface/transformers/blob/2bd7a27a671fd1d98059124024f580f8f5c0f3b5/src/transformers/models/roberta/modeling_roberta.py#L598) können Sie sehen, wie `RobertaPreTrainedModel._init_weights` von der gleichen Methode in `BertPreTrainedModel` mit dem Kommentar kopiert wird:
+
+```py
+# Copied from transformers.models.bert.modeling_bert.BertPreTrainedModel._init_weights
+```
+
+Manchmal ist die Kopie bis auf die Namen genau gleich: zum Beispiel verwenden wir in `RobertaAttention` `RobertaSelfAttention` anstelle von `BertSelfAttention`, aber ansonsten ist der Code genau derselbe. Aus diesem Grund unterstützt `#Copied from` einfache String-Ersetzungen mit der folgenden Syntax: `Kopiert von xxx mit foo->bar`. Das bedeutet, dass der Code kopiert wird, wobei alle Instanzen von "foo" durch "bar" ersetzt werden. Sie können sehen, wie es [hier](https://github.com/huggingface/transformers/blob/2bd7a27a671fd1d98059124024f580f8f5c0f3b5/src/transformers/models/roberta/modeling_roberta.py#L304C1-L304C86) in `RobertaAttention` mit dem Kommentar verwendet wird:
+
+```py
+# Copied from transformers.models.bert.modeling_bert.BertAttention with Bert->Roberta
+```
+
+Beachten Sie, dass um den Pfeil herum keine Leerzeichen stehen sollten (es sei denn, das Leerzeichen ist Teil des zu ersetzenden Musters, natürlich).
+
+Sie können mehrere Muster durch ein Komma getrennt hinzufügen. Zum Beispiel ist hier `CamemberForMaskedLM` eine direkte Kopie von `RobertaForMaskedLM` mit zwei Ersetzungen: `Roberta` zu `Camembert` und `ROBERTA` zu `CAMEMBERT`. Sie können [hier](https://github.com/huggingface/transformers/blob/15082a9dc6950ecae63a0d3e5060b2fc7f15050a/src/transformers/models/camembert/modeling_camembert.py#L929) sehen, wie dies mit dem Kommentar gemacht wird:
+
+```py
+# Copied from transformers.models.roberta.modeling_roberta.RobertaForMaskedLM with Roberta->Camembert, ROBERTA->CAMEMBERT
+```
+
+Wenn die Reihenfolge eine Rolle spielt (weil eine der Ersetzungen mit einer vorherigen in Konflikt geraten könnte), werden die Ersetzungen von links nach rechts ausgeführt.
+
+<Tip>
+
+Wenn die Ersetzungen die Formatierung ändern (wenn Sie z.B. einen kurzen Namen durch einen sehr langen Namen ersetzen), wird die Kopie nach Anwendung des automatischen Formats überprüft.
+
+</Tip>
+
+Eine andere Möglichkeit, wenn es sich bei den Mustern nur um verschiedene Umschreibungen derselben Ersetzung handelt (mit einer groß- und einer kleingeschriebenen Variante), besteht darin, die Option `all-casing` hinzuzufügen. [Hier](https://github.com/huggingface/transformers/blob/15082a9dc6950ecae63a0d3e5060b2fc7f15050a/src/transformers/models/mobilebert/modeling_mobilebert.py#L1237) ist ein Beispiel in `MobileBertForSequenceClassification` mit dem Kommentar:
+
+```py
+# Copied from transformers.models.bert.modeling_bert.BertForSequenceClassification with Bert->MobileBert all-casing
+```
+
+In diesem Fall wird der Code von `BertForSequenceClassification` kopiert, indem er ersetzt wird:
+- `Bert` durch `MobileBert` (zum Beispiel bei der Verwendung von `MobileBertModel` in der Init)
+- `bert` durch `mobilebert` (zum Beispiel bei der Definition von `self.mobilebert`)
+- `BERT` durch `MOBILEBERT` (in der Konstante `MOBILEBERT_INPUTS_DOCSTRING`)

docs/source/de/run_scripts.md

@@ -0,0 +1,351 @@
+<!--Copyright 2022 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Trainieren mit einem Skript
+
+Neben den 🤗 Transformers [notebooks](./noteboks/README) gibt es auch Beispielskripte, die zeigen, wie man ein Modell für eine Aufgabe mit [PyTorch](https://github.com/huggingface/transformers/tree/main/examples/pytorch), [TensorFlow](https://github.com/huggingface/transformers/tree/main/examples/tensorflow) oder [JAX/Flax](https://github.com/huggingface/transformers/tree/main/examples/flax) trainiert.
+
+Sie werden auch Skripte finden, die wir in unseren [Forschungsprojekten](https://github.com/huggingface/transformers/tree/main/examples/research_projects) und [Legacy-Beispielen](https://github.com/huggingface/transformers/tree/main/examples/legacy) verwendet haben und die größtenteils von der Community stammen. Diese Skripte werden nicht aktiv gepflegt und erfordern eine bestimmte Version von 🤗 Transformers, die höchstwahrscheinlich nicht mit der neuesten Version der Bibliothek kompatibel ist.
+
+Es wird nicht erwartet, dass die Beispielskripte bei jedem Problem sofort funktionieren. Möglicherweise müssen Sie das Skript an das Problem anpassen, das Sie zu lösen versuchen. Um Ihnen dabei zu helfen, legen die meisten Skripte vollständig offen, wie die Daten vorverarbeitet werden, so dass Sie sie nach Bedarf für Ihren Anwendungsfall bearbeiten können.
+
+Für jede Funktion, die Sie in einem Beispielskript implementieren möchten, diskutieren Sie bitte im [Forum] (https://discuss.huggingface.co/) oder in einem [issue] (https://github.com/huggingface/transformers/issues), bevor Sie einen Pull Request einreichen. Wir freuen uns zwar über Fehlerkorrekturen, aber es ist unwahrscheinlich, dass wir einen Pull Request zusammenführen, der mehr Funktionalität auf Kosten der Lesbarkeit hinzufügt.
+
+Diese Anleitung zeigt Ihnen, wie Sie ein Beispiel für ein Trainingsskript zur Zusammenfassung in [PyTorch](https://github.com/huggingface/transformers/tree/main/examples/pytorch/summarization) und [TensorFlow](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/summarization) ausführen können. Sofern nicht anders angegeben, sollten alle Beispiele mit beiden Frameworks funktionieren.
+
+## Einrichtung
+
+Um die neueste Version der Beispielskripte erfolgreich auszuführen, **müssen Sie 🤗 Transformers aus dem Quellcode** in einer neuen virtuellen Umgebung installieren:
+
+```bash
+git clone https://github.com/huggingface/transformers
+cd transformers
+pip install .
+```
+
+Für ältere Versionen der Beispielskripte klicken Sie auf die Umschalttaste unten:
+
+<details>
+  <summary>Beispiele für ältere Versionen von 🤗 Transformers</summary>
+	<ul>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.5.1/examples">v4.5.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.4.2/examples">v4.4.2</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.3.3/examples">v4.3.3</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.2.2/examples">v4.2.2</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.1.1/examples">v4.1.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.0.1/examples">v4.0.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.5.1/examples">v3.5.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.4.0/examples">v3.4.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.3.1/examples">v3.3.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.2.0/examples">v3.2.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.1.0/examples">v3.1.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.0.2/examples">v3.0.2</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.11.0/examples">v2.11.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.10.0/examples">v2.10.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.9.1/examples">v2.9.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.8.0/examples">v2.8.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.7.0/examples">v2.7.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.6.0/examples">v2.6.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.5.1/examples">v2.5.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.4.0/examples">v2.4.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.3.0/examples">v2.3.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.2.0/examples">v2.2.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.1.0/examples">v2.1.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.0.0/examples">v2.0.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v1.2.0/examples">v1.2.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v1.1.0/examples">v1.1.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v1.0.0/examples">v1.0.0</a></li>
+	</ul>
+</details>
+
+Dann stellen Sie Ihren aktuellen Klon von 🤗 Transformers auf eine bestimmte Version um, z.B. v3.5.1:
+
+```bash
+git checkout tags/v3.5.1
+```
+
+Nachdem Sie die richtige Bibliotheksversion eingerichtet haben, navigieren Sie zu dem Beispielordner Ihrer Wahl und installieren die beispielspezifischen Anforderungen:
+
+```bash
+pip install -r requirements.txt
+```
+
+## Ein Skript ausführen
+
+<frameworkcontent>
+<pt>
+Das Beispielskript lädt einen Datensatz aus der 🤗 [Datasets](https://huggingface.co/docs/datasets/) Bibliothek herunter und verarbeitet ihn vor. Dann nimmt das Skript eine Feinabstimmung eines Datensatzes mit dem [Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) auf einer Architektur vor, die eine Zusammenfassung unterstützt. Das folgende Beispiel zeigt, wie die Feinabstimmung von [T5-small](https://huggingface.co/t5-small) auf dem Datensatz [CNN/DailyMail](https://huggingface.co/datasets/cnn_dailymail) durchgeführt wird. Das T5-Modell benötigt aufgrund der Art und Weise, wie es trainiert wurde, ein zusätzliches Argument `source_prefix`. Mit dieser Eingabeaufforderung weiß T5, dass es sich um eine Zusammenfassungsaufgabe handelt.
+
+```bash
+python examples/pytorch/summarization/run_summarization.py \
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+</pt>
+<tf>
+Das Beispielskript lädt einen Datensatz aus der 🤗 [Datasets](https://huggingface.co/docs/datasets/) Bibliothek herunter und verarbeitet ihn vor. Anschließend nimmt das Skript die Feinabstimmung eines Datensatzes mit Keras auf einer Architektur vor, die die Zusammenfassung unterstützt. Das folgende Beispiel zeigt, wie die Feinabstimmung von [T5-small](https://huggingface.co/t5-small) auf dem [CNN/DailyMail](https://huggingface.co/datasets/cnn_dailymail) Datensatz durchgeführt wird. Das T5-Modell benötigt aufgrund der Art und Weise, wie es trainiert wurde, ein zusätzliches Argument `source_prefix`. Mit dieser Eingabeaufforderung weiß T5, dass es sich um eine Zusammenfassungsaufgabe handelt.
+
+```bash
+python examples/tensorflow/summarization/run_summarization.py  \
+    --model_name_or_path t5-small \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --output_dir /tmp/tst-summarization  \
+    --per_device_train_batch_size 8 \
+    --per_device_eval_batch_size 16 \
+    --num_train_epochs 3 \
+    --do_train \
+    --do_eval
+```
+</tf>
+</frameworkcontent>
+
+## Verteiltes Training und gemischte Präzision
+
+Der [Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) unterstützt verteiltes Training und gemischte Präzision, d.h. Sie können ihn auch in einem Skript verwenden. So aktivieren Sie diese beiden Funktionen:
+
+- Fügen Sie das Argument `fp16` hinzu, um gemischte Genauigkeit zu aktivieren.
+- Legen Sie die Anzahl der zu verwendenden GPUs mit dem Argument `nproc_per_node` fest.
+
+```bash
+python -m torch.distributed.launch \
+    --nproc_per_node 8 pytorch/summarization/run_summarization.py \
+    --fp16 \
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+
+TensorFlow-Skripte verwenden eine [`MirroredStrategy`](https://www.tensorflow.org/guide/distributed_training#mirroredstrategy) für verteiltes Training, und Sie müssen dem Trainingsskript keine zusätzlichen Argumente hinzufügen. Das TensorFlow-Skript verwendet standardmäßig mehrere GPUs, wenn diese verfügbar sind.
+
+## Ein Skript auf einer TPU ausführen
+
+<frameworkcontent>
+<pt>
+Tensor Processing Units (TPUs) sind speziell für die Beschleunigung der Leistung konzipiert. PyTorch unterstützt TPUs mit dem [XLA](https://www.tensorflow.org/xla) Deep Learning Compiler (siehe [hier](https://github.com/pytorch/xla/blob/master/README.md) für weitere Details). Um eine TPU zu verwenden, starten Sie das Skript `xla_spawn.py` und verwenden das Argument `num_cores`, um die Anzahl der TPU-Kerne festzulegen, die Sie verwenden möchten.
+
+```bash
+python xla_spawn.py --num_cores 8 \
+    summarization/run_summarization.py \
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+</pt>
+<tf>
+Tensor Processing Units (TPUs) sind speziell für die Beschleunigung der Leistung konzipiert. TensorFlow Skripte verwenden eine [`TPUStrategy`](https://www.tensorflow.org/guide/distributed_training#tpustrategy) für das Training auf TPUs. Um eine TPU zu verwenden, übergeben Sie den Namen der TPU-Ressource an das Argument `tpu`.
+
+```bash
+python run_summarization.py  \
+    --tpu name_of_tpu_resource \
+    --model_name_or_path t5-small \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --output_dir /tmp/tst-summarization  \
+    --per_device_train_batch_size 8 \
+    --per_device_eval_batch_size 16 \
+    --num_train_epochs 3 \
+    --do_train \
+    --do_eval
+```
+</tf>
+</frameworkcontent>
+
+## Führen Sie ein Skript mit 🤗 Accelerate aus.
+
+🤗 [Accelerate](https://huggingface.co/docs/accelerate) ist eine reine PyTorch-Bibliothek, die eine einheitliche Methode für das Training eines Modells auf verschiedenen Arten von Setups (nur CPU, mehrere GPUs, TPUs) bietet und dabei die vollständige Transparenz der PyTorch-Trainingsschleife beibehält. Stellen Sie sicher, dass Sie 🤗 Accelerate installiert haben, wenn Sie es nicht bereits haben:
+
+> Hinweis: Da Accelerate schnell weiterentwickelt wird, muss die Git-Version von Accelerate installiert sein, um die Skripte auszuführen.
+```bash
+pip install git+https://github.com/huggingface/accelerate
+```
+
+Anstelle des Skripts `run_summarization.py` müssen Sie das Skript `run_summarization_no_trainer.py` verwenden. Die von Accelerate unterstützten Skripte haben eine Datei `task_no_trainer.py` im Ordner. Beginnen Sie mit dem folgenden Befehl, um eine Konfigurationsdatei zu erstellen und zu speichern:
+
+```bash
+accelerate config
+```
+
+Testen Sie Ihre Einrichtung, um sicherzustellen, dass sie korrekt konfiguriert ist:
+
+```bash
+accelerate test
+```
+
+Jetzt sind Sie bereit, das Training zu starten:
+
+```bash
+accelerate launch run_summarization_no_trainer.py \
+    --model_name_or_path t5-small \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir ~/tmp/tst-summarization
+```
+
+## Verwenden Sie einen benutzerdefinierten Datensatz
+
+Das Verdichtungsskript unterstützt benutzerdefinierte Datensätze, solange es sich um eine CSV- oder JSON-Line-Datei handelt. Wenn Sie Ihren eigenen Datensatz verwenden, müssen Sie mehrere zusätzliche Argumente angeben:
+
+- `train_file` und `validation_file` geben den Pfad zu Ihren Trainings- und Validierungsdateien an.
+- text_column` ist der Eingabetext, der zusammengefasst werden soll.
+- Summary_column" ist der auszugebende Zieltext.
+
+Ein Zusammenfassungsskript, das einen benutzerdefinierten Datensatz verwendet, würde wie folgt aussehen:
+
+```bash
+python examples/pytorch/summarization/run_summarization.py \
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --train_file path_to_csv_or_jsonlines_file \
+    --validation_file path_to_csv_or_jsonlines_file \
+    --text_column text_column_name \
+    --summary_column summary_column_name \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --overwrite_output_dir \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --predict_with_generate
+```
+
+## Testen Sie ein Skript
+
+Es ist oft eine gute Idee, Ihr Skript an einer kleineren Anzahl von Beispielen für Datensätze auszuführen, um sicherzustellen, dass alles wie erwartet funktioniert, bevor Sie sich auf einen ganzen Datensatz festlegen, dessen Fertigstellung Stunden dauern kann. Verwenden Sie die folgenden Argumente, um den Datensatz auf eine maximale Anzahl von Stichproben zu beschränken:
+
+- `max_train_samples`
+- `max_eval_samples`
+- `max_predict_samples`
+
+```bash
+python examples/pytorch/summarization/run_summarization.py \
+    --model_name_or_path t5-small \
+    --max_train_samples 50 \
+    --max_eval_samples 50 \
+    --max_predict_samples 50 \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+
+Nicht alle Beispielskripte unterstützen das Argument `max_predict_samples`. Wenn Sie sich nicht sicher sind, ob Ihr Skript dieses Argument unterstützt, fügen Sie das Argument `-h` hinzu, um dies zu überprüfen:
+
+```bash
+examples/pytorch/summarization/run_summarization.py -h
+```
+
+## Training vom Kontrollpunkt fortsetzen
+
+Eine weitere hilfreiche Option, die Sie aktivieren können, ist die Wiederaufnahme des Trainings von einem früheren Kontrollpunkt aus. Auf diese Weise können Sie im Falle einer Unterbrechung Ihres Trainings dort weitermachen, wo Sie aufgehört haben, ohne von vorne beginnen zu müssen. Es gibt zwei Methoden, um das Training von einem Kontrollpunkt aus wieder aufzunehmen.
+
+Die erste Methode verwendet das Argument `output_dir previous_output_dir`, um das Training ab dem letzten in `output_dir` gespeicherten Kontrollpunkt wieder aufzunehmen. In diesem Fall sollten Sie `overwrite_output_dir` entfernen:
+
+```bash
+python examples/pytorch/summarization/run_summarization.py
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --output_dir previous_output_dir \
+    --predict_with_generate
+```
+
+Die zweite Methode verwendet das Argument `Resume_from_checkpoint path_to_specific_checkpoint`, um das Training ab einem bestimmten Checkpoint-Ordner wieder aufzunehmen.
+
+```bash
+python examples/pytorch/summarization/run_summarization.py
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --resume_from_checkpoint path_to_specific_checkpoint \
+    --predict_with_generate
+```
+
+## Teilen Sie Ihr Modell
+
+Alle Skripte können Ihr endgültiges Modell in den [Model Hub](https://huggingface.co/models) hochladen. Stellen Sie sicher, dass Sie bei Hugging Face angemeldet sind, bevor Sie beginnen:
+
+```bash
+huggingface-cli login
+```
+
+Dann fügen Sie dem Skript das Argument `push_to_hub` hinzu. Mit diesem Argument wird ein Repository mit Ihrem Hugging Face-Benutzernamen und dem in `output_dir` angegebenen Ordnernamen erstellt.
+
+Wenn Sie Ihrem Repository einen bestimmten Namen geben möchten, fügen Sie ihn mit dem Argument `push_to_hub_model_id` hinzu. Das Repository wird automatisch unter Ihrem Namensraum aufgeführt.
+
+Das folgende Beispiel zeigt, wie Sie ein Modell mit einem bestimmten Repository-Namen hochladen können:
+
+```bash
+python examples/pytorch/summarization/run_summarization.py
+    --model_name_or_path t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --push_to_hub \
+    --push_to_hub_model_id finetuned-t5-cnn_dailymail \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```

docs/source/de/transformers_agents.md

@@ -0,0 +1,323 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Transformers Agents
+
+<Tip warning={true}>
+
+Transformers Agents ist eine experimentelle API, die jederzeit geändert werden kann. Die von den Agenten zurückgegebenen Ergebnisse
+zurückgegeben werden, können variieren, da sich die APIs oder die zugrunde liegenden Modelle ändern können.
+
+</Tip>
+
+Transformers Version v4.29.0, die auf dem Konzept von *Tools* und *Agenten* aufbaut. Sie können damit spielen in
+[dieses Colab](https://colab.research.google.com/drive/1c7MHD-T1forUPGcC_jlwsIptOzpG3hSj).
+
+Kurz gesagt, es bietet eine API für natürliche Sprache auf der Grundlage von Transformers: Wir definieren eine Reihe von kuratierten Tools und entwerfen einen 
+Agenten, um natürliche Sprache zu interpretieren und diese Werkzeuge zu verwenden. Es ist von vornherein erweiterbar; wir haben einige relevante Tools kuratiert, 
+aber wir werden Ihnen zeigen, wie das System einfach erweitert werden kann, um jedes von der Community entwickelte Tool zu verwenden.
+
+Beginnen wir mit einigen Beispielen dafür, was mit dieser neuen API erreicht werden kann. Sie ist besonders leistungsfähig, wenn es um 
+Sie ist besonders leistungsstark, wenn es um multimodale Aufgaben geht. Lassen Sie uns also eine Runde drehen, um Bilder zu erzeugen und Text vorzulesen.
+
+```py
+agent.run("Caption the following image", image=image)
+```
+
+| **Input**                                                                                                                   | **Output**                        |
+|-----------------------------------------------------------------------------------------------------------------------------|-----------------------------------|
+| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/beaver.png" width=200> | A beaver is swimming in the water |
+
+---
+
+```py
+agent.run("Read the following text out loud", text=text)
+```
+| **Input**                                                                                                               | **Output**                                   |
+|-------------------------------------------------------------------------------------------------------------------------|----------------------------------------------|
+| A beaver is swimming in the water | <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tts_example.wav" type="audio/wav"> your browser does not support the audio element. </audio>
+
+---
+
+```py
+agent.run(
+    "In the following `document`, where will the TRRF Scientific Advisory Council Meeting take place?",
+    document=document,
+)
+```
+| **Input**                                                                                                                   | **Output**     |
+|-----------------------------------------------------------------------------------------------------------------------------|----------------|
+| <img src="https://datasets-server.huggingface.co/assets/hf-internal-testing/example-documents/--/hf-internal-testing--example-documents/test/0/image/image.jpg" width=200> | ballroom foyer |
+
+## Schnellstart
+
+Bevor Sie `agent.run` verwenden können, müssen Sie einen Agenten instanziieren, der ein großes Sprachmodell (LLM) ist. 
+Wir bieten Unterstützung für openAI-Modelle sowie für OpenSource-Alternativen von BigCode und OpenAssistant. Die openAI
+Modelle sind leistungsfähiger (erfordern aber einen openAI-API-Schlüssel, können also nicht kostenlos verwendet werden); Hugging Face
+bietet kostenlosen Zugang zu Endpunkten für BigCode- und OpenAssistant-Modelle.
+
+To start with, please install the `agents` extras in order to install all default dependencies.
+```bash
+pip install transformers[agents]
+```
+
+Um openAI-Modelle zu verwenden, instanziieren Sie einen [`OpenAiAgent`], nachdem Sie die `openai`-Abhängigkeit installiert haben:
+
+```bash
+pip install openai
+```
+
+
+```py
+from transformers import OpenAiAgent
+
+agent = OpenAiAgent(model="text-davinci-003", api_key="<your_api_key>")
+```
+
+Um BigCode oder OpenAssistant zu verwenden, melden Sie sich zunächst an, um Zugriff auf die Inference API zu erhalten:
+
+```py
+from huggingface_hub import login
+
+login("<YOUR_TOKEN>")
+```
+
+Dann instanziieren Sie den Agenten
+
+```py
+from transformers import HfAgent
+
+# Starcoder
+agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
+# StarcoderBase
+# agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoderbase")
+# OpenAssistant
+# agent = HfAgent(url_endpoint="https://api-inference.huggingface.co/models/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5")
+```
+
+Dies geschieht mit der Inferenz-API, die Hugging Face derzeit kostenlos zur Verfügung stellt. Wenn Sie Ihren eigenen Inferenz
+Endpunkt für dieses Modell (oder einen anderen) haben, können Sie die obige URL durch Ihren URL-Endpunkt ersetzen.
+
+<Tip>
+
+StarCoder und OpenAssistant sind kostenlos und leisten bei einfachen Aufgaben bewundernswert gute Arbeit. Allerdings halten die Kontrollpunkte
+nicht, wenn es um komplexere Aufforderungen geht. Wenn Sie mit einem solchen Problem konfrontiert sind, empfehlen wir Ihnen, das OpenAI
+Modell auszuprobieren, das zwar leider nicht quelloffen ist, aber zur Zeit eine bessere Leistung erbringt.
+
+</Tip>
+
+Sie sind jetzt startklar! Lassen Sie uns in die beiden APIs eintauchen, die Ihnen jetzt zur Verfügung stehen.
+
+### Einzelne Ausführung (run)
+
+Die Methode der einmaligen Ausführung ist die Verwendung der [`~Agent.run`] Methode des Agenten:
+
+```py
+agent.run("Draw me a picture of rivers and lakes.")
+```
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200>
+
+Es wählt automatisch das (oder die) Werkzeug(e) aus, das (die) für die von Ihnen gewünschte Aufgabe geeignet ist (sind) und führt es (sie) entsprechend aus. Es
+kann eine oder mehrere Aufgaben in der gleichen Anweisung ausführen (je komplexer Ihre Anweisung ist, desto wahrscheinlicher ist ein
+der Agent scheitern).
+
+```py
+agent.run("Draw me a picture of the sea then transform the picture to add an island")
+```
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/sea_and_island.png" width=200>
+
+<br/>
+
+
+Jede [`~Agent.run`] Operation ist unabhängig, so dass Sie sie mehrmals hintereinander mit unterschiedlichen Aufgaben ausführen können.
+
+Beachten Sie, dass Ihr `Agent` nur ein großsprachiges Modell ist, so dass kleine Variationen in Ihrer Eingabeaufforderung völlig unterschiedliche Ergebnisse liefern können.
+unterschiedliche Ergebnisse liefern. Es ist wichtig, dass Sie die Aufgabe, die Sie ausführen möchten, so genau wie möglich erklären. Wir gehen noch weiter ins Detail
+wie man gute Prompts schreibt [hier](custom_tools#writing-good-user-inputs).
+
+Wenn Sie einen Status über Ausführungszeiten hinweg beibehalten oder dem Agenten Nicht-Text-Objekte übergeben möchten, können Sie dies tun, indem Sie
+Variablen, die der Agent verwenden soll. Sie könnten zum Beispiel das erste Bild von Flüssen und Seen erzeugen, 
+und das Modell bitten, dieses Bild zu aktualisieren und eine Insel hinzuzufügen, indem Sie Folgendes tun:
+
+```python
+picture = agent.run("Generate a picture of rivers and lakes.")
+updated_picture = agent.run("Transform the image in `picture` to add an island to it.", picture=picture)
+```
+
+<Tip>
+
+Dies kann hilfreich sein, wenn das Modell Ihre Anfrage nicht verstehen kann und die Werkzeuge verwechselt. Ein Beispiel wäre:
+
+```py
+agent.run("Draw me the picture of a capybara swimming in the sea")
+```
+
+Hier könnte das Modell auf zwei Arten interpretieren:
+- Die Funktion `Text-zu-Bild` erzeugt ein Wasserschwein, das im Meer schwimmt.
+- Oder Sie lassen das `Text-zu-Bild` ein Wasserschwein erzeugen und verwenden dann das Werkzeug `Bildtransformation`, um es im Meer schwimmen zu lassen.
+
+Falls Sie das erste Szenario erzwingen möchten, können Sie dies tun, indem Sie die Eingabeaufforderung als Argument übergeben:
+
+```py
+agent.run("Draw me a picture of the `prompt`", prompt="a capybara swimming in the sea")
+```
+
+</Tip>
+
+
+### Chat-basierte Ausführung (Chat)
+
+Der Agent verfügt auch über einen Chat-basierten Ansatz, der die Methode [`~Agent.chat`] verwendet:
+
+```py
+agent.chat("Generate a picture of rivers and lakes")
+```
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200> 
+
+```py
+agent.chat("Transform the picture so that there is a rock in there")
+```
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_and_beaver.png" width=200>
+
+<br/>
+
+Dies ist ein interessanter Ansatz, wenn Sie den Zustand über Anweisungen hinweg beibehalten möchten. Er ist besser für Experimente geeignet, 
+eignet sich aber eher für einzelne Anweisungen als für komplexe Anweisungen (die die [`~Agent.run`]
+Methode besser verarbeiten kann).
+
+Diese Methode kann auch Argumente entgegennehmen, wenn Sie Nicht-Text-Typen oder bestimmte Aufforderungen übergeben möchten.
+
+### ⚠️ Fernausführung
+
+Zu Demonstrationszwecken und damit es mit allen Setups verwendet werden kann, haben wir Remote-Executors für mehrere 
+der Standard-Tools erstellt, auf die der Agent in dieser Version Zugriff hat. Diese werden erstellt mit 
+[inference endpoints](https://huggingface.co/inference-endpoints).
+
+Wir haben diese vorerst deaktiviert, aber um zu sehen, wie Sie selbst Remote Executors Tools einrichten können,
+empfehlen wir die Lektüre des [custom tool guide](./custom_tools).
+
+### Was passiert hier? Was sind Tools und was sind Agenten?
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/diagram.png">
+
+#### Agenten
+
+Der "Agent" ist hier ein großes Sprachmodell, das wir auffordern, Zugang zu einem bestimmten Satz von Tools zu erhalten.
+
+LLMs sind ziemlich gut darin, kleine Codeproben zu erzeugen. Diese API macht sich das zunutze, indem sie das 
+LLM ein kleines Codebeispiel gibt, das eine Aufgabe mit einer Reihe von Werkzeugen ausführt. Diese Aufforderung wird dann ergänzt durch die 
+Aufgabe, die Sie Ihrem Agenten geben, und die Beschreibung der Werkzeuge, die Sie ihm geben. Auf diese Weise erhält er Zugriff auf die Dokumentation der 
+Tools, insbesondere die erwarteten Eingaben und Ausgaben, und kann den entsprechenden Code generieren.
+
+#### Tools
+
+Tools sind sehr einfach: Sie bestehen aus einer einzigen Funktion mit einem Namen und einer Beschreibung. Wir verwenden dann die Beschreibungen dieser Tools 
+um den Agenten aufzufordern. Anhand der Eingabeaufforderung zeigen wir dem Agenten, wie er die Tools nutzen kann, um das zu tun, was in der 
+in der Abfrage angefordert wurde.
+
+Dies geschieht mit brandneuen Tools und nicht mit Pipelines, denn der Agent schreibt besseren Code mit sehr atomaren Tools. 
+Pipelines sind stärker refaktorisiert und fassen oft mehrere Aufgaben in einer einzigen zusammen. Tools sind dafür gedacht, sich auf
+eine einzige, sehr einfache Aufgabe konzentrieren.
+
+#### Code-Ausführung?!
+
+Dieser Code wird dann mit unserem kleinen Python-Interpreter auf den mit Ihren Tools übergebenen Eingaben ausgeführt. 
+Wir hören Sie schon schreien "Willkürliche Codeausführung!", aber lassen Sie uns erklären, warum das nicht der Fall ist.
+
+Die einzigen Funktionen, die aufgerufen werden können, sind die von Ihnen zur Verfügung gestellten Tools und die Druckfunktion, so dass Sie bereits eingeschränkt sind 
+eingeschränkt, was ausgeführt werden kann. Sie sollten sicher sein, wenn es sich auf die Werkzeuge für das Umarmungsgesicht beschränkt. 
+
+Dann lassen wir keine Attributsuche oder Importe zu (die ohnehin nicht benötigt werden, um die 
+Inputs/Outputs an eine kleine Gruppe von Funktionen), so dass alle offensichtlichen Angriffe (und Sie müssten den LLM 
+dazu auffordern, sie auszugeben) kein Problem darstellen sollten. Wenn Sie auf Nummer sicher gehen wollen, können Sie die 
+run()-Methode mit dem zusätzlichen Argument return_code=True ausführen. In diesem Fall gibt der Agent nur den auszuführenden Code 
+zur Ausführung zurück und Sie können entscheiden, ob Sie ihn ausführen möchten oder nicht.
+
+Die Ausführung bricht bei jeder Zeile ab, in der versucht wird, eine illegale Operation auszuführen, oder wenn ein regulärer Python-Fehler 
+mit dem vom Agenten generierten Code.
+
+### Ein kuratierter Satz von Tools
+
+Wir haben eine Reihe von Tools identifiziert, die solche Agenten unterstützen können. Hier ist eine aktualisierte Liste der Tools, die wir integriert haben 
+in `transformers` integriert haben:
+
+- **Beantwortung von Fragen zu Dokumenten**: Beantworten Sie anhand eines Dokuments (z.B. PDF) im Bildformat eine Frage zu diesem Dokument ([Donut](./model_doc/donut))
+- Beantworten von Textfragen**: Geben Sie einen langen Text und eine Frage an, beantworten Sie die Frage im Text ([Flan-T5](./model_doc/flan-t5))
+- **Unbedingte Bildunterschriften**: Beschriften Sie das Bild! ([BLIP](./model_doc/blip))
+- **Bildfragebeantwortung**: Beantworten Sie bei einem Bild eine Frage zu diesem Bild ([VILT](./model_doc/vilt))
+- **Bildsegmentierung**: Geben Sie ein Bild und einen Prompt an und geben Sie die Segmentierungsmaske dieses Prompts aus ([CLIPSeg](./model_doc/clipseg))
+- **Sprache in Text**: Geben Sie eine Audioaufnahme einer sprechenden Person an und transkribieren Sie die Sprache in Text ([Whisper](./model_doc/whisper))
+- **Text in Sprache**: wandelt Text in Sprache um ([SpeechT5](./model_doc/speecht5))
+- **Zero-Shot-Textklassifizierung**: Ermitteln Sie anhand eines Textes und einer Liste von Bezeichnungen, welcher Bezeichnung der Text am ehesten entspricht ([BART](./model_doc/bart))
+- **Textzusammenfassung**: fassen Sie einen langen Text in einem oder wenigen Sätzen zusammen ([BART](./model_doc/bart))
+- **Übersetzung**: Übersetzen des Textes in eine bestimmte Sprache ([NLLB](./model_doc/nllb))
+
+Diese Tools sind in Transformatoren integriert und können auch manuell verwendet werden, zum Beispiel:
+
+```py
+from transformers import load_tool
+
+tool = load_tool("text-to-speech")
+audio = tool("This is a text to speech tool")
+```
+
+### Benutzerdefinierte Tools
+
+Wir haben zwar eine Reihe von Tools identifiziert, sind aber der festen Überzeugung, dass der Hauptwert dieser Implementierung darin besteht 
+die Möglichkeit, benutzerdefinierte Tools schnell zu erstellen und weiterzugeben.
+
+Indem Sie den Code eines Tools in einen Hugging Face Space oder ein Modell-Repository stellen, können Sie das Tool 
+direkt mit dem Agenten nutzen. Wir haben ein paar neue Funktionen hinzugefügt 
+**transformers-agnostic** Tools zur [`huggingface-tools` Organisation](https://huggingface.co/huggingface-tools) hinzugefügt:
+
+- **Text-Downloader**: zum Herunterladen eines Textes von einer Web-URL
+- **Text zu Bild**: erzeugt ein Bild nach einer Eingabeaufforderung und nutzt dabei stabile Diffusion
+- **Bildtransformation**: verändert ein Bild anhand eines Ausgangsbildes und einer Eingabeaufforderung, unter Ausnutzung der stabilen pix2pix-Diffusion
+- **Text zu Video**: Erzeugen eines kleinen Videos nach einer Eingabeaufforderung, unter Verwendung von damo-vilab
+
+Das Text-zu-Bild-Tool, das wir von Anfang an verwendet haben, ist ein Remote-Tool, das sich in 
+[*huggingface-tools/text-to-image*](https://huggingface.co/spaces/huggingface-tools/text-to-image)! Wir werden
+weiterhin solche Tools für diese und andere Organisationen veröffentlichen, um diese Implementierung weiter zu verbessern.
+
+Die Agenten haben standardmäßig Zugriff auf die Tools, die sich auf [*huggingface-tools*](https://huggingface.co/huggingface-tools) befinden.
+Wie Sie Ihre eigenen Tools schreiben und freigeben können und wie Sie jedes benutzerdefinierte Tool, das sich auf dem Hub befindet, nutzen können, erklären wir in [folgender Anleitung](custom_tools).
+
+### Code-Erzeugung
+
+Bisher haben wir gezeigt, wie Sie die Agenten nutzen können, um Aktionen für Sie durchzuführen. Der Agent generiert jedoch nur Code
+den wir dann mit einem sehr eingeschränkten Python-Interpreter ausführen. Falls Sie den generierten Code in einer anderen Umgebung verwenden möchten 
+einer anderen Umgebung verwenden möchten, können Sie den Agenten auffordern, den Code zusammen mit einer Tooldefinition und genauen Importen zurückzugeben.
+
+Zum Beispiel die folgende Anweisung
+```python
+agent.run("Draw me a picture of rivers and lakes", return_code=True)
+```
+
+gibt den folgenden Code zurück
+
+```python
+from transformers import load_tool
+
+image_generator = load_tool("huggingface-tools/text-to-image")
+
+image = image_generator(prompt="rivers and lakes")
+```
+
+die Sie dann selbst ändern und ausführen können.

docs/source/en/_toctree.yml

@@ -71,6 +71,10 @@
       title: Zero-shot image classification
     - local: tasks/monocular_depth_estimation
       title: Depth estimation
+    - local: tasks/image_to_image
+      title: Image-to-Image
+    - local: tasks/knowledge_distillation_for_image_classification
+      title: Knowledge Distillation for Computer Vision
     title: Computer Vision
   - isExpanded: false
     sections:
@@ -88,6 +92,13 @@
     - local: generation_strategies
       title: Customize the generation strategy
     title: Generation
+  - isExpanded: false
+    sections:
+    - local: tasks/idefics
+      title: Image tasks with IDEFICS
+    - local: tasks/prompting
+      title: LLM prompting guide
+    title: Prompting
   title: Task Guides
 - sections:
   - local: fast_tokenizers
@@ -98,6 +109,8 @@
     title: Use model-specific APIs
   - local: custom_models
     title: Share a custom model
+  - local: chat_templating
+    title: Templates for chat models
   - local: sagemaker
     title: Run training on Amazon SageMaker
   - local: serialization
@@ -198,6 +211,8 @@
     title: Pipelines for webserver inference
   - local: model_memory_anatomy
     title: Model training anatomy
+  - local: llm_tutorial_optimization
+    title: Getting the most out of LLMs
   title: Conceptual guides
 - sections:
   - sections:
@@ -313,6 +328,8 @@
         title: ErnieM
       - local: model_doc/esm
         title: ESM
+      - local: model_doc/falcon
+        title: Falcon
       - local: model_doc/flan-t5
         title: FLAN-T5
       - local: model_doc/flan-ul2
@@ -325,6 +342,8 @@
         title: FSMT
       - local: model_doc/funnel
         title: Funnel Transformer
+      - local: model_doc/fuyu
+        title: Fuyu
       - local: model_doc/openai-gpt
         title: GPT
       - local: model_doc/gpt_neo
@@ -375,6 +394,8 @@
         title: MegatronBERT
       - local: model_doc/megatron_gpt2
         title: MegatronGPT2
+      - local: model_doc/mistral
+        title: Mistral
       - local: model_doc/mluke
         title: mLUKE
       - local: model_doc/mobilebert
@@ -405,6 +426,8 @@
         title: Pegasus
       - local: model_doc/pegasus_x
         title: PEGASUS-X
+      - local: model_doc/persimmon
+        title: Persimmon
       - local: model_doc/phobert
         title: PhoBERT
       - local: model_doc/plbart
@@ -495,7 +518,7 @@
       - local: model_doc/dinat
         title: DiNAT
       - local: model_doc/dinov2
-        title: DINO V2
+        title: DINOV2
       - local: model_doc/dit
         title: DiT
       - local: model_doc/dpt
@@ -562,6 +585,8 @@
         title: ViTDet
       - local: model_doc/vit_mae
         title: ViTMAE
+      - local: model_doc/vitmatte
+        title: ViTMatte
       - local: model_doc/vit_msn
         title: ViTMSN
       - local: model_doc/vivit
@@ -603,6 +628,8 @@
         title: UniSpeech
       - local: model_doc/unispeech-sat
         title: UniSpeech-SAT
+      - local: model_doc/vits
+        title: VITS
       - local: model_doc/wav2vec2
         title: Wav2Vec2
       - local: model_doc/wav2vec2-conformer
@@ -630,6 +657,8 @@
         title: BLIP-2
       - local: model_doc/bridgetower
         title: BridgeTower
+      - local: model_doc/bros
+        title: BROS
       - local: model_doc/chinese_clip
         title: Chinese-CLIP
       - local: model_doc/clip
@@ -668,10 +697,14 @@
         title: MatCha
       - local: model_doc/mgp-str
         title: MGP-STR
+      - local: model_doc/nougat
+        title: Nougat
       - local: model_doc/oneformer
         title: OneFormer
       - local: model_doc/owlvit
         title: OWL-ViT
+      - local: model_doc/owlv2
+        title: OWLv2
       - local: model_doc/perceiver
         title: Perceiver
       - local: model_doc/pix2struct

docs/source/en/accelerate.md

@@ -133,4 +133,4 @@ accelerate launch train.py
 >>> notebook_launcher(training_function)
 ```
 
-For more information about 🤗 Accelerate and it's rich features, refer to the [documentation](https://huggingface.co/docs/accelerate).
+For more information about 🤗 Accelerate and its rich features, refer to the [documentation](https://huggingface.co/docs/accelerate).

docs/source/en/add_new_model.md

@@ -52,7 +52,7 @@ A good first starting point to better understand the library is to read the [doc
 
 In our opinion, the library's code is not just a means to provide a product, *e.g.* the ability to use BERT for
 inference, but also as the very product that we want to improve. Hence, when adding a model, the user is not only the
-person that will use your model, but also everybody that will read, try to understand, and possibly tweak your code.
+person who will use your model, but also everybody who will read, try to understand, and possibly tweak your code.
 
 With this in mind, let's go a bit deeper into the general library design.
 
@@ -131,9 +131,9 @@ From experience, we can tell you that the most important things to keep in mind
   friends. Note that it might very well happen that your model's tokenizer is based on one model implementation, and
   your model's modeling code on another one. *E.g.* FSMT's modeling code is based on BART, while FSMT's tokenizer code
   is based on XLM.
--  It's more of an engineering challenge than a scientific challenge. You should spend more time on creating an
-  efficient debugging environment than trying to understand all theoretical aspects of the model in the paper.
--  Ask for help, when you're stuck! Models are the core component of 🤗 Transformers so that we at Hugging Face are more
+-  It's more of an engineering challenge than a scientific challenge. You should spend more time creating an
+  efficient debugging environment rather than trying to understand all theoretical aspects of the model in the paper.
+-  Ask for help, when you're stuck! Models are the core component of 🤗 Transformers so we at Hugging Face are more
   than happy to help you at every step to add your model. Don't hesitate to ask if you notice you are not making
   progress.
 
@@ -157,9 +157,9 @@ List:
 ☐ Submitted the pull request<br>
 ☐ (Optional) Added a demo notebook
 
-To begin with, we usually recommend to start by getting a good theoretical understanding of `BrandNewBert`. However,
+To begin with, we usually recommend starting by getting a good theoretical understanding of `BrandNewBert`. However,
 if you prefer to understand the theoretical aspects of the model *on-the-job*, then it is totally fine to directly dive
-into the `BrandNewBert`'s code-base. This option might suit you better, if your engineering skills are better than
+into the `BrandNewBert`'s code-base. This option might suit you better if your engineering skills are better than
 your theoretical skill, if you have trouble understanding `BrandNewBert`'s paper, or if you just enjoy programming
 much more than reading scientific papers.
 
@@ -175,7 +175,7 @@ theoretical aspects, but rather focus on the practical ones, namely:
   encoder-decoder model? Look at the [model_summary](model_summary) if you're not familiar with the differences between those.
 -  What are the applications of *brand_new_bert*? Text classification? Text generation? Seq2Seq tasks, *e.g.,*
   summarization?
--  What is the novel feature of the model making it different from BERT/GPT-2/BART?
+-  What is the novel feature of the model that makes it different from BERT/GPT-2/BART?
 -  Which of the already existing [🤗 Transformers models](https://huggingface.co/transformers/#contents) is most
   similar to *brand_new_bert*?
 -  What type of tokenizer is used? A sentencepiece tokenizer? Word piece tokenizer? Is it the same tokenizer as used
@@ -261,7 +261,7 @@ figure out the following:
 - How can you debug the model in the original environment of the repo? Do you have to add *print* statements, can you
   work with an interactive debugger like *ipdb*, or should you use an efficient IDE to debug the model, like PyCharm?
 
-It is very important that before you start the porting process, that you can **efficiently** debug code in the original
+It is very important that before you start the porting process, you can **efficiently** debug code in the original
 repository! Also, remember that you are working with an open-source library, so do not hesitate to open an issue, or
 even a pull request in the original repository. The maintainers of this repository are most likely very happy about
 someone looking into their code!
@@ -280,10 +280,10 @@ In general, there are two possible debugging environments for running the origin
 Jupyter notebooks have the advantage that they allow for cell-by-cell execution which can be helpful to better split
 logical components from one another and to have faster debugging cycles as intermediate results can be stored. Also,
 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 Jupyter notebooks, we strongly recommend you to work with them.
+Face team for help. If you are familiar with Jupyter notebooks, we strongly recommend you work with them.
 
 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
+some time adjusting to the new programming environment and you might not be able to use your known debugging tools
 anymore, like `ipdb`.
 
 For each code-base, a good first step is always to load a **small** pretrained checkpoint and to be able to reproduce a
@@ -329,7 +329,7 @@ example is [T5's MeshTensorFlow](https://github.com/tensorflow/mesh/tree/master/
 very complex and does not offer a simple way to decompose the model into its sub-components. For such libraries, one
 often relies on verifying print statements.
 
-No matter which strategy you choose, the recommended procedure is often the same in that you should start to debug the
+No matter which strategy you choose, the recommended procedure is often the same that you should start to debug the
 starting layers first and the ending layers last.
 
 It is recommended that you retrieve the output, either by print statements or sub-component functions, of the following
@@ -361,10 +361,10 @@ We expect that every model added to 🤗 Transformers passes a couple of integra
 model and the reimplemented version in 🤗 Transformers have to give the exact same output up to a precision of 0.001!
 Since it is normal that the exact same model written in different libraries can give a slightly different output
 depending on the library framework, we accept an error tolerance of 1e-3 (0.001). It is not enough if the model gives
-nearly the same output, they have to be the almost identical. Therefore, you will certainly compare the intermediate
+nearly the same output, they have to be almost identical. Therefore, you will certainly compare the intermediate
 outputs of the 🤗 Transformers version multiple times against the intermediate outputs of the original implementation of
 *brand_new_bert* in which case an **efficient** debugging environment of the original repository is absolutely
-important. Here is some advice is to make your debugging environment as efficient as possible.
+important. Here is some advice to make your debugging environment as efficient as possible.
 
 - Find the best way of debugging intermediate results. Is the original repository written in PyTorch? Then you should
   probably take the time to write a longer script that decomposes the original model into smaller sub-components to
@@ -409,7 +409,7 @@ Otherwise, let's start generating a new model. You have two choices here:
 - `transformers-cli add-new-model-like` to add a new model like an existing one
 - `transformers-cli add-new-model` to add a new model from our template (will look like BERT or Bart depending on the type of model you select)
 
-In both cases, you will be prompted with a questionnaire to fill the basic information of your model. The second command requires to install `cookiecutter`, you can find more information on it [here](https://github.com/huggingface/transformers/tree/main/templates/adding_a_new_model).
+In both cases, you will be prompted with a questionnaire to fill in the basic information of your model. The second command requires to install `cookiecutter`, you can find more information on it [here](https://github.com/huggingface/transformers/tree/main/templates/adding_a_new_model).
 
 **Open a Pull Request on the main huggingface/transformers repo**
 
@@ -451,7 +451,7 @@ git push -u origin a-descriptive-name-for-my-changes
 
 6. Change the PR into a draft by clicking on “Convert to draft” on the right of the GitHub pull request web page.
 
-In the following, whenever you have done some progress, don't forget to commit your work and push it to your account so
+In the following, whenever you have made some progress, don't forget to commit your work and push it to your account so
 that it shows in the pull request. Additionally, you should make sure to update your work with the current main from
 time to time by doing:
 
@@ -483,7 +483,7 @@ Now you can finally start coding :). The generated code in
 `src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` will either have the same architecture as BERT if
 it's an encoder-only model or BART if it's an encoder-decoder model. At this point, you should remind yourself what
 you've learned in the beginning about the theoretical aspects of the model: *How is the model different from BERT or
-BART?*". Implement those changes which often means to change the *self-attention* layer, the order of the normalization
+BART?*". Implement those changes which often means changing the *self-attention* layer, the order of the normalization
 layer, etc… Again, it is often useful to look at the similar architecture of already existing models in Transformers to
 get a better feeling of how your model should be implemented.
 
@@ -665,7 +665,7 @@ PyTorch's implementation of a layer requires the weight to be transposed beforeh
 
 Finally, you should also check that **all** required weights are initialized and print out all checkpoint weights that
 were not used for initialization to make sure the model is correctly converted. It is completely normal, that the
-conversion trials fail with either a wrong shape statement or wrong name assignment. This is most likely because either
+conversion trials fail with either a wrong shape statement or a wrong name assignment. This is most likely because either
 you used incorrect parameters in `BrandNewBertConfig()`, have a wrong architecture in the 🤗 Transformers
 implementation, you have a bug in the `init()` functions of one of the components of the 🤗 Transformers
 implementation or you need to transpose one of the checkpoint weights.
@@ -722,7 +722,7 @@ in the 🤗 Transformers implementation. From our experience, a simple and effic
 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 presentations.
 
-When you're confident that both implementations yield the same output, verifying the outputs with
+When you're confident that both implementations yield the same output, verify the outputs with
 `torch.allclose(original_output, output, atol=1e-3)`, you're done with the most difficult part! Congratulations - the
 work left to be done should be a cakewalk 😊.
 
@@ -744,7 +744,7 @@ Having fixed all common tests, it is now crucial to ensure that all the nice wor
 - b) Future changes to your model will not break any important feature of the model.
 
 At first, integration tests should be added. Those integration tests essentially do the same as the debugging scripts
-you used earlier to implement the model to 🤗 Transformers. A template of those model tests is already added by the
+you used earlier to implement the model to 🤗 Transformers. A template of those model tests has already added by the
 Cookiecutter, called `BrandNewBertModelIntegrationTests` and only has to be filled out by you. To ensure that those
 tests are passing, run
 
@@ -769,7 +769,7 @@ ways:
 
 **9. Implement the tokenizer**
 
-Next, we should add the tokenizer of *brand_new_bert*. Usually, the tokenizer is equivalent or very similar to an
+Next, we should add the tokenizer of *brand_new_bert*. Usually, the tokenizer is equivalent to or very similar to an
 already existing tokenizer of 🤗 Transformers.
 
 It is very important to find/extract the original tokenizer file and to manage to load this file into the 🤗
@@ -890,6 +890,6 @@ reviewer.
 Now, it's time to get some credit from the community for your work! Having completed a model addition is a major
 contribution to Transformers and the whole NLP community. Your code and the ported pre-trained models will certainly be
 used by hundreds and possibly even thousands of developers and researchers. You should be proud of your work and share
-your achievement with the community.
+your achievements with the community.
 
 **You have made another model that is super easy to access for everyone in the community! 🤯**

docs/source/en/add_new_pipeline.md

@@ -111,8 +111,8 @@ def _sanitize_parameters(self, **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)
+without requiring users to understand new kinds of objects. It's also relatively common to support many different types
+of arguments for ease of use (audio files, which can be filenames, URLs or pure bytes)
 
 
 
@@ -219,8 +219,8 @@ repo.push_to_hub()
 ```
 
 This will copy the file where you defined `PairClassificationPipeline` inside the folder `"test-dynamic-pipeline"`,
-along with saving the model and tokenizer of the pipeline, before pushing everything in the repository
-`{your_username}/test-dynamic-pipeline`. After that anyone can use it as long as they provide the option
+along with saving the model and tokenizer of the pipeline, before pushing everything into the repository
+`{your_username}/test-dynamic-pipeline`. After that, anyone can use it as long as they provide the option
 `trust_remote_code=True`:
 
 ```py
@@ -232,9 +232,9 @@ classifier = pipeline(model="{your_username}/test-dynamic-pipeline", trust_remot
 ## Add the pipeline to 🤗 Transformers
 
 If you want to contribute your pipeline to 🤗 Transformers, you will need to add a new module in the `pipelines` submodule
-with the code of your pipeline, then add it in the list of tasks defined in `pipelines/__init__.py`.
+with the code of your pipeline, then add it to the list of tasks defined in `pipelines/__init__.py`.
 
-Then you will need to add tests. Create a new file `tests/test_pipelines_MY_PIPELINE.py` with example with the other tests.
+Then you will need to add tests. Create a new file `tests/test_pipelines_MY_PIPELINE.py` with examples of the other tests.
 
 The `run_pipeline_test` function will be very generic and run on small random models on every possible
 architecture as defined by `model_mapping` and `tf_model_mapping`.

docs/source/en/add_tensorflow_model.md

@@ -56,7 +56,7 @@ you might recall from our [general overview of 🤗 Transformers](add_new_model#
 that we are an opinionated bunch - the ease of use of 🤗 Transformers relies on consistent design choices. From
 experience, we can tell you a few important things about adding TensorFlow models:
 
-- Don't reinvent the wheel! More often that not, there are at least two reference implementations you should check: the
+- Don't reinvent the wheel! More often than not, there are at least two reference implementations you should check: the
 PyTorch equivalent of the model you are implementing and other TensorFlow models for the same class of problems.
 - Great model implementations survive the test of time. This doesn't happen because the code is pretty, but rather
 because the code is clear, easy to debug and build upon. If you make the life of the maintainers easy with your
@@ -101,7 +101,7 @@ TensorFlow-related pull request.
 
 **2. Prepare transformers dev environment**
 
-Having selected the model architecture, open an draft PR to signal your intention to work on it. Follow the
+Having selected the model architecture, open a draft PR to signal your intention to work on it. Follow the
 instructions below to set up your environment and open a draft PR.
 
 1. Fork the [repository](https://github.com/huggingface/transformers) by clicking on the 'Fork' button on the
@@ -229,7 +229,6 @@ documentation pages. You can complete this part entirely following the patterns
 changes:
 - Include all public classes of *BrandNewBert* in `src/transformers/__init__.py`
 - Add *BrandNewBert* classes to the corresponding Auto classes in `src/transformers/models/auto/modeling_tf_auto.py`
-- Include the modeling file in the documentation test file list in `utils/documentation_tests.txt`
 - Add the lazy loading classes related to *BrandNewBert* in `src/transformers/utils/dummy_tf_objects.py`
 - Update the import structures for the public classes in `src/transformers/models/brand_new_bert/__init__.py`
 - Add the documentation pointers to the public methods of *BrandNewBert* in `docs/source/en/model_doc/brand_new_bert.md`
@@ -328,7 +327,7 @@ That's it! 🎉
 ## Debugging mismatches across ML frameworks 🐛
 
 At some point, when adding a new architecture or when creating TensorFlow weights for an existing architecture, you
-might come across errors compaining about mismatches between PyTorch and TensorFlow. You might even decide to open the
+might come across errors complaining about mismatches between PyTorch and TensorFlow. You might even decide to open the
 model architecture code for the two frameworks, and find that they look identical. What's going on? 🤔
 
 First of all, let's talk about why understanding these mismatches matters. Many community members will use 🤗
@@ -351,7 +350,7 @@ ingredient here is patience. Here is our suggested workflow for when you come ac
    that you'll have to venture into the source implementation of said instruction. In some cases, you might find an
    issue with a reference implementation - don't abstain from opening an issue in the upstream repository.
 
-In some cases, in dicussion with the 🤗 Transformers team, we might find that the fixing the mismatch is infeasible.
+In some cases, in discussion with the 🤗 Transformers team, we might find that fixing the mismatch is infeasible.
 When the mismatch is very small in the output layers of the model (but potentially large in the hidden states), we
 might decide to ignore it in favor of distributing the model. The `pt-to-tf` CLI mentioned above has a `--max-error`
 flag to override the error message at weight conversion time.

docs/source/en/autoclass_tutorial.md

@@ -16,7 +16,7 @@ rendered properly in your Markdown viewer.
 
 # Load pretrained instances with an AutoClass
 
-With so many different Transformer architectures, it can be challenging to create one for your checkpoint. As a part of 🤗 Transformers core philosophy to make the library easy, simple and flexible to use, an `AutoClass` automatically infer and load the correct architecture from a given checkpoint. The `from_pretrained()` method lets you quickly load a pretrained model for any architecture so you don't have to devote time and resources to train a model from scratch. Producing this type of checkpoint-agnostic code means if your code works for one checkpoint, it will work with another checkpoint - as long as it was trained for a similar task - even if the architecture is different.
+With so many different Transformer architectures, it can be challenging to create one for your checkpoint. As a part of 🤗 Transformers core philosophy to make the library easy, simple and flexible to use, an `AutoClass` automatically infers and loads the correct architecture from a given checkpoint. The `from_pretrained()` method lets you quickly load a pretrained model for any architecture so you don't have to devote time and resources to train a model from scratch. Producing this type of checkpoint-agnostic code means if your code works for one checkpoint, it will work with another checkpoint - as long as it was trained for a similar task - even if the architecture is different.
 
 <Tip>
 

docs/source/en/big_models.md

@@ -23,11 +23,11 @@ from PyTorch is:
 2. Load your pretrained weights.
 3. Put those pretrained weights in your random model.
 
-Step 1 and 2 both require a full version of the model in memory, which is not a problem in most cases, but if your model starts weighing several GigaBytes, those two copies can make you got our of RAM. Even worse, if you are using `torch.distributed` to launch a distributed training, each process will load the pretrained model and store these two copies in RAM.
+Step 1 and 2 both require a full version of the model in memory, which is not a problem in most cases, but if your model starts weighing several GigaBytes, those two copies can make you get out of RAM. Even worse, if you are using `torch.distributed` to launch a distributed training, each process will load the pretrained model and store these two copies in RAM.
 
 <Tip>
 
-Note that the randomly created model is initialized with "empty" tensors, which take the space in memory without filling it (thus the random values are whatever was in this chunk of memory at a given time). The random initialization following the appropriate distribution for the kind of model/parameters instatiated (like a normal distribution for instance) is only performed after step 3 on the non-initialized weights, to be as fast as possible! 
+Note that the randomly created model is initialized with "empty" tensors, which take the space in memory without filling it (thus the random values are whatever was in this chunk of memory at a given time). The random initialization following the appropriate distribution for the kind of model/parameters instantiated (like a normal distribution for instance) is only performed after step 3 on the non-initialized weights, to be as fast as possible! 
 
 </Tip>
 
@@ -120,4 +120,4 @@ If you want to directly load such a sharded checkpoint inside a model without us
 
 Sharded checkpoints reduce the memory usage during step 2 of the workflow mentioned above, but in order to use that model in a low memory setting, we recommend leveraging our tools based on the Accelerate library.
 
-Please read the following guide for more information: [Large model loading using Accelerate](./main_classes/model#large-model-loading)
+Please read the following guide for more information: [Large model loading using Accelerate](./main_classes/model#large-model-loading)

docs/source/en/chat_templating.md

@@ -0,0 +1,367 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Templates for Chat Models
+
+## Introduction
+
+An increasingly common use case for LLMs is **chat**. In a chat context, rather than continuing a single string
+of text (as is the case with a standard language model), the model instead continues a conversation that consists
+of one or more **messages**, each of which includes a **role** as well as message text.
+
+Most commonly, these roles are "user" for messages sent by the user, and "assistant" for messages sent by the model.
+Some models also support a "system" role. System messages are usually sent at the beginning of the conversation
+and include directives about how the model should behave in the subsequent chat.
+
+All language models, including models fine-tuned for chat, operate on linear sequences of tokens and do not intrinsically
+have special handling for roles. This means that role information is usually injected by adding control tokens
+between messages, to indicate both the message boundary and the relevant roles.
+
+Unfortunately, there isn't (yet!) a standard for which tokens to use, and so different models have been trained
+with wildly different formatting and control tokens for chat. This can be a real problem for users - if you use the
+wrong format, then the model will be confused by your input, and your performance will be a lot worse than it should be.
+This is the problem that **chat templates** aim to resolve. 
+
+Chat conversations are typically represented as a list of dictionaries, where each dictionary contains `role`
+and `content` keys, and represents a single chat message. Chat templates are strings containing a Jinja template that
+specifies how to format a conversation for a given model into a single tokenizable sequence. By storing this information
+with the tokenizer, we can ensure that models get input data in the format they expect.
+
+Let's make this concrete with a quick example using the `BlenderBot` model. BlenderBot has an extremely simple default 
+template, which mostly just adds whitespace between rounds of dialogue:
+
+```python
+>>> from transformers import AutoTokenizer
+>>> tokenizer = AutoTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
+
+>>> chat = [
+...   {"role": "user", "content": "Hello, how are you?"},
+...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
+...   {"role": "user", "content": "I'd like to show off how chat templating works!"},
+... ]
+
+>>> tokenizer.apply_chat_template(chat, tokenize=False)
+" Hello, how are you?  I'm doing great. How can I help you today?   I'd like to show off how chat templating works!</s>"
+```
+
+Notice how the entire chat is condensed into a single string. If we use `tokenize=True`, which is the default setting,
+that string will also be tokenized for us. To see a more complex template in action, though, let's use the 
+`meta-llama/Llama-2-7b-chat-hf` model. Note that this model has gated access, so you will have to
+[request access on the repo](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) if you want to run this code yourself:
+
+```python
+>> from transformers import AutoTokenizer
+>> tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
+
+>> chat = [
+...   {"role": "user", "content": "Hello, how are you?"},
+...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
+...   {"role": "user", "content": "I'd like to show off how chat templating works!"},
+... ]
+
+>> tokenizer.use_default_system_prompt = False
+>> tokenizer.apply_chat_template(chat, tokenize=False)
+"<s>[INST] Hello, how are you? [/INST] I'm doing great. How can I help you today? </s><s>[INST] I'd like to show off how chat templating works! [/INST]"
+```
+
+Note that this time, the tokenizer has added the control tokens [INST] and [/INST] to indicate the start and end of 
+user messages (but not assistant messages!)
+
+## How do chat templates work?
+
+The chat template for a model is stored on the `tokenizer.chat_template` attribute. If no chat template is set, the
+default template for that model class is used instead. Let's take a look at the template for `BlenderBot`:
+
+```python
+
+>>> from transformers import AutoTokenizer
+>>> tokenizer = AutoTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
+
+>>> tokenizer.default_chat_template
+"{% for message in messages %}{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}{{ message['content'] }}{% if not loop.last %}{{ '  ' }}{% endif %}{% endfor %}{{ eos_token }}"
+```
+
+That's kind of intimidating. Let's add some newlines and indentation to make it more readable. Note that the first
+newline after each block as well as any preceding whitespace before a block are ignored by default, using the 
+Jinja `trim_blocks` and `lstrip_blocks` flags. However, be cautious - although leading whitespace on each
+line is stripped, spaces between blocks on the same line are not. We strongly recommend checking that your template
+isn't printing extra spaces where it shouldn't be!
+
+```
+{% for message in messages %}
+    {% if message['role'] == 'user' %}
+        {{ ' ' }}
+    {% endif %}
+    {{ message['content'] }}
+    {% if not loop.last %}
+        {{ '  ' }}
+    {% endif %}
+{% endfor %}
+{{ eos_token }}
+```
+
+If you've never seen one of these before, this is a [Jinja template](https://jinja.palletsprojects.com/en/3.1.x/templates/).
+Jinja is a templating language that allows you to write simple code that generates text. In many ways, the code and
+syntax resembles Python. In pure Python, this template would look something like this:
+
+```python
+for idx, message in enumerate(messages):
+    if message['role'] == 'user':
+        print(' ')
+    print(message['content'])
+    if not idx == len(messages) - 1:  # Check for the last message in the conversation
+        print('  ')
+print(eos_token)
+```
+
+Effectively, the template does three things:
+1. For each message, if the message is a user message, add a blank space before it, otherwise print nothing.
+2. Add the message content
+3. If the message is not the last message, add two spaces after it. After the final message, print the EOS token.
+
+This is a pretty simple template - it doesn't add any control tokens, and it doesn't support "system" messages, which 
+are a common way to give the model directives about how it should behave in the subsequent conversation.
+But Jinja gives you a lot of flexibility to do those things! Let's see a Jinja template that can format inputs
+similarly to the way LLaMA formats them (note that the real LLaMA template includes handling for default system
+messages and slightly different system message handling in general - don't use this one in your actual code!)
+
+```
+{% for message in messages %}
+    {% if message['role'] == 'user' %}
+        {{ bos_token + '[INST] ' + message['content'] + ' [/INST]' }}
+    {% elif message['role'] == 'system' %}
+        {{ '<<SYS>>\\n' + message['content'] + '\\n<</SYS>>\\n\\n' }}
+    {% elif message['role'] == 'assistant' %}
+        {{ ' '  + message['content'] + ' ' + eos_token }}
+    {% endif %}
+{% endfor %}
+```
+
+Hopefully if you stare at this for a little bit you can see what this template is doing - it adds specific tokens based
+on the "role" of each message, which represents who sent it. User, assistant and system messages are clearly
+distinguishable to the model because of the tokens they're wrapped in.
+
+## How do I create a chat template?
+
+Simple, just write a jinja template and set `tokenizer.chat_template`. You may find it easier to start with an 
+existing template from another model and simply edit it for your needs! For example, we could take the LLaMA template
+above and add "[ASST]" and "[/ASST]" to assistant messages:
+
+```
+{% for message in messages %}
+    {% if message['role'] == 'user' %}
+        {{ bos_token + '[INST] ' + message['content'].strip() + ' [/INST]' }}
+    {% elif message['role'] == 'system' %}
+        {{ '<<SYS>>\\n' + message['content'].strip() + '\\n<</SYS>>\\n\\n' }}
+    {% elif message['role'] == 'assistant' %}
+        {{ '[ASST] '  + message['content'] + ' [/ASST]' + eos_token }}
+    {% endif %}
+{% endfor %}
+```
+
+Now, simply set the `tokenizer.chat_template` attribute. Next time you use [`~PreTrainedTokenizer.apply_chat_template`], it will
+use your new template! This attribute will be saved in the `tokenizer_config.json` file, so you can use
+[`~utils.PushToHubMixin.push_to_hub`] to upload your new template to the Hub and make sure everyone's using the right
+template for your model!
+
+```python
+template = tokenizer.chat_template
+template = template.replace("SYS", "SYSTEM")  # Change the system token
+tokenizer.chat_template = template  # Set the new template
+tokenizer.push_to_hub("model_name")  # Upload your new template to the Hub!
+```
+
+The method [`~PreTrainedTokenizer.apply_chat_template`] which uses your chat template is called by the [`ConversationalPipeline`] class, so 
+once you set the correct chat template, your model will automatically become compatible with [`ConversationalPipeline`].
+
+## What are "default" templates?
+
+Before the introduction of chat templates, chat handling was hardcoded at the model class level. For backwards 
+compatibility, we have retained this class-specific handling as default templates, also set at the class level. If a
+model does not have a chat template set, but there is a default template for its model class, the `ConversationalPipeline`
+class and methods like `apply_chat_template` will use the class template instead. You can find out what the default
+template for your tokenizer is by checking the `tokenizer.default_chat_template` attribute.
+
+This is something we do purely for backward compatibility reasons, to avoid breaking any existing workflows. Even when
+the class template is appropriate for your model, we strongly recommend overriding the default template by
+setting the `chat_template` attribute explicitly to make it clear to users that your model has been correctly configured
+for chat, and to future-proof in case the default templates are ever altered or deprecated.
+
+## What template should I use?
+
+When setting the template for a model that's already been trained for chat, you should ensure that the template
+exactly matches the message formatting that the model saw during training, or else you will probably experience
+performance degradation. This is true even if you're training the model further - you will probably get the best 
+performance if you keep the chat tokens constant. This is very analogous to tokenization - you generally get the
+best performance for inference or fine-tuning when you precisely match the tokenization used during training.
+
+If you're training a model from scratch, or fine-tuning a base language model for chat, on the other hand,
+you have a lot of freedom to choose an appropriate template! LLMs are smart enough to learn to handle lots of different
+input formats. Our default template for models that don't have a class-specific template follows the 
+[ChatML format](https://github.com/openai/openai-python/blob/main/chatml.md), and this is a good, flexible choice for many use-cases. It looks like this:
+
+```
+{% for message in messages %}
+    {{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}
+{% endfor %}
+```
+
+If you like this one, here it is in one-liner form, ready to copy into your code. The one-liner also includes
+handy support for "generation prompts" - see the next section for more!
+
+```
+tokenizer.chat_template = "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
+```
+
+This template wraps each message in `<|im_start|>` and `<|im_end|>` tokens, and simply writes the role as a string, which
+allows for flexibility in the roles you train with. The output looks like this:
+
+```
+<|im_start|>system
+You are a helpful chatbot that will do its best not to say anything so stupid that people tweet about it.<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+I'm doing great!<|im_end|>
+```
+
+The "user", "system" and "assistant" roles are the standard for chat, and we recommend using them when it makes sense,
+particularly if you want your model to operate well with [`ConversationalPipeline`]. However, you are not limited
+to these roles - templating is extremely flexible, and any string can be a role.
+
+## What are "generation prompts"?
+
+You may notice that the `apply_chat_template` method has an `add_generation_prompt` argument. This argument tells
+the template to add tokens that indicate the start of a bot response. For example, consider the following chat:
+
+```python
+messages = [
+    {"role": "user", "content": "Hi there!"},
+    {"role": "assistant", "content": "Nice to meet you!"},
+    {"role": "user", "content": "Can I ask a question?"}
+]
+```
+
+Here's what this will look like without a generation prompt, using the ChatML template we described above:
+
+```python
+>> tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=False)
+"""<|im_start|>user
+Hi there!<|im_end|>
+<|im_start|>assistant
+Nice to meet you!<|im_end|>
+<|im_start|>user
+Can I ask a question?<|im_end|>
+"""
+```
+
+And here's what it looks like **with** a generation prompt:
+
+```python
+>> tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+"""<|im_start|>user
+Hi there!<|im_end|>
+<|im_start|>assistant
+Nice to meet you!<|im_end|>
+<|im_start|>user
+Can I ask a question?<|im_end|>
+<|im_start|>assistant
+"""
+```
+
+Note that this time, we've added the tokens that indicate the start of a bot response. This ensures that when the model
+generates text it will write a bot response instead of doing something unexpected, like continuing the user's 
+message. Remember, chat models are still just language models - they're trained to continue text, and chat is just a 
+special kind of text to them! You need to guide them with the appropriate control tokens so they know what they're 
+supposed to be doing.
+
+Not all models require generation prompts. Some models, like BlenderBot and LLaMA, don't have any
+special tokens before bot responses. In these cases, the `add_generation_prompt` argument will have no effect. The exact
+effect that `add_generation_prompt` has will depend on the template being used.
+
+## I want to use chat templates! How should I get started?
+
+If you have any chat models, you should set their `tokenizer.chat_template` attribute and test it using
+[`~PreTrainedTokenizer.apply_chat_template`]. This applies even if you're not the model owner - if you're using a model
+with an empty chat template, or one that's still using the default class template, please open a [pull request](https://huggingface.co/docs/hub/repositories-pull-requests-discussions) to
+the model repository so that this attribute can be set properly!
+
+Once the attribute is set, that's it, you're done! `tokenizer.apply_chat_template` will now work correctly for that
+model, which means it is also automatically supported in places like `ConversationalPipeline`!
+
+By ensuring that models have this attribute, we can make sure that the whole community gets to use the full power of
+open-source models. Formatting mismatches have been haunting the field and silently harming performance for too long - 
+it's time to put an end to them!
+
+## Template writing tips
+
+If you're unfamiliar with Jinja, we generally find that the easiest way to write a chat template is to first
+write a short Python script that formats messages the way you want, and then convert that script into a template.
+
+Remember that the template handler will receive the conversation history as a variable called `messages`. Each
+message is a dictionary with two keys, `role` and `content`. You will be able to access `messages` in your template
+just like you can in Python, which means you can loop over it with `{% for message in messages %}` or access
+individual messages with, for example, `{{ messages[0] }}`.
+
+You can also use the following tips to convert your code to Jinja:
+
+### For loops
+
+For loops in Jinja look like this:
+
+```
+{% for message in messages %}
+{{ message['content'] }}
+{% endfor %}
+```
+
+Note that whatever's inside the {{ expression block }} will be printed to the output. You can use operators like
+`+` to combine strings inside expression blocks.
+
+### If statements
+
+If statements in Jinja look like this:
+
+```
+{% if message['role'] == 'user' %}
+{{ message['content'] }}
+{% endif %}
+```
+
+Note how where Python uses whitespace to mark the beginnings and ends of `for` and `if` blocks, Jinja requires you
+to explicitly end them with `{% endfor %}` and `{% endif %}`.
+
+### Special variables
+
+Inside your template, you will have access to the list of `messages`, but you can also access several other special
+variables. These include special tokens like `bos_token` and `eos_token`, as well as the `add_generation_prompt`
+variable that we discussed above. You can also use the `loop` variable to access information about the current loop
+iteration, for example  using `{% if loop.last %}` to check if the current message is the last message in the 
+conversation. Here's an example that puts these ideas together to add a generation prompt at the end of the
+conversation if add_generation_prompt is `True`:
+
+```
+{% if loop.last and add_generation_prompt %}
+{{ bos_token + 'Assistant:\n' }}
+{% endif %}
+```
+
+### Notes on whitespace
+
+As much as possible, we've tried to get Jinja to ignore whitespace outside of {{ expressions }}. However, be aware
+that Jinja is a general-purpose templating engine, and it may treat whitespace between blocks on the same line
+as significant and print it to the output. We **strongly** recommend checking that your template isn't printing extra
+spaces where it shouldn't be before you upload it!

docs/source/en/community.md

@@ -1,4 +1,4 @@
-<!--⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+<!--⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
 rendered properly in your Markdown viewer.
 -->
 
@@ -10,7 +10,7 @@ This page regroups resources around 🤗 Transformers developed by the community
 
 | Resource     |      Description      |      Author      |
 |:----------|:-------------|------:|
-| [Hugging Face Transformers Glossary Flashcards](https://www.darigovresearch.com/huggingface-transformers-glossary-flashcards) | A set of flashcards based on the [Transformers Docs Glossary](glossary) that has been put into a form which can be easily learnt/revised using [Anki ](https://apps.ankiweb.net/) an open source, cross platform app specifically designed for long term knowledge retention. See this [Introductory video on how to use the flashcards](https://www.youtube.com/watch?v=Dji_h7PILrw). | [Darigov Research](https://www.darigovresearch.com/) |
+| [Hugging Face Transformers Glossary Flashcards](https://www.darigovresearch.com/huggingface-transformers-glossary-flashcards) | A set of flashcards based on the [Transformers Docs Glossary](glossary) that has been put into a form which can be easily learned/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:
 
@@ -35,7 +35,7 @@ This page regroups resources around 🤗 Transformers developed by the community
 |[Speed up Fine-Tuning in Transformers with Dynamic Padding / Bucketing](https://github.com/ELS-RD/transformers-notebook/blob/master/Divide_Hugging_Face_Transformers_training_time_by_2_or_more.ipynb)|How to speed up fine-tuning by a factor of 2 using dynamic padding / bucketing|[Michael Benesty](https://github.com/pommedeterresautee) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1CBfRU1zbfu7-ijiOqAAQUA-RJaxfcJoO?usp=sharing)|
 |[Pretrain Reformer for Masked Language Modeling](https://github.com/patrickvonplaten/notebooks/blob/master/Reformer_For_Masked_LM.ipynb)| How to train a Reformer model with bi-directional self-attention layers | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1tzzh0i8PgDQGV3SMFUGxM7_gGae3K-uW?usp=sharing)|
 |[Expand and Fine Tune Sci-BERT](https://github.com/lordtt13/word-embeddings/blob/master/COVID-19%20Research%20Data/COVID-SciBERT.ipynb)| How to increase vocabulary of a pretrained SciBERT model from AllenAI on the CORD dataset and pipeline it. | [Tanmay Thakur](https://github.com/lordtt13) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1rqAR40goxbAfez1xvF3hBJphSCsvXmh8)|
-|[Fine Tune BlenderBotSmall for Summarization using the Trainer API](https://github.com/lordtt13/transformers-experiments/blob/master/Custom%20Tasks/fine-tune-blenderbot_small-for-summarization.ipynb)| How to fine tune BlenderBotSmall for summarization on a custom dataset, using the Trainer API. | [Tanmay Thakur](https://github.com/lordtt13) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/19Wmupuls7mykSGyRN_Qo6lPQhgp56ymq?usp=sharing)|
+|[Fine Tune BlenderBotSmall for Summarization using the Trainer API](https://github.com/lordtt13/transformers-experiments/blob/master/Custom%20Tasks/fine-tune-blenderbot_small-for-summarization.ipynb)| How to fine-tune BlenderBotSmall for summarization on a custom dataset, using the Trainer API. | [Tanmay Thakur](https://github.com/lordtt13) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/19Wmupuls7mykSGyRN_Qo6lPQhgp56ymq?usp=sharing)|
 |[Fine-tune Electra and interpret with Integrated Gradients](https://github.com/elsanns/xai-nlp-notebooks/blob/master/electra_fine_tune_interpret_captum_ig.ipynb) | How to fine-tune Electra for sentiment analysis and interpret predictions with Captum Integrated Gradients | [Eliza Szczechla](https://elsanns.github.io) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/elsanns/xai-nlp-notebooks/blob/master/electra_fine_tune_interpret_captum_ig.ipynb)|
 |[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)|

docs/source/en/create_a_model.md

@@ -209,7 +209,7 @@ Easily reuse this checkpoint for another task by switching to a different model
 The last base class you need before using a model for textual data is a [tokenizer](main_classes/tokenizer) to convert raw text to tensors. There are two types of tokenizers you can use with 🤗 Transformers:
 
 - [`PreTrainedTokenizer`]: a Python implementation of a tokenizer.
-- [`PreTrainedTokenizerFast`]: a tokenizer from our Rust-based [🤗 Tokenizer](https://huggingface.co/docs/tokenizers/python/latest/) library. This tokenizer type is significantly faster - especially during batch tokenization - due to it's Rust implementation. The fast tokenizer also offers additional methods like *offset mapping* which maps tokens to their original words or characters.
+- [`PreTrainedTokenizerFast`]: a tokenizer from our Rust-based [🤗 Tokenizer](https://huggingface.co/docs/tokenizers/python/latest/) library. This tokenizer type is significantly faster - especially during batch tokenization - due to its Rust implementation. The fast tokenizer also offers additional methods like *offset mapping* which maps tokens to their original words or characters.
 
 Both tokenizers support common methods such as encoding and decoding, adding new tokens, and managing special tokens.
 

docs/source/en/custom_models.md

@@ -341,7 +341,7 @@ model. This is different from pushing the code to the Hub in the sense that user
 get the custom models (contrarily to automatically downloading the model code from the Hub).
 
 As long as your config has a `model_type` attribute that is different from existing model types, and that your model
-classes have the right `config_class` attributes, you can just add them to the auto classes likes this:
+classes have the right `config_class` attributes, you can just add them to the auto classes like this:
 
 ```py
 from transformers import AutoConfig, AutoModel, AutoModelForImageClassification

docs/source/en/custom_tools.md

@@ -25,7 +25,7 @@ If you are not aware of what tools and agents are in the context of transformers
 
 <Tip warning={true}>
 
-Transformers Agent is an experimental API that is subject to change at any time. Results returned by the agents
+Transformers Agents is an experimental API that is subject to change at any time. Results returned by the agents
 can vary as the APIs or underlying models are prone to change.
 
 </Tip>

docs/source/en/generation_strategies.md

@@ -55,12 +55,10 @@ When you load a model explicitly, you can inspect the generation configuration t
 >>> from transformers import AutoModelForCausalLM
 
 >>> model = AutoModelForCausalLM.from_pretrained("distilgpt2")
->>> model.generation_config  # doctest: +IGNORE_RESULT
+>>> model.generation_config
 GenerationConfig {
-    "_from_model_config": true,
     "bos_token_id": 50256,
     "eos_token_id": 50256,
-    "transformers_version": "4.26.0.dev0"
 }
 ```
 
@@ -84,7 +82,8 @@ Even if the default decoding strategy mostly works for your task, you can still
 commonly adjusted parameters include:
 
 - `max_new_tokens`: the maximum number of tokens to generate. In other words, the size of the output sequence, not
-including the tokens in the prompt.
+including the tokens in the prompt. As an alternative to using the output's length as a stopping criteria, you can choose 
+to stop generation whenever the full generation exceeds some amount of time. To learn more, check [`StoppingCriteria`].
 - `num_beams`: by specifying a number of beams higher than 1, you are effectively switching from greedy search to
 beam search. This strategy evaluates several hypotheses at each time step and eventually chooses the hypothesis that
 has the overall highest probability for the entire sequence. This has the advantage of identifying high-probability
@@ -92,7 +91,7 @@ sequences that start with a lower probability initial tokens and would've been i
 - `do_sample`: if set to `True`, this parameter enables decoding strategies such as multinomial sampling, beam-search
 multinomial sampling, Top-K sampling and Top-p sampling. All these strategies select the next token from the probability
 distribution over the entire vocabulary with various strategy-specific adjustments.
-- `num_return_sequences`: the number of sequence candidates to return for each input. This options is only available for
+- `num_return_sequences`: the number of sequence candidates to return for each input. This option is only available for
 the decoding strategies that support multiple sequence candidates, e.g. variations of beam search and sampling. Decoding
 strategies like greedy search and contrastive search return a single output sequence.
 
@@ -146,7 +145,7 @@ one for summarization with beam search). You must have the right Hub permissions
 
 ## Streaming
 
-The `generate()` supports streaming, through its `streamer` input. The `streamer` input is compatible any instance
+The `generate()` supports streaming, through its `streamer` input. The `streamer` input is compatible with any instance
 from a class that has the following methods: `put()` and `end()`. Internally, `put()` is used to push new tokens and
 `end()` is used to flag the end of text generation.
 
@@ -301,7 +300,7 @@ the `num_beams` greater than 1, and set `do_sample=True` to use this decoding st
 The diverse beam search decoding strategy is an extension of the beam search strategy that allows for generating a more diverse
 set of beam sequences to choose from. To learn how it works, refer to [Diverse Beam Search: Decoding Diverse Solutions from Neural Sequence Models](https://arxiv.org/pdf/1610.02424.pdf).
 This approach has three main parameters: `num_beams`, `num_beam_groups`, and `diversity_penalty`.
-The diversily penalty ensures the outputs are distinct across groups, and beam search is used within each group.
+The diversity penalty ensures the outputs are distinct across groups, and beam search is used within each group.
 
 
 ```python
@@ -332,7 +331,8 @@ The diversily penalty ensures the outputs are distinct across groups, and beam s
 
 >>> outputs = model.generate(**inputs, num_beams=5, num_beam_groups=5, max_new_tokens=30, diversity_penalty=1.0)
 >>> tokenizer.decode(outputs[0], skip_special_tokens=True)
-'The aim of this project is to create a new type of living system, one that is more sustainable and efficient than the current one.'
+'The Design Principles are a set of universal design principles that can be applied to any location, climate and
+culture, and they allow us to design the'
 ```
 
 This guide illustrates the main parameters that enable various decoding strategies. More advanced parameters exist for the
@@ -366,7 +366,7 @@ To enable assisted decoding, set the `assistant_model` argument with a model.
 ['Alice and Bob are sitting in a bar. Alice is drinking a beer and Bob is drinking a']
 ```
 
-When using assisted decoding with sampling methods, you can use the `temperarure` argument to control the randomness
+When using assisted decoding with sampling methods, you can use the `temperature` argument to control the randomness
 just like in multinomial sampling. However, in assisted decoding, reducing the temperature will help improving latency.
 
 ```python

docs/source/en/glossary.md

@@ -187,7 +187,7 @@ The model head refers to the last layer of a neural network that accepts the raw
 
 ### image patch
 
-Vision-based Transformers models split an image into smaller patches which are linearly embedded, and then passed as a sequence to the model. You can find the `patch_size` - or resolution - of the model in it's configuration.
+Vision-based Transformers models split an image into smaller patches which are linearly embedded, and then passed as a sequence to the model. You can find the `patch_size` - or resolution - of the model in its configuration.
 
 ### inference
 

docs/source/en/hpo_train.md

@@ -54,6 +54,18 @@ For optuna, see optuna [object_parameter](https://optuna.readthedocs.io/en/stabl
 ...     }
 ```
 
+Optuna provides multi-objective HPO. You can pass `direction` in `hyperparameter_search` and define your own compute_objective to return multiple objective values. The Pareto Front (`List[BestRun]`) will be returned in hyperparameter_search, you should refer to the test case `TrainerHyperParameterMultiObjectOptunaIntegrationTest` in [test_trainer](https://github.com/huggingface/transformers/blob/main/tests/trainer/test_trainer.py). It's like following
+
+```py
+>>> best_trials = trainer.hyperparameter_search(
+...     direction=["minimize", "maximize"],
+...     backend="optuna",
+...     hp_space=optuna_hp_space,
+...     n_trials=20,
+...     compute_objective=compute_objective,
+... )
+```
+
 For raytune, see raytune [object_parameter](https://docs.ray.io/en/latest/tune/api/search_space.html), it's like following:
 
 ```py

docs/source/en/index.md

@@ -48,235 +48,8 @@ The documentation is organized into five sections:
   - **MODELS** details the classes and functions related to each model implemented in the library.
   - **INTERNAL HELPERS** details utility classes and functions used internally.
 
-### 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. **[ALIGN](model_doc/align)** (from Google Research) released with the paper [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.
-1. **[AltCLIP](model_doc/altclip)** (from BAAI) released with the paper [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679) by Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell.
-1. **[Audio Spectrogram Transformer](model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
-1. **[Autoformer](model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
-1. **[Bark](model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
-1. **[BART](model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BioGpt](model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
-1. **[BiT](model_doc/bit)** (from Google AI) released with the paper [Big Transfer (BiT): General Visual Representation Learning](https://arxiv.org/abs/1912.11370) by Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil Houlsby.
-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. **[BLIP](model_doc/blip)** (from Salesforce) released with the paper [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://arxiv.org/abs/2201.12086) by Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi.
-1. **[BLIP-2](model_doc/blip-2)** (from Salesforce) released with the paper [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://arxiv.org/abs/2301.12597) by Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi.
-1. **[BLOOM](model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[BridgeTower](model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
-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. **[Chinese-CLIP](model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
-1. **[CLAP](model_doc/clap)** (from LAION-AI) released with the paper [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://arxiv.org/abs/2211.06687) by Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, Shlomo Dubnov.
-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. **[CLIPSeg](model_doc/clipseg)** (from University of Göttingen) released with the paper [Image Segmentation Using Text and Image Prompts](https://arxiv.org/abs/2112.10003) by Timo Lüddecke and Alexander Ecker.
-1. **[CodeGen](model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[CodeLlama](model_doc/llama_code)** (from MetaAI) released with the paper [Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/) by Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong, Alexandre Défossez, Jade Copet, Faisal Azhar, Hugo Touvron, Louis Martin, Nicolas Usunier, Thomas Scialom, Gabriel Synnaeve.
-1. **[Conditional DETR](model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[ConvNeXTV2](model_doc/convnextv2)** (from Facebook AI) released with the paper [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) by Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, Saining Xie.
-1. **[CPM](model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CPM-Ant](model_doc/cpmant)** (from OpenBMB) released by the [OpenBMB](https://www.openbmb.org/).
-1. **[CTRL](model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](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. **[DePlot](model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
-1. **[DETA](model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
-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. **[DiNAT](model_doc/dinat)** (from SHI Labs) released with the paper [Dilated Neighborhood Attention Transformer](https://arxiv.org/abs/2209.15001) by Ali Hassani and Humphrey Shi.
-1. **[DINOv2](model_doc/dinov2)** (from Meta AI) released with the paper [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
-1. **[DistilBERT](model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
-1. **[DiT](model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[EfficientFormer](model_doc/efficientformer)** (from Snap Research) released with the paper [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191) by Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren.
-1. **[EfficientNet](model_doc/efficientnet)** (from Google Brain) released with the paper [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946) by Mingxing Tan, Quoc V. Le.
-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. **[EnCodec](model_doc/encodec)** (from Meta AI) released with the paper [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) by Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.
-1. **[EncoderDecoder](model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ErnieM](model_doc/ernie_m)** (from Baidu) released with the paper [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) by Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang.
-1. **[ESM](model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2 and ESMFold** were released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[Falcon](model_doc/falcon)** (from Technology Innovation Institute) by Almazrouei, Ebtesam and Alobeidli, Hamza and Alshamsi, Abdulaziz and Cappelli, Alessandro and Cojocaru, Ruxandra and Debbah, Merouane and Goffinet, Etienne and Heslow, Daniel and Launay, Julien and Malartic, Quentin and Noune, Badreddine and Pannier, Baptiste and Penedo, Guilherme.
-1. **[FLAN-T5](model_doc/flan-t5)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-t5-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
-1. **[FLAN-UL2](model_doc/flan-ul2)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-ul2-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
-1. **[FlauBERT](model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[FocalNet](model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
-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. **[GIT](model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
-1. **[GLPN](model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-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-Sw3](model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
-1. **[GPTBigCode](model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
-1. **[GPTSAN-japanese](model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
-1. **[Graphormer](model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
-1. **[GroupViT](model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[Hubert](model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[IDEFICS](model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh.
-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. **[Informer](model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
-1. **[InstructBLIP](model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
-1. **[Jukebox](model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[LayoutLM](model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[LLaMA](model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
-1. **[Llama2](model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/XXX) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
-1. **[Longformer](model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[Mask2Former](model_doc/mask2former)** (from FAIR and UIUC) released with the paper [Masked-attention Mask Transformer for Universal Image Segmentation](https://arxiv.org/abs/2112.01527) by Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, Rohit Girdhar.
-1. **[MaskFormer](model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[MatCha](model_doc/matcha)** (from Google AI) released with the paper [MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering](https://arxiv.org/abs/2212.09662) by Fangyu Liu, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Yasemin Altun, Nigel Collier, Julian Martin Eisenschlos.
-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. **[MEGA](model_doc/mega)** (from Meta/USC/CMU/SJTU) released with the paper [Mega: Moving Average Equipped Gated Attention](https://arxiv.org/abs/2209.10655) by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer.
-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. **[MGP-STR](model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
-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. **[MMS](model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
-1. **[MobileBERT](model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileNetV1](model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
-1. **[MobileNetV2](model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
-1. **[MobileViT](model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MobileViTV2](model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MPT](model_doc/mpt)** (from MosaiML) released with the repository [llm-foundry](https://github.com/mosaicml/llm-foundry/) by the MosaicML NLP Team.
-1. **[MRA](model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
-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. **[MusicGen](model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
-1. **[MVP](model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[NAT](model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
-1. **[Nezha](model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[NLLB-MOE](model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nyströmformer](model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OneFormer](model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
-1. **[OPT](master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
-1. **[Perceiver IO](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. **[Pix2Struct](model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
-1. **[PLBart](model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
-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. **[PVT](model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
-1. **[QDQBert](model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoBERTa-PreLayerNorm](model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
-1. **[RoCBert](model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
-1. **[RoFormer](model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[RWKV](model_doc/rwkv)** (from Bo Peng), released on [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
-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. **[Segment Anything](model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
-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. **[SpeechT5](model_doc/speecht5)** (from Microsoft Research) released with the paper [SpeechT5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing](https://arxiv.org/abs/2110.07205) by Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, Zhihua Wei, Yao Qian, Jinyu Li, Furu Wei.
-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. **[SwiftFormer](model_doc/swiftformer)** (from MBZUAI) released with the paper [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) by Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.
-1. **[Swin Transformer](model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[Swin2SR](model_doc/swin2sr)** (from University of Würzburg) released with the paper [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) by Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte.
-1. **[SwitchTransformers](model_doc/switch_transformers)** (from Google) released with the paper [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961) by William Fedus, Barret Zoph, Noam Shazeer.
-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. **[Table Transformer](model_doc/table-transformer)** (from Microsoft Research) released with the paper [PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents](https://arxiv.org/abs/2110.00061) by Brandon Smock, Rohith Pesala, Robin Abraham.
-1. **[TAPAS](model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](model_doc/time_series_transformer)** (from HuggingFace).
-1. **[TimeSformer](model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
-1. **[Trajectory Transformer](model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[TVLT](model_doc/tvlt)** (from UNC Chapel Hill) released with the paper [TVLT: Textless Vision-Language Transformer](https://arxiv.org/abs/2209.14156) by Zineng Tang, Jaemin Cho, Yixin Nie, Mohit Bansal.
-1. **[UL2](model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UMT5](model_doc/umt5)** (from Google Research) released with the paper [UniMax: Fairer and More Effective Language Sampling for Large-Scale Multilingual Pretraining](https://openreview.net/forum?id=kXwdL1cWOAi) by Hyung Won Chung, Xavier Garcia, Adam Roberts, Yi Tay, Orhan Firat, Sharan Narang, Noah Constant.
-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. **[UPerNet](model_doc/upernet)** (from Peking University) released with the paper [Unified Perceptual Parsing for Scene Understanding](https://arxiv.org/abs/1807.10221) by Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, Jian Sun.
-1. **[VAN](model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViT Hybrid](model_doc/vit_hybrid)** (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. **[VitDet](model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
-1. **[ViTMAE](model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMSN](model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[ViViT](model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
-1. **[Wav2Vec2](model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[X-MOD](model_doc/xmod)** (from Meta AI) released with the paper [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) by Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe.
-1. **[XGLM](model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLM-V](model_doc/xlm-v)** (from Meta AI) released with the paper [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://arxiv.org/abs/2301.10472) by Davis Liang, Hila Gonen, Yuning Mao, Rui Hou, Naman Goyal, Marjan Ghazvininejad, Luke Zettlemoyer, Madian Khabsa.
-1. **[XLNet](model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-
-
-### Supported frameworks
+## Supported models and 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
@@ -284,211 +57,245 @@ Flax), PyTorch, and/or TensorFlow.
 
 <!--This table is updated automatically from the auto modules with _make fix-copies_. Do not update manually!-->
 
-|             Model             | PyTorch support | TensorFlow support | Flax Support |
-|:-----------------------------:|:---------------:|:------------------:|:------------:|
-|            ALBERT             |       ✅        |         ✅         |      ✅      |
-|             ALIGN             |       ✅        |         ❌         |      ❌      |
-|            AltCLIP            |       ✅        |         ❌         |      ❌      |
-| Audio Spectrogram Transformer |       ✅        |         ❌         |      ❌      |
-|          Autoformer           |       ✅        |         ❌         |      ❌      |
-|             Bark              |       ✅        |         ❌         |      ❌      |
-|             BART              |       ✅        |         ✅         |      ✅      |
-|             BEiT              |       ✅        |         ❌         |      ✅      |
-|             BERT              |       ✅        |         ✅         |      ✅      |
-|        Bert Generation        |       ✅        |         ❌         |      ❌      |
-|            BigBird            |       ✅        |         ❌         |      ✅      |
-|        BigBird-Pegasus        |       ✅        |         ❌         |      ❌      |
-|            BioGpt             |       ✅        |         ❌         |      ❌      |
-|              BiT              |       ✅        |         ❌         |      ❌      |
-|          Blenderbot           |       ✅        |         ✅         |      ✅      |
-|        BlenderbotSmall        |       ✅        |         ✅         |      ✅      |
-|             BLIP              |       ✅        |         ✅         |      ❌      |
-|            BLIP-2             |       ✅        |         ❌         |      ❌      |
-|             BLOOM             |       ✅        |         ❌         |      ✅      |
-|          BridgeTower          |       ✅        |         ❌         |      ❌      |
-|           CamemBERT           |       ✅        |         ✅         |      ❌      |
-|            CANINE             |       ✅        |         ❌         |      ❌      |
-|         Chinese-CLIP          |       ✅        |         ❌         |      ❌      |
-|             CLAP              |       ✅        |         ❌         |      ❌      |
-|             CLIP              |       ✅        |         ✅         |      ✅      |
-|            CLIPSeg            |       ✅        |         ❌         |      ❌      |
-|            CodeGen            |       ✅        |         ❌         |      ❌      |
-|           CodeLlama           |       ✅        |         ❌         |      ❌      |
-|       Conditional DETR        |       ✅        |         ❌         |      ❌      |
-|           ConvBERT            |       ✅        |         ✅         |      ❌      |
-|           ConvNeXT            |       ✅        |         ✅         |      ❌      |
-|          ConvNeXTV2           |       ✅        |         ❌         |      ❌      |
-|            CPM-Ant            |       ✅        |         ❌         |      ❌      |
-|             CTRL              |       ✅        |         ✅         |      ❌      |
-|              CvT              |       ✅        |         ✅         |      ❌      |
-|         Data2VecAudio         |       ✅        |         ❌         |      ❌      |
-|         Data2VecText          |       ✅        |         ❌         |      ❌      |
-|        Data2VecVision         |       ✅        |         ✅         |      ❌      |
-|            DeBERTa            |       ✅        |         ✅         |      ❌      |
-|          DeBERTa-v2           |       ✅        |         ✅         |      ❌      |
-|     Decision Transformer      |       ✅        |         ❌         |      ❌      |
-|        Deformable DETR        |       ✅        |         ❌         |      ❌      |
-|             DeiT              |       ✅        |         ✅         |      ❌      |
-|             DETA              |       ✅        |         ❌         |      ❌      |
-|             DETR              |       ✅        |         ❌         |      ❌      |
-|             DiNAT             |       ✅        |         ❌         |      ❌      |
-|            DINOv2             |       ✅        |         ❌         |      ❌      |
-|          DistilBERT           |       ✅        |         ✅         |      ✅      |
-|           DonutSwin           |       ✅        |         ❌         |      ❌      |
-|              DPR              |       ✅        |         ✅         |      ❌      |
-|              DPT              |       ✅        |         ❌         |      ❌      |
-|        EfficientFormer        |       ✅        |         ✅         |      ❌      |
-|         EfficientNet          |       ✅        |         ❌         |      ❌      |
-|            ELECTRA            |       ✅        |         ✅         |      ✅      |
-|            EnCodec            |       ✅        |         ❌         |      ❌      |
-|        Encoder decoder        |       ✅        |         ✅         |      ✅      |
-|             ERNIE             |       ✅        |         ❌         |      ❌      |
-|            ErnieM             |       ✅        |         ❌         |      ❌      |
-|              ESM              |       ✅        |         ✅         |      ❌      |
-|  FairSeq Machine-Translation  |       ✅        |         ❌         |      ❌      |
-|            Falcon             |       ✅        |         ❌         |      ❌      |
-|           FlauBERT            |       ✅        |         ✅         |      ❌      |
-|             FLAVA             |       ✅        |         ❌         |      ❌      |
-|             FNet              |       ✅        |         ❌         |      ❌      |
-|           FocalNet            |       ✅        |         ❌         |      ❌      |
-|      Funnel Transformer       |       ✅        |         ✅         |      ❌      |
-|              GIT              |       ✅        |         ❌         |      ❌      |
-|             GLPN              |       ✅        |         ❌         |      ❌      |
-|            GPT Neo            |       ✅        |         ❌         |      ✅      |
-|           GPT NeoX            |       ✅        |         ❌         |      ❌      |
-|       GPT NeoX Japanese       |       ✅        |         ❌         |      ❌      |
-|             GPT-J             |       ✅        |         ✅         |      ✅      |
-|            GPT-Sw3            |       ✅        |         ✅         |      ✅      |
-|          GPTBigCode           |       ✅        |         ❌         |      ❌      |
-|        GPTSAN-japanese        |       ✅        |         ❌         |      ❌      |
-|          Graphormer           |       ✅        |         ❌         |      ❌      |
-|           GroupViT            |       ✅        |         ✅         |      ❌      |
-|            Hubert             |       ✅        |         ✅         |      ❌      |
-|            I-BERT             |       ✅        |         ❌         |      ❌      |
-|            IDEFICS            |       ✅        |         ❌         |      ❌      |
-|           ImageGPT            |       ✅        |         ❌         |      ❌      |
-|           Informer            |       ✅        |         ❌         |      ❌      |
-|         InstructBLIP          |       ✅        |         ❌         |      ❌      |
-|            Jukebox            |       ✅        |         ❌         |      ❌      |
-|           LayoutLM            |       ✅        |         ✅         |      ❌      |
-|          LayoutLMv2           |       ✅        |         ❌         |      ❌      |
-|          LayoutLMv3           |       ✅        |         ✅         |      ❌      |
-|              LED              |       ✅        |         ✅         |      ❌      |
-|             LeViT             |       ✅        |         ❌         |      ❌      |
-|             LiLT              |       ✅        |         ❌         |      ❌      |
-|             LLaMA             |       ✅        |         ❌         |      ❌      |
-|          Longformer           |       ✅        |         ✅         |      ❌      |
-|            LongT5             |       ✅        |         ❌         |      ✅      |
-|             LUKE              |       ✅        |         ❌         |      ❌      |
-|            LXMERT             |       ✅        |         ✅         |      ❌      |
-|            M-CTC-T            |       ✅        |         ❌         |      ❌      |
-|            M2M100             |       ✅        |         ❌         |      ❌      |
-|            Marian             |       ✅        |         ✅         |      ✅      |
-|           MarkupLM            |       ✅        |         ❌         |      ❌      |
-|          Mask2Former          |       ✅        |         ❌         |      ❌      |
-|          MaskFormer           |       ✅        |         ❌         |      ❌      |
-|        MaskFormerSwin         |       ❌        |         ❌         |      ❌      |
-|             mBART             |       ✅        |         ✅         |      ✅      |
-|             MEGA              |       ✅        |         ❌         |      ❌      |
-|         Megatron-BERT         |       ✅        |         ❌         |      ❌      |
-|            MGP-STR            |       ✅        |         ❌         |      ❌      |
-|          MobileBERT           |       ✅        |         ✅         |      ❌      |
-|          MobileNetV1          |       ✅        |         ❌         |      ❌      |
-|          MobileNetV2          |       ✅        |         ❌         |      ❌      |
-|           MobileViT           |       ✅        |         ✅         |      ❌      |
-|          MobileViTV2          |       ✅        |         ❌         |      ❌      |
-|             MPNet             |       ✅        |         ✅         |      ❌      |
-|              MPT              |       ✅        |         ❌         |      ❌      |
-|              MRA              |       ✅        |         ❌         |      ❌      |
-|              MT5              |       ✅        |         ✅         |      ✅      |
-|           MusicGen            |       ✅        |         ❌         |      ❌      |
-|              MVP              |       ✅        |         ❌         |      ❌      |
-|              NAT              |       ✅        |         ❌         |      ❌      |
-|             Nezha             |       ✅        |         ❌         |      ❌      |
-|           NLLB-MOE            |       ✅        |         ❌         |      ❌      |
-|         Nyströmformer         |       ✅        |         ❌         |      ❌      |
-|           OneFormer           |       ✅        |         ❌         |      ❌      |
-|          OpenAI GPT           |       ✅        |         ✅         |      ❌      |
-|         OpenAI GPT-2          |       ✅        |         ✅         |      ✅      |
-|           OpenLlama           |       ✅        |         ❌         |      ❌      |
-|              OPT              |       ✅        |         ✅         |      ✅      |
-|            OWL-ViT            |       ✅        |         ❌         |      ❌      |
-|            Pegasus            |       ✅        |         ✅         |      ✅      |
-|           PEGASUS-X           |       ✅        |         ❌         |      ❌      |
-|           Perceiver           |       ✅        |         ❌         |      ❌      |
-|          Pix2Struct           |       ✅        |         ❌         |      ❌      |
-|            PLBart             |       ✅        |         ❌         |      ❌      |
-|          PoolFormer           |       ✅        |         ❌         |      ❌      |
-|           Pop2Piano           |       ✅        |         ❌         |      ❌      |
-|          ProphetNet           |       ✅        |         ❌         |      ❌      |
-|              PVT              |       ✅        |         ❌         |      ❌      |
-|            QDQBert            |       ✅        |         ❌         |      ❌      |
-|              RAG              |       ✅        |         ✅         |      ❌      |
-|             REALM             |       ✅        |         ❌         |      ❌      |
-|           Reformer            |       ✅        |         ❌         |      ❌      |
-|            RegNet             |       ✅        |         ✅         |      ✅      |
-|            RemBERT            |       ✅        |         ✅         |      ❌      |
-|            ResNet             |       ✅        |         ✅         |      ✅      |
-|           RetriBERT           |       ✅        |         ❌         |      ❌      |
-|            RoBERTa            |       ✅        |         ✅         |      ✅      |
-|     RoBERTa-PreLayerNorm      |       ✅        |         ✅         |      ✅      |
-|            RoCBert            |       ✅        |         ❌         |      ❌      |
-|           RoFormer            |       ✅        |         ✅         |      ✅      |
-|             RWKV              |       ✅        |         ❌         |      ❌      |
-|              SAM              |       ✅        |         ✅         |      ❌      |
-|           SegFormer           |       ✅        |         ✅         |      ❌      |
-|              SEW              |       ✅        |         ❌         |      ❌      |
-|             SEW-D             |       ✅        |         ❌         |      ❌      |
-|    Speech Encoder decoder     |       ✅        |         ❌         |      ✅      |
-|          Speech2Text          |       ✅        |         ✅         |      ❌      |
-|         Speech2Text2          |       ❌        |         ❌         |      ❌      |
-|           SpeechT5            |       ✅        |         ❌         |      ❌      |
-|           Splinter            |       ✅        |         ❌         |      ❌      |
-|          SqueezeBERT          |       ✅        |         ❌         |      ❌      |
-|          SwiftFormer          |       ✅        |         ❌         |      ❌      |
-|       Swin Transformer        |       ✅        |         ✅         |      ❌      |
-|      Swin Transformer V2      |       ✅        |         ❌         |      ❌      |
-|            Swin2SR            |       ✅        |         ❌         |      ❌      |
-|      SwitchTransformers       |       ✅        |         ❌         |      ❌      |
-|              T5               |       ✅        |         ✅         |      ✅      |
-|       Table Transformer       |       ✅        |         ❌         |      ❌      |
-|             TAPAS             |       ✅        |         ✅         |      ❌      |
-|    Time Series Transformer    |       ✅        |         ❌         |      ❌      |
-|          TimeSformer          |       ✅        |         ❌         |      ❌      |
-|         TimmBackbone          |       ❌        |         ❌         |      ❌      |
-|    Trajectory Transformer     |       ✅        |         ❌         |      ❌      |
-|        Transformer-XL         |       ✅        |         ✅         |      ❌      |
-|             TrOCR             |       ✅        |         ❌         |      ❌      |
-|             TVLT              |       ✅        |         ❌         |      ❌      |
-|             UMT5              |       ✅        |         ❌         |      ❌      |
-|           UniSpeech           |       ✅        |         ❌         |      ❌      |
-|         UniSpeechSat          |       ✅        |         ❌         |      ❌      |
-|            UPerNet            |       ✅        |         ❌         |      ❌      |
-|              VAN              |       ✅        |         ❌         |      ❌      |
-|           VideoMAE            |       ✅        |         ❌         |      ❌      |
-|             ViLT              |       ✅        |         ❌         |      ❌      |
-|    Vision Encoder decoder     |       ✅        |         ✅         |      ✅      |
-|     VisionTextDualEncoder     |       ✅        |         ✅         |      ✅      |
-|          VisualBERT           |       ✅        |         ❌         |      ❌      |
-|              ViT              |       ✅        |         ✅         |      ✅      |
-|          ViT Hybrid           |       ✅        |         ❌         |      ❌      |
-|            VitDet             |       ✅        |         ❌         |      ❌      |
-|            ViTMAE             |       ✅        |         ✅         |      ❌      |
-|            ViTMSN             |       ✅        |         ❌         |      ❌      |
-|             ViViT             |       ✅        |         ❌         |      ❌      |
-|           Wav2Vec2            |       ✅        |         ✅         |      ✅      |
-|      Wav2Vec2-Conformer       |       ✅        |         ❌         |      ❌      |
-|             WavLM             |       ✅        |         ❌         |      ❌      |
-|            Whisper            |       ✅        |         ✅         |      ✅      |
-|            X-CLIP             |       ✅        |         ❌         |      ❌      |
-|             X-MOD             |       ✅        |         ❌         |      ❌      |
-|             XGLM              |       ✅        |         ✅         |      ✅      |
-|              XLM              |       ✅        |         ✅         |      ❌      |
-|        XLM-ProphetNet         |       ✅        |         ❌         |      ❌      |
-|          XLM-RoBERTa          |       ✅        |         ✅         |      ✅      |
-|        XLM-RoBERTa-XL         |       ✅        |         ❌         |      ❌      |
-|             XLNet             |       ✅        |         ✅         |      ❌      |
-|             YOLOS             |       ✅        |         ❌         |      ❌      |
-|             YOSO              |       ✅        |         ❌         |      ❌      |
+|                                  Model                                   | PyTorch support | TensorFlow support | Flax Support |
+|:------------------------------------------------------------------------:|:---------------:|:------------------:|:------------:|
+|                        [ALBERT](model_doc/albert)                        |       ✅        |         ✅         |      ✅      |
+|                         [ALIGN](model_doc/align)                         |       ✅        |         ❌         |      ❌      |
+|                       [AltCLIP](model_doc/altclip)                       |       ✅        |         ❌         |      ❌      |
+| [Audio Spectrogram Transformer](model_doc/audio-spectrogram-transformer) |       ✅        |         ❌         |      ❌      |
+|                    [Autoformer](model_doc/autoformer)                    |       ✅        |         ❌         |      ❌      |
+|                          [Bark](model_doc/bark)                          |       ✅        |         ❌         |      ❌      |
+|                          [BART](model_doc/bart)                          |       ✅        |         ✅         |      ✅      |
+|                       [BARThez](model_doc/barthez)                       |       ✅        |         ✅         |      ✅      |
+|                       [BARTpho](model_doc/bartpho)                       |       ✅        |         ✅         |      ✅      |
+|                          [BEiT](model_doc/beit)                          |       ✅        |         ❌         |      ✅      |
+|                          [BERT](model_doc/bert)                          |       ✅        |         ✅         |      ✅      |
+|               [Bert Generation](model_doc/bert-generation)               |       ✅        |         ❌         |      ❌      |
+|                 [BertJapanese](model_doc/bert-japanese)                  |       ✅        |         ✅         |      ✅      |
+|                      [BERTweet](model_doc/bertweet)                      |       ✅        |         ✅         |      ✅      |
+|                      [BigBird](model_doc/big_bird)                       |       ✅        |         ❌         |      ✅      |
+|               [BigBird-Pegasus](model_doc/bigbird_pegasus)               |       ✅        |         ❌         |      ❌      |
+|                        [BioGpt](model_doc/biogpt)                        |       ✅        |         ❌         |      ❌      |
+|                           [BiT](model_doc/bit)                           |       ✅        |         ❌         |      ❌      |
+|                    [Blenderbot](model_doc/blenderbot)                    |       ✅        |         ✅         |      ✅      |
+|              [BlenderbotSmall](model_doc/blenderbot-small)               |       ✅        |         ✅         |      ✅      |
+|                          [BLIP](model_doc/blip)                          |       ✅        |         ✅         |      ❌      |
+|                        [BLIP-2](model_doc/blip-2)                        |       ✅        |         ❌         |      ❌      |
+|                         [BLOOM](model_doc/bloom)                         |       ✅        |         ❌         |      ✅      |
+|                          [BORT](model_doc/bort)                          |       ✅        |         ✅         |      ✅      |
+|                   [BridgeTower](model_doc/bridgetower)                   |       ✅        |         ❌         |      ❌      |
+|                          [BROS](model_doc/bros)                          |       ✅        |         ❌         |      ❌      |
+|                          [ByT5](model_doc/byt5)                          |       ✅        |         ✅         |      ✅      |
+|                     [CamemBERT](model_doc/camembert)                     |       ✅        |         ✅         |      ❌      |
+|                        [CANINE](model_doc/canine)                        |       ✅        |         ❌         |      ❌      |
+|                  [Chinese-CLIP](model_doc/chinese_clip)                  |       ✅        |         ❌         |      ❌      |
+|                          [CLAP](model_doc/clap)                          |       ✅        |         ❌         |      ❌      |
+|                          [CLIP](model_doc/clip)                          |       ✅        |         ✅         |      ✅      |
+|                       [CLIPSeg](model_doc/clipseg)                       |       ✅        |         ❌         |      ❌      |
+|                       [CodeGen](model_doc/codegen)                       |       ✅        |         ❌         |      ❌      |
+|                    [CodeLlama](model_doc/code_llama)                     |       ✅        |         ❌         |      ❌      |
+|              [Conditional DETR](model_doc/conditional_detr)              |       ✅        |         ❌         |      ❌      |
+|                      [ConvBERT](model_doc/convbert)                      |       ✅        |         ✅         |      ❌      |
+|                      [ConvNeXT](model_doc/convnext)                      |       ✅        |         ✅         |      ❌      |
+|                    [ConvNeXTV2](model_doc/convnextv2)                    |       ✅        |         ❌         |      ❌      |
+|                           [CPM](model_doc/cpm)                           |       ✅        |         ✅         |      ✅      |
+|                       [CPM-Ant](model_doc/cpmant)                        |       ✅        |         ❌         |      ❌      |
+|                          [CTRL](model_doc/ctrl)                          |       ✅        |         ✅         |      ❌      |
+|                           [CvT](model_doc/cvt)                           |       ✅        |         ✅         |      ❌      |
+|                   [Data2VecAudio](model_doc/data2vec)                    |       ✅        |         ❌         |      ❌      |
+|                    [Data2VecText](model_doc/data2vec)                    |       ✅        |         ❌         |      ❌      |
+|                   [Data2VecVision](model_doc/data2vec)                   |       ✅        |         ✅         |      ❌      |
+|                       [DeBERTa](model_doc/deberta)                       |       ✅        |         ✅         |      ❌      |
+|                    [DeBERTa-v2](model_doc/deberta-v2)                    |       ✅        |         ✅         |      ❌      |
+|          [Decision Transformer](model_doc/decision_transformer)          |       ✅        |         ❌         |      ❌      |
+|               [Deformable DETR](model_doc/deformable_detr)               |       ✅        |         ❌         |      ❌      |
+|                          [DeiT](model_doc/deit)                          |       ✅        |         ✅         |      ❌      |
+|                        [DePlot](model_doc/deplot)                        |       ✅        |         ❌         |      ❌      |
+|                          [DETA](model_doc/deta)                          |       ✅        |         ❌         |      ❌      |
+|                          [DETR](model_doc/detr)                          |       ✅        |         ❌         |      ❌      |
+|                      [DialoGPT](model_doc/dialogpt)                      |       ✅        |         ✅         |      ✅      |
+|                         [DiNAT](model_doc/dinat)                         |       ✅        |         ❌         |      ❌      |
+|                        [DINOv2](model_doc/dinov2)                        |       ✅        |         ❌         |      ❌      |
+|                    [DistilBERT](model_doc/distilbert)                    |       ✅        |         ✅         |      ✅      |
+|                           [DiT](model_doc/dit)                           |       ✅        |         ❌         |      ✅      |
+|                       [DonutSwin](model_doc/donut)                       |       ✅        |         ❌         |      ❌      |
+|                           [DPR](model_doc/dpr)                           |       ✅        |         ✅         |      ❌      |
+|                           [DPT](model_doc/dpt)                           |       ✅        |         ❌         |      ❌      |
+|               [EfficientFormer](model_doc/efficientformer)               |       ✅        |         ✅         |      ❌      |
+|                  [EfficientNet](model_doc/efficientnet)                  |       ✅        |         ❌         |      ❌      |
+|                       [ELECTRA](model_doc/electra)                       |       ✅        |         ✅         |      ✅      |
+|                       [EnCodec](model_doc/encodec)                       |       ✅        |         ❌         |      ❌      |
+|               [Encoder decoder](model_doc/encoder-decoder)               |       ✅        |         ✅         |      ✅      |
+|                         [ERNIE](model_doc/ernie)                         |       ✅        |         ❌         |      ❌      |
+|                       [ErnieM](model_doc/ernie_m)                        |       ✅        |         ❌         |      ❌      |
+|                           [ESM](model_doc/esm)                           |       ✅        |         ✅         |      ❌      |
+|              [FairSeq Machine-Translation](model_doc/fsmt)               |       ✅        |         ❌         |      ❌      |
+|                        [Falcon](model_doc/falcon)                        |       ✅        |         ❌         |      ❌      |
+|                       [FLAN-T5](model_doc/flan-t5)                       |       ✅        |         ✅         |      ✅      |
+|                      [FLAN-UL2](model_doc/flan-ul2)                      |       ✅        |         ✅         |      ✅      |
+|                      [FlauBERT](model_doc/flaubert)                      |       ✅        |         ✅         |      ❌      |
+|                         [FLAVA](model_doc/flava)                         |       ✅        |         ❌         |      ❌      |
+|                          [FNet](model_doc/fnet)                          |       ✅        |         ❌         |      ❌      |
+|                      [FocalNet](model_doc/focalnet)                      |       ✅        |         ❌         |      ❌      |
+|                  [Funnel Transformer](model_doc/funnel)                  |       ✅        |         ✅         |      ❌      |
+|                          [Fuyu](model_doc/fuyu)                          |       ✅        |         ❌         |      ❌      |
+|                           [GIT](model_doc/git)                           |       ✅        |         ❌         |      ❌      |
+|                          [GLPN](model_doc/glpn)                          |       ✅        |         ❌         |      ❌      |
+|                       [GPT Neo](model_doc/gpt_neo)                       |       ✅        |         ❌         |      ✅      |
+|                      [GPT NeoX](model_doc/gpt_neox)                      |       ✅        |         ❌         |      ❌      |
+|             [GPT NeoX Japanese](model_doc/gpt_neox_japanese)             |       ✅        |         ❌         |      ❌      |
+|                         [GPT-J](model_doc/gptj)                          |       ✅        |         ✅         |      ✅      |
+|                       [GPT-Sw3](model_doc/gpt-sw3)                       |       ✅        |         ✅         |      ✅      |
+|                   [GPTBigCode](model_doc/gpt_bigcode)                    |       ✅        |         ❌         |      ❌      |
+|               [GPTSAN-japanese](model_doc/gptsan-japanese)               |       ✅        |         ❌         |      ❌      |
+|                    [Graphormer](model_doc/graphormer)                    |       ✅        |         ❌         |      ❌      |
+|                      [GroupViT](model_doc/groupvit)                      |       ✅        |         ✅         |      ❌      |
+|                       [HerBERT](model_doc/herbert)                       |       ✅        |         ✅         |      ✅      |
+|                        [Hubert](model_doc/hubert)                        |       ✅        |         ✅         |      ❌      |
+|                        [I-BERT](model_doc/ibert)                         |       ✅        |         ❌         |      ❌      |
+|                       [IDEFICS](model_doc/idefics)                       |       ✅        |         ❌         |      ❌      |
+|                      [ImageGPT](model_doc/imagegpt)                      |       ✅        |         ❌         |      ❌      |
+|                      [Informer](model_doc/informer)                      |       ✅        |         ❌         |      ❌      |
+|                  [InstructBLIP](model_doc/instructblip)                  |       ✅        |         ❌         |      ❌      |
+|                       [Jukebox](model_doc/jukebox)                       |       ✅        |         ❌         |      ❌      |
+|                      [LayoutLM](model_doc/layoutlm)                      |       ✅        |         ✅         |      ❌      |
+|                    [LayoutLMv2](model_doc/layoutlmv2)                    |       ✅        |         ❌         |      ❌      |
+|                    [LayoutLMv3](model_doc/layoutlmv3)                    |       ✅        |         ✅         |      ❌      |
+|                     [LayoutXLM](model_doc/layoutxlm)                     |       ✅        |         ❌         |      ❌      |
+|                           [LED](model_doc/led)                           |       ✅        |         ✅         |      ❌      |
+|                         [LeViT](model_doc/levit)                         |       ✅        |         ❌         |      ❌      |
+|                          [LiLT](model_doc/lilt)                          |       ✅        |         ❌         |      ❌      |
+|                         [LLaMA](model_doc/llama)                         |       ✅        |         ❌         |      ❌      |
+|                        [Llama2](model_doc/llama2)                        |       ✅        |         ❌         |      ❌      |
+|                    [Longformer](model_doc/longformer)                    |       ✅        |         ✅         |      ❌      |
+|                        [LongT5](model_doc/longt5)                        |       ✅        |         ❌         |      ✅      |
+|                          [LUKE](model_doc/luke)                          |       ✅        |         ❌         |      ❌      |
+|                        [LXMERT](model_doc/lxmert)                        |       ✅        |         ✅         |      ❌      |
+|                        [M-CTC-T](model_doc/mctct)                        |       ✅        |         ❌         |      ❌      |
+|                       [M2M100](model_doc/m2m_100)                        |       ✅        |         ❌         |      ❌      |
+|                        [Marian](model_doc/marian)                        |       ✅        |         ✅         |      ✅      |
+|                      [MarkupLM](model_doc/markuplm)                      |       ✅        |         ❌         |      ❌      |
+|                   [Mask2Former](model_doc/mask2former)                   |       ✅        |         ❌         |      ❌      |
+|                    [MaskFormer](model_doc/maskformer)                    |       ✅        |         ❌         |      ❌      |
+|                        [MatCha](model_doc/matcha)                        |       ✅        |         ❌         |      ❌      |
+|                         [mBART](model_doc/mbart)                         |       ✅        |         ✅         |      ✅      |
+|                      [mBART-50](model_doc/mbart50)                       |       ✅        |         ✅         |      ✅      |
+|                          [MEGA](model_doc/mega)                          |       ✅        |         ❌         |      ❌      |
+|                 [Megatron-BERT](model_doc/megatron-bert)                 |       ✅        |         ❌         |      ❌      |
+|                 [Megatron-GPT2](model_doc/megatron_gpt2)                 |       ✅        |         ✅         |      ✅      |
+|                       [MGP-STR](model_doc/mgp-str)                       |       ✅        |         ❌         |      ❌      |
+|                       [Mistral](model_doc/mistral)                       |       ✅        |         ❌         |      ❌      |
+|                         [mLUKE](model_doc/mluke)                         |       ✅        |         ❌         |      ❌      |
+|                           [MMS](model_doc/mms)                           |       ✅        |         ✅         |      ✅      |
+|                    [MobileBERT](model_doc/mobilebert)                    |       ✅        |         ✅         |      ❌      |
+|                  [MobileNetV1](model_doc/mobilenet_v1)                   |       ✅        |         ❌         |      ❌      |
+|                  [MobileNetV2](model_doc/mobilenet_v2)                   |       ✅        |         ❌         |      ❌      |
+|                     [MobileViT](model_doc/mobilevit)                     |       ✅        |         ✅         |      ❌      |
+|                   [MobileViTV2](model_doc/mobilevitv2)                   |       ✅        |         ❌         |      ❌      |
+|                         [MPNet](model_doc/mpnet)                         |       ✅        |         ✅         |      ❌      |
+|                           [MPT](model_doc/mpt)                           |       ✅        |         ❌         |      ❌      |
+|                           [MRA](model_doc/mra)                           |       ✅        |         ❌         |      ❌      |
+|                           [MT5](model_doc/mt5)                           |       ✅        |         ✅         |      ✅      |
+|                      [MusicGen](model_doc/musicgen)                      |       ✅        |         ❌         |      ❌      |
+|                           [MVP](model_doc/mvp)                           |       ✅        |         ❌         |      ❌      |
+|                           [NAT](model_doc/nat)                           |       ✅        |         ❌         |      ❌      |
+|                         [Nezha](model_doc/nezha)                         |       ✅        |         ❌         |      ❌      |
+|                          [NLLB](model_doc/nllb)                          |       ✅        |         ❌         |      ❌      |
+|                      [NLLB-MOE](model_doc/nllb-moe)                      |       ✅        |         ❌         |      ❌      |
+|                        [Nougat](model_doc/nougat)                        |       ✅        |         ✅         |      ✅      |
+|                 [Nyströmformer](model_doc/nystromformer)                 |       ✅        |         ❌         |      ❌      |
+|                     [OneFormer](model_doc/oneformer)                     |       ✅        |         ❌         |      ❌      |
+|                    [OpenAI GPT](model_doc/openai-gpt)                    |       ✅        |         ✅         |      ❌      |
+|                      [OpenAI GPT-2](model_doc/gpt2)                      |       ✅        |         ✅         |      ✅      |
+|                    [OpenLlama](model_doc/open-llama)                     |       ✅        |         ❌         |      ❌      |
+|                           [OPT](model_doc/opt)                           |       ✅        |         ✅         |      ✅      |
+|                       [OWL-ViT](model_doc/owlvit)                        |       ✅        |         ❌         |      ❌      |
+|                         [OWLv2](model_doc/owlv2)                         |       ✅        |         ❌         |      ❌      |
+|                       [Pegasus](model_doc/pegasus)                       |       ✅        |         ✅         |      ✅      |
+|                     [PEGASUS-X](model_doc/pegasus_x)                     |       ✅        |         ❌         |      ❌      |
+|                     [Perceiver](model_doc/perceiver)                     |       ✅        |         ❌         |      ❌      |
+|                     [Persimmon](model_doc/persimmon)                     |       ✅        |         ❌         |      ❌      |
+|                       [PhoBERT](model_doc/phobert)                       |       ✅        |         ✅         |      ✅      |
+|                    [Pix2Struct](model_doc/pix2struct)                    |       ✅        |         ❌         |      ❌      |
+|                        [PLBart](model_doc/plbart)                        |       ✅        |         ❌         |      ❌      |
+|                    [PoolFormer](model_doc/poolformer)                    |       ✅        |         ❌         |      ❌      |
+|                     [Pop2Piano](model_doc/pop2piano)                     |       ✅        |         ❌         |      ❌      |
+|                    [ProphetNet](model_doc/prophetnet)                    |       ✅        |         ❌         |      ❌      |
+|                           [PVT](model_doc/pvt)                           |       ✅        |         ❌         |      ❌      |
+|                       [QDQBert](model_doc/qdqbert)                       |       ✅        |         ❌         |      ❌      |
+|                           [RAG](model_doc/rag)                           |       ✅        |         ✅         |      ❌      |
+|                         [REALM](model_doc/realm)                         |       ✅        |         ❌         |      ❌      |
+|                      [Reformer](model_doc/reformer)                      |       ✅        |         ❌         |      ❌      |
+|                        [RegNet](model_doc/regnet)                        |       ✅        |         ✅         |      ✅      |
+|                       [RemBERT](model_doc/rembert)                       |       ✅        |         ✅         |      ❌      |
+|                        [ResNet](model_doc/resnet)                        |       ✅        |         ✅         |      ✅      |
+|                     [RetriBERT](model_doc/retribert)                     |       ✅        |         ❌         |      ❌      |
+|                       [RoBERTa](model_doc/roberta)                       |       ✅        |         ✅         |      ✅      |
+|          [RoBERTa-PreLayerNorm](model_doc/roberta-prelayernorm)          |       ✅        |         ✅         |      ✅      |
+|                      [RoCBert](model_doc/roc_bert)                       |       ✅        |         ❌         |      ❌      |
+|                      [RoFormer](model_doc/roformer)                      |       ✅        |         ✅         |      ✅      |
+|                          [RWKV](model_doc/rwkv)                          |       ✅        |         ❌         |      ❌      |
+|                           [SAM](model_doc/sam)                           |       ✅        |         ✅         |      ❌      |
+|                     [SegFormer](model_doc/segformer)                     |       ✅        |         ✅         |      ❌      |
+|                           [SEW](model_doc/sew)                           |       ✅        |         ❌         |      ❌      |
+|                         [SEW-D](model_doc/sew-d)                         |       ✅        |         ❌         |      ❌      |
+|        [Speech Encoder decoder](model_doc/speech-encoder-decoder)        |       ✅        |         ❌         |      ✅      |
+|                 [Speech2Text](model_doc/speech_to_text)                  |       ✅        |         ✅         |      ❌      |
+|                      [SpeechT5](model_doc/speecht5)                      |       ✅        |         ❌         |      ❌      |
+|                      [Splinter](model_doc/splinter)                      |       ✅        |         ❌         |      ❌      |
+|                   [SqueezeBERT](model_doc/squeezebert)                   |       ✅        |         ❌         |      ❌      |
+|                   [SwiftFormer](model_doc/swiftformer)                   |       ✅        |         ❌         |      ❌      |
+|                    [Swin Transformer](model_doc/swin)                    |       ✅        |         ✅         |      ❌      |
+|                 [Swin Transformer V2](model_doc/swinv2)                  |       ✅        |         ❌         |      ❌      |
+|                       [Swin2SR](model_doc/swin2sr)                       |       ✅        |         ❌         |      ❌      |
+|           [SwitchTransformers](model_doc/switch_transformers)            |       ✅        |         ❌         |      ❌      |
+|                            [T5](model_doc/t5)                            |       ✅        |         ✅         |      ✅      |
+|                        [T5v1.1](model_doc/t5v1.1)                        |       ✅        |         ✅         |      ✅      |
+|             [Table Transformer](model_doc/table-transformer)             |       ✅        |         ❌         |      ❌      |
+|                         [TAPAS](model_doc/tapas)                         |       ✅        |         ✅         |      ❌      |
+|                         [TAPEX](model_doc/tapex)                         |       ✅        |         ✅         |      ✅      |
+|       [Time Series Transformer](model_doc/time_series_transformer)       |       ✅        |         ❌         |      ❌      |
+|                   [TimeSformer](model_doc/timesformer)                   |       ✅        |         ❌         |      ❌      |
+|        [Trajectory Transformer](model_doc/trajectory_transformer)        |       ✅        |         ❌         |      ❌      |
+|                  [Transformer-XL](model_doc/transfo-xl)                  |       ✅        |         ✅         |      ❌      |
+|                         [TrOCR](model_doc/trocr)                         |       ✅        |         ❌         |      ❌      |
+|                          [TVLT](model_doc/tvlt)                          |       ✅        |         ❌         |      ❌      |
+|                           [UL2](model_doc/ul2)                           |       ✅        |         ✅         |      ✅      |
+|                          [UMT5](model_doc/umt5)                          |       ✅        |         ❌         |      ❌      |
+|                     [UniSpeech](model_doc/unispeech)                     |       ✅        |         ❌         |      ❌      |
+|                 [UniSpeechSat](model_doc/unispeech-sat)                  |       ✅        |         ❌         |      ❌      |
+|                       [UPerNet](model_doc/upernet)                       |       ✅        |         ❌         |      ❌      |
+|                           [VAN](model_doc/van)                           |       ✅        |         ❌         |      ❌      |
+|                      [VideoMAE](model_doc/videomae)                      |       ✅        |         ❌         |      ❌      |
+|                          [ViLT](model_doc/vilt)                          |       ✅        |         ❌         |      ❌      |
+|        [Vision Encoder decoder](model_doc/vision-encoder-decoder)        |       ✅        |         ✅         |      ✅      |
+|       [VisionTextDualEncoder](model_doc/vision-text-dual-encoder)        |       ✅        |         ✅         |      ✅      |
+|                   [VisualBERT](model_doc/visual_bert)                    |       ✅        |         ❌         |      ❌      |
+|                           [ViT](model_doc/vit)                           |       ✅        |         ✅         |      ✅      |
+|                    [ViT Hybrid](model_doc/vit_hybrid)                    |       ✅        |         ❌         |      ❌      |
+|                        [VitDet](model_doc/vitdet)                        |       ✅        |         ❌         |      ❌      |
+|                       [ViTMAE](model_doc/vit_mae)                        |       ✅        |         ✅         |      ❌      |
+|                      [ViTMatte](model_doc/vitmatte)                      |       ✅        |         ❌         |      ❌      |
+|                       [ViTMSN](model_doc/vit_msn)                        |       ✅        |         ❌         |      ❌      |
+|                          [VITS](model_doc/vits)                          |       ✅        |         ❌         |      ❌      |
+|                         [ViViT](model_doc/vivit)                         |       ✅        |         ❌         |      ❌      |
+|                      [Wav2Vec2](model_doc/wav2vec2)                      |       ✅        |         ✅         |      ✅      |
+|            [Wav2Vec2-Conformer](model_doc/wav2vec2-conformer)            |       ✅        |         ❌         |      ❌      |
+|              [Wav2Vec2Phoneme](model_doc/wav2vec2_phoneme)               |       ✅        |         ✅         |      ✅      |
+|                         [WavLM](model_doc/wavlm)                         |       ✅        |         ❌         |      ❌      |
+|                       [Whisper](model_doc/whisper)                       |       ✅        |         ✅         |      ✅      |
+|                        [X-CLIP](model_doc/xclip)                         |       ✅        |         ❌         |      ❌      |
+|                         [X-MOD](model_doc/xmod)                          |       ✅        |         ❌         |      ❌      |
+|                          [XGLM](model_doc/xglm)                          |       ✅        |         ✅         |      ✅      |
+|                           [XLM](model_doc/xlm)                           |       ✅        |         ✅         |      ❌      |
+|                [XLM-ProphetNet](model_doc/xlm-prophetnet)                |       ✅        |         ❌         |      ❌      |
+|                   [XLM-RoBERTa](model_doc/xlm-roberta)                   |       ✅        |         ✅         |      ✅      |
+|                [XLM-RoBERTa-XL](model_doc/xlm-roberta-xl)                |       ✅        |         ❌         |      ❌      |
+|                         [XLM-V](model_doc/xlm-v)                         |       ✅        |         ✅         |      ✅      |
+|                         [XLNet](model_doc/xlnet)                         |       ✅        |         ✅         |      ❌      |
+|                         [XLS-R](model_doc/xls_r)                         |       ✅        |         ✅         |      ✅      |
+|                 [XLSR-Wav2Vec2](model_doc/xlsr_wav2vec2)                 |       ✅        |         ✅         |      ✅      |
+|                         [YOLOS](model_doc/yolos)                         |       ✅        |         ❌         |      ❌      |
+|                          [YOSO](model_doc/yoso)                          |       ✅        |         ❌         |      ❌      |
 
 <!-- End table-->

docs/source/en/installation.md

@@ -169,28 +169,28 @@ Pretrained models are downloaded and locally cached at: `~/.cache/huggingface/hu
 
 ## Offline mode
 
-🤗 Transformers is able to run in a firewalled or offline environment by only using local files. Set the environment variable `TRANSFORMERS_OFFLINE=1` to enable this behavior.
+Run 🤗 Transformers in a firewalled or offline environment with locally cached files by setting the environment variable `TRANSFORMERS_OFFLINE=1`.
 
 <Tip>
 
-Add [🤗 Datasets](https://huggingface.co/docs/datasets/) to your offline training workflow by setting the environment variable `HF_DATASETS_OFFLINE=1`.
+Add [🤗 Datasets](https://huggingface.co/docs/datasets/) to your offline training workflow with the environment variable `HF_DATASETS_OFFLINE=1`.
 
 </Tip>
 
-For example, you would typically run a program on a normal network firewalled to external instances with the following command:
-
-```bash
-python examples/pytorch/translation/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
-```
-
-Run this same program in an offline instance with:
-
 ```bash
 HF_DATASETS_OFFLINE=1 TRANSFORMERS_OFFLINE=1 \
 python examples/pytorch/translation/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
 ```
 
-The script should now run without hanging or waiting to timeout because it knows it should only look for local files.
+This script should run without hanging or waiting to timeout because it won't attempt to download the model from the Hub.
+
+You can also bypass loading a model from the Hub from each [`~PreTrainedModel.from_pretrained`] call with the [`local_files_only`] parameter. When set to `True`, only local files are loaded:
+
+```py
+from transformers import T5Model
+
+model = T5Model.from_pretrained("./path/to/local/directory", local_files_only=True)
+```
 
 ### Fetch models and tokenizers to use offline
 

docs/source/en/llm_tutorial.md

@@ -27,7 +27,7 @@ This tutorial will show you how to:
 
 * Generate text with an LLM
 * Avoid common pitfalls
-* Next steps to help you get the most out your LLM
+* Next steps to help you get the most out of your LLM
 
 Before you begin, make sure you have all the necessary libraries installed:
 
@@ -74,14 +74,13 @@ If you're interested in basic LLM usage, our high-level [`Pipeline`](pipeline_tu
 
 </Tip>
 
-<!-- TODO: update example to llama 2 (or a newer popular baseline) when it becomes ungated -->
 First, you need to load the model.
 
 ```py
 >>> from transformers import AutoModelForCausalLM
 
 >>> model = AutoModelForCausalLM.from_pretrained(
-...     "openlm-research/open_llama_7b", device_map="auto", load_in_4bit=True
+...     "mistralai/Mistral-7B-v0.1", device_map="auto", load_in_4bit=True
 ... )
 ```
 
@@ -97,18 +96,31 @@ Next, you need to preprocess your text input with a [tokenizer](tokenizer_summar
 ```py
 >>> from transformers import AutoTokenizer
 
->>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1", padding_side="left")
 >>> model_inputs = tokenizer(["A list of colors: red, blue"], return_tensors="pt").to("cuda")
 ```
 
 The `model_inputs` variable holds the tokenized text input, as well as the attention mask. While [`~generation.GenerationMixin.generate`] does its best effort to infer the attention mask when it is not passed, we recommend passing it whenever possible for optimal results.
 
-Finally, call the [`~generation.GenerationMixin.generate`] method to returns the generated tokens, which should be converted to text before printing.
+After tokenizing the inputs, you can call the [`~generation.GenerationMixin.generate`] method to returns the generated tokens. The generated tokens then should be converted to text before printing.
 
 ```py
 >>> generated_ids = model.generate(**model_inputs)
 >>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'A list of colors: red, blue, green, yellow, black, white, and brown'
+'A list of colors: red, blue, green, yellow, orange, purple, pink,'
+```
+
+Finally, you don't need to do it one sequence at a time! You can batch your inputs, which will greatly improve the throughput at a small latency and memory cost. All you need to do is to make sure you pad your inputs properly (more on that below).
+
+```py
+>>> tokenizer.pad_token = tokenizer.eos_token  # Most LLMs don't have a pad token by default
+>>> model_inputs = tokenizer(
+...     ["A list of colors: red, blue", "Portugal is"], return_tensors="pt", padding=True
+... ).to("cuda")
+>>> generated_ids = model.generate(**model_inputs)
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
+['A list of colors: red, blue, green, yellow, orange, purple, pink,',
+'Portugal is a country in southwestern Europe, on the Iber']
 ```
 
 And that's it! In a few lines of code, you can harness the power of an LLM.
@@ -121,10 +133,10 @@ There are many [generation strategies](generation_strategies), and sometimes the
 ```py
 >>> from transformers import AutoModelForCausalLM, AutoTokenizer
 
->>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
->>> tokenizer.pad_token = tokenizer.eos_token  # Llama has no pad token by default
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
+>>> tokenizer.pad_token = tokenizer.eos_token  # Most LLMs don't have a pad token by default
 >>> model = AutoModelForCausalLM.from_pretrained(
-...     "openlm-research/open_llama_7b", device_map="auto", load_in_4bit=True
+...     "mistralai/Mistral-7B-v0.1", device_map="auto", load_in_4bit=True
 ... )
 ```
 
@@ -154,7 +166,7 @@ By default, and unless specified in the [`~generation.GenerationConfig`] file, `
 ```py
 >>> # Set seed or reproducibility -- you don't need this unless you want full reproducibility
 >>> from transformers import set_seed
->>> set_seed(0)
+>>> set_seed(42)
 
 >>> model_inputs = tokenizer(["I am a cat."], return_tensors="pt").to("cuda")
 
@@ -166,7 +178,7 @@ By default, and unless specified in the [`~generation.GenerationConfig`] file, `
 >>> # With sampling, the output becomes more creative!
 >>> generated_ids = model.generate(**model_inputs, do_sample=True)
 >>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'I am a cat.\nI just need to be. I am always.\nEvery time'
+'I am a cat.  Specifically, I am an indoor-only cat.  I'
 ```
 
 ### Wrong padding side
@@ -175,17 +187,17 @@ LLMs are [decoder-only](https://huggingface.co/learn/nlp-course/chapter1/6?fw=pt
 
 ```py
 >>> # The tokenizer initialized above has right-padding active by default: the 1st sequence,
->>> # which is shorter, has padding on the right side. Generation fails.
+>>> # which is shorter, has padding on the right side. Generation fails to capture the logic.
 >>> model_inputs = tokenizer(
 ...     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
 ... ).to("cuda")
 >>> generated_ids = model.generate(**model_inputs)
->>> tokenizer.batch_decode(generated_ids[0], skip_special_tokens=True)[0]
-''
+>>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+'1, 2, 33333333333'
 
 >>> # With left-padding, it works as expected!
->>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b", padding_side="left")
->>> tokenizer.pad_token = tokenizer.eos_token  # Llama has no pad token by default
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1", padding_side="left")
+>>> tokenizer.pad_token = tokenizer.eos_token  # Most LLMs don't have a pad token by default
 >>> model_inputs = tokenizer(
 ...     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
 ... ).to("cuda")
@@ -194,26 +206,61 @@ LLMs are [decoder-only](https://huggingface.co/learn/nlp-course/chapter1/6?fw=pt
 '1, 2, 3, 4, 5, 6,'
 ```
 
-<!-- TODO: when the prompting guide is ready, mention the importance of setting the right prompt in this section -->
+### Wrong prompt
+
+Some models and tasks expect a certain input prompt format to work properly. When this format is not applied, you will get a silent performance degradation: the model kinda works, but not as well as if you were following the expected prompt. More information about prompting, including which models and tasks need to be careful, is available in this [guide](tasks/prompting). Let's see an example with a chat LLM, which makes use of [chat templating](chat_templating):
+
+```python
+>>> tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-alpha")
+>>> model = AutoModelForCausalLM.from_pretrained(
+...     "HuggingFaceH4/zephyr-7b-alpha", device_map="auto", load_in_4bit=True
+... )
+>>> set_seed(0)
+>>> prompt = """How many helicopters can a human eat in one sitting? Reply as a thug."""
+>>> model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
+>>> input_length = model_inputs.input_ids.shape[1]
+>>> generated_ids = model.generate(**model_inputs, max_new_tokens=20)
+>>> print(tokenizer.batch_decode(generated_ids[:, input_length:], skip_special_tokens=True)[0])
+"I'm not a thug, but i can tell you that a human cannot eat"
+>>> # Oh no, it did not follow our instruction to reply as a thug! Let's see what happens when we write
+>>> # a better prompt and use the right template for this model (through `tokenizer.apply_chat_template`)
+
+>>> set_seed(0)
+>>> messages = [
+...     {
+...         "role": "system",
+...         "content": "You are a friendly chatbot who always responds in the style of a thug",
+...     },
+...     {"role": "user", "content": "How many helicopters can a human eat in one sitting?"},
+... ]
+>>> model_inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to("cuda")
+>>> input_length = model_inputs.shape[1]
+>>> generated_ids = model.generate(model_inputs, do_sample=True, max_new_tokens=20)
+>>> print(tokenizer.batch_decode(generated_ids[:, input_length:], skip_special_tokens=True)[0])
+'None, you thug. How bout you try to focus on more useful questions?'
+>>> # As we can see, it followed a proper thug style 😎
+```
 
 ## Further resources
 
 While the autoregressive generation process is relatively straightforward, making the most out of your LLM can be a challenging endeavor because there are many moving parts. For your next steps to help you dive deeper into LLM usage and understanding:
 
-<!-- TODO: complete with new guides -->
 ### Advanced generate usage
 
 1. [Guide](generation_strategies) on how to control different generation methods, how to set up the generation configuration file, and how to stream the output;
-2. API reference on [`~generation.GenerationConfig`], [`~generation.GenerationMixin.generate`], and [generate-related classes](internal/generation_utils).
+2. [Guide](chat_templating) on the prompt template for chat LLMs;
+3. [Guide](tasks/prompting) on to get the most of prompt design;
+4. API reference on [`~generation.GenerationConfig`], [`~generation.GenerationMixin.generate`], and [generate-related classes](internal/generation_utils).
 
 ### LLM leaderboards
 
 1. [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard), which focuses on the quality of the open-source models;
 2. [Open LLM-Perf Leaderboard](https://huggingface.co/spaces/optimum/llm-perf-leaderboard), which focuses on LLM throughput.
 
-### Latency and throughput
+### Latency, throughput and memory utilization
 
-1. [Guide](main_classes/quantization) on dynamic quantization, which shows you how to drastically reduce your memory requirements.
+1. [Guide](llm_tutorial_optimization) on how to optimize LLMs for speed and memory;
+2. [Guide](main_classes/quantization) on quantization such as bitsandbytes and autogptq, which shows you how to drastically reduce your memory requirements.
 
 ### Related libraries
 

docs/source/en/llm_tutorial_optimization.md

@@ -0,0 +1,739 @@
+<!--Copyright 2023 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.
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+-->
+# Optimizing LLMs for Speed and Memory
+
+[[open-in-colab]]
+
+Large Language Models (LLMs) such as GPT3/4, [Falcon](https://huggingface.co/tiiuae/falcon-40b), and [Llama](https://huggingface.co/meta-llama/Llama-2-70b-hf) are rapidly advancing in their ability to tackle human-centric tasks, establishing themselves as essential tools in modern knowledge-based industries.
+Deploying these models in real-world tasks remains challenging, however:
+
+-   To exhibit near-human text understanding and generation capabilities, LLMs currently require to be composed of billions of parameters (see [Kaplan et al](https://arxiv.org/abs/2001.08361), [Wei et. al](https://arxiv.org/abs/2206.07682)). This consequently amplifies the memory demands for inference.
+-   In many real-world tasks, LLMs need to be given extensive contextual information. This necessitates the model's capability to manage very long input sequences during inference.
+
+The crux of these challenges lies in augmenting the computational and memory capabilities of LLMs, especially when handling expansive input sequences.
+
+In this guide, we will go over the effective techniques for efficient LLM deployment:
+
+1.  **Lower Precision**: Research has shown that operating at reduced numerical precision, namely [8-bit and 4-bit](./main_classes/quantization.md) can achieve computational advantages without a considerable decline in model performance.
+
+2.  **Flash Attention:** Flash Attention is a variation of the attention algorithm that not only provides a more memory-efficient approach but also realizes increased efficiency due to optimized GPU memory utilization.
+
+3.  **Architectural Innovations:** Considering that LLMs are always deployed in the same way during inference, namely autoregressive text generation with a long input context, specialized model architectures have been proposed that allow for more efficient inference. The most important advancement in model architectures hereby are [Alibi](https://arxiv.org/abs/2108.12409), [Rotary embeddings](https://arxiv.org/abs/2104.09864), [Multi-Query Attention (MQA)](https://arxiv.org/abs/1911.02150) and [Grouped-Query-Attention (GQA)]((https://arxiv.org/abs/2305.13245)).
+
+Throughout this guide, we will offer an analysis of auto-regressive generation from a tensor's perspective. We delve into the pros and cons of adopting lower precision, provide a comprehensive exploration of the latest attention algorithms, and discuss improved LLM architectures. While doing so, we run practical examples showcasing each of the feature improvements.
+
+## 1. Lower Precision
+
+Memory requirements of LLMs can be best understood by seeing the LLM as a set of weight matrices and vectors and the text inputs as a sequence of vectors. In the following, the definition *weights* will be used to signify all model weight matrices and vectors.
+
+At the time of writing this guide, LLMs consist of at least a couple billion parameters. Each parameter thereby is made of a decimal number, e.g. `4.5689` which is usually stored in either [float32](https://en.wikipedia.org/wiki/Single-precision_floating-point_format), [bfloat16](https://en.wikipedia.org/wiki/Bfloat16_floating-point_format), or [float16](https://en.wikipedia.org/wiki/Half-precision_floating-point_format) format. This allows us to easily compute the memory requirement to load the LLM into memory:
+
+> *Loading the weights of a model having X billion parameters requires roughly 4 * X GB of VRAM in float32 precision*
+
+Nowadays, models are however rarely trained in full float32 precision, but usually in bfloat16 precision or less frequently in float16 precision. Therefore the rule of thumb becomes:
+
+> *Loading the weights of a model having X billion parameters requires roughly 2 * X GB of VRAM in bfloat16/float16 precision*
+
+For shorter text inputs (less than 1024 tokens), the memory requirement for inference is very much dominated by the memory requirement to load the weights. Therefore, for now, let's assume that the memory requirement for inference is equal to the memory requirement to load the model into the GPU VRAM.
+
+To give some examples of how much VRAM it roughly takes to load a model in bfloat16:
+
+-   **GPT3** requires 2 \* 175 GB = **350 GB** VRAM
+-   [**Bloom**](https://huggingface.co/bigscience/bloom) requires 2 \* 176 GB = **352 GB** VRAM
+-   [**Llama-2-70b**](https://huggingface.co/meta-llama/Llama-2-70b-hf) requires 2 \* 70 GB = **140 GB** VRAM
+-   [**Falcon-40b**](https://huggingface.co/tiiuae/falcon-40b) requires 2 \* 40 GB = **80 GB** VRAM
+-   [**MPT-30b**](https://huggingface.co/mosaicml/mpt-30b) requires 2 \* 30 GB = **60 GB** VRAM
+-   [**bigcode/starcoder**](https://huggingface.co/bigcode/starcoder) requires 2 \* 15.5 = **31 GB** VRAM
+
+As of writing this document, the largest GPU chip on the market is the A100 & H100 offering 80GB of VRAM. Most of the models listed before require more than 80GB just to be loaded and therefore necessarily require [tensor parallelism](https://huggingface.co/docs/transformers/perf_train_gpu_many#tensor-parallelism) and/or [pipeline parallelism](https://huggingface.co/docs/transformers/perf_train_gpu_many#naive-model-parallelism-vertical-and-pipeline-parallelism).
+
+🤗 Transformers does not support tensor parallelism out of the box as it requires the model architecture to be written in a specific way. If you're interested in writing models in a tensor-parallelism-friendly way, feel free to have a look at [the text-generation-inference library](https://github.com/huggingface/text-generation-inference/tree/main/server/text_generation_server/models/custom_modeling).
+
+Naive pipeline parallelism is supported out of the box. For this, simply load the model with `device="auto"` which will automatically place the different layers on the available GPUs as explained [here](https://huggingface.co/docs/accelerate/v0.22.0/en/concept_guides/big_model_inference).
+Note, however that while very effective, this naive pipeline parallelism does not tackle the issues of GPU idling. For this more advanced pipeline parallelism is required as explained [here](https://huggingface.co/docs/transformers/v4.34.0/en/perf_train_gpu_many#naive-model-parallelism-vertical-and-pipeline-parallelism).
+
+If you have access to an 8 x 80GB A100 node, you could load BLOOM as follows
+
+```bash
+!pip install transformers accelerate bitsandbytes optimum
+```
+```python
+from transformers import AutoModelForCausalLM
+
+model = AutoModelForCausalLM.from_pretrained("bigscience/bloom", device_map="auto", pad_token_id=0)
+```
+
+By using `device_map="auto"` the attention layers would be equally distributed over all available GPUs.
+
+In this guide, we will use [bigcode/octocoder](https://huggingface.co/bigcode/octocoder) as it can be run on a single 40 GB A100 GPU device chip. Note that all memory and speed optimizations that we will apply going forward, are equally applicable to models that require model or tensor parallelism.
+
+Since the model is loaded in bfloat16 precision, using our rule of thumb above, we would expect the memory requirement to run inference with `bigcode/octocoder` to be around 31 GB VRAM. Let's give it a try.
+
+We first load the model and tokenizer and then pass both to Transformers' [pipeline](https://huggingface.co/docs/transformers/main_classes/pipelines) object.
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+import torch
+
+model = AutoModelForCausalLM.from_pretrained("bigcode/octocoder", torch_dtype=torch.bfloat16, device_map="auto", pad_token_id=0)
+tokenizer = AutoTokenizer.from_pretrained("bigcode/octocoder")
+
+pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
+```
+
+```python
+prompt = "Question: Please write a function in Python that transforms bytes to Giga bytes.\n\nAnswer:"
+
+result = pipe(prompt, max_new_tokens=60)[0]["generated_text"][len(prompt):]
+result
+```
+
+**Output**:
+```
+Here is a Python function that transforms bytes to Giga bytes:\n\n```python\ndef bytes_to_giga_bytes(bytes):\n    return bytes / 1024 / 1024 / 1024\n```\n\nThis function takes a single
+```
+
+Nice, we can now directly use the result to convert bytes into Gigabytes.
+
+```python
+def bytes_to_giga_bytes(bytes):
+  return bytes / 1024 / 1024 / 1024
+```
+
+Let's call [`torch.cuda.max_memory_allocated`](https://pytorch.org/docs/stable/generated/torch.cuda.max_memory_allocated.html) to measure the peak GPU memory allocation.
+
+```python
+bytes_to_giga_bytes(torch.cuda.max_memory_allocated())
+```
+
+**Output**:
+```bash
+29.0260648727417
+```
+
+Close enough to our back-of-the-envelope computation! We can see the number is not exactly correct as going from bytes to kilobytes requires a multiplication of 1024 instead of 1000. Therefore the back-of-the-envelope formula can also be understood as an "at most X GB" computation.
+Note that if we had tried to run the model in full float32 precision, a whopping 64 GB of VRAM would have been required.
+
+> Almost all models are trained in bfloat16 nowadays, there is no reason to run the model in full float32 precision if [your GPU supports bfloat16](https://discuss.pytorch.org/t/bfloat16-native-support/117155/5). Float32 won't give better inference results than the precision that was used to train the model.
+
+If you are unsure in which format the model weights are stored on the Hub, you can always look into the checkpoint's config under `"torch_dtype"`, *e.g.* [here](https://huggingface.co/meta-llama/Llama-2-7b-hf/blob/6fdf2e60f86ff2481f2241aaee459f85b5b0bbb9/config.json#L21). It is recommended to set the model to the same precision type as written in the config when loading with `from_pretrained(..., torch_dtype=...)` except when the original type is float32 in which case one can use both `float16` or `bfloat16` for inference.
+
+
+Let's define a `flush(...)` function to free all allocated memory so that we can accurately measure the peak allocated GPU memory.
+
+```python
+del pipe
+del model
+
+import gc
+import torch
+
+def flush():
+  gc.collect()
+  torch.cuda.empty_cache()
+  torch.cuda.reset_peak_memory_stats()
+```
+
+Let's call it now for the next experiment.
+
+```python
+flush()
+```
+In the recent version of the accelerate library, you can also use an utility method called `release_memory()`
+
+```python
+from accelerate.utils import release_memory
+# ...
+
+release_memory(model)
+```
+
+Now what if your GPU does not have 32 GB of VRAM? It has been found that model weights can be quantized to 8-bit or 4-bits without a significant loss in performance (see [Dettmers et al.](https://arxiv.org/abs/2208.07339)).
+Model can be quantized to even 3 or 2 bits with an acceptable loss in performance as shown in the recent [GPTQ paper](https://arxiv.org/abs/2210.17323) 🤯.
+
+Without going into too many details, quantization schemes aim at reducing the precision of weights while trying to keep the model's inference results as accurate as possible (*a.k.a* as close as possible to bfloat16).
+Note that quantization works especially well for text generation since all we care about is choosing the *set of most likely next tokens* and don't really care about the exact values of the next token *logit* distribution.
+All that matters is that the next token *logit* distribution stays roughly the same so that an `argmax` or `topk` operation gives the same results.
+
+There are various quantization techniques, which we won't discuss in detail here, but in general, all quantization techniques work as follows:
+
+-   1.  Quantize all weights to the target precision
+-   2.  Load the quantized weights, and pass the input sequence of vectors in bfloat16 precision
+-   3.  Dynamically dequantize weights to bfloat16 to perform the computation with their input vectors in bfloat16 precision
+
+In a nutshell, this means that *inputs-weight matrix* multiplications, with \\( X \\) being the *inputs*, \\( W \\) being a weight matrix and \\( Y \\) being the output:
+
+$$ Y = X * W $$
+
+are changed to
+
+$$ Y = X * \text{dequantize}(W) $$
+
+for every matrix multiplication. Dequantization and re-quantization is performed sequentially for all weight matrices as the inputs run through the network graph.
+
+Therefore, inference time is often **not** reduced when using quantized weights, but rather increases.
+Enough theory, let's give it a try! To quantize the weights with Transformers, you need to make sure that
+the [`bitsandbytes`](https://github.com/TimDettmers/bitsandbytes) library is installed.
+
+```bash
+!pip install bitsandbytes
+```
+
+We can then load models in 8-bit quantization by simply adding a `load_in_8bit=True` flag to `from_pretrained`.
+
+```python
+model = AutoModelForCausalLM.from_pretrained("bigcode/octocoder", load_in_8bit=True, pad_token_id=0)
+```
+
+Now, let's run our example again and measure the memory usage.
+
+```python
+pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
+
+result = pipe(prompt, max_new_tokens=60)[0]["generated_text"][len(prompt):]
+result
+```
+
+**Output**:
+```
+Here is a Python function that transforms bytes to Giga bytes:\n\n```python\ndef bytes_to_giga_bytes(bytes):\n    return bytes / 1024 / 1024 / 1024\n```\n\nThis function takes a single
+```
+
+Nice, we're getting the same result as before, so no loss in accuracy! Let's look at how much memory was used this time.
+
+```python
+bytes_to_giga_bytes(torch.cuda.max_memory_allocated())
+```
+
+**Output**:
+```
+15.219234466552734
+```
+
+Significantly less! We're down to just a bit over 15 GBs and could therefore run this model on consumer GPUs like the 4090.
+We're seeing a very nice gain in memory efficiency and more or less no degradation to the model's output. However, we can also notice a slight slow-down during inference.
+
+
+We delete the models and flush the memory again.
+```python
+del model
+del pipe
+```
+
+```python
+flush()
+```
+
+Let's see what peak GPU memory consumption 4-bit quantization gives. Quantizing the model to 4-bit can be done with the same API as before - this time by passing `load_in_4bit=True` instead of `load_in_8bit=True`.
+
+```python
+model = AutoModelForCausalLM.from_pretrained("bigcode/octocoder", load_in_4bit=True, low_cpu_mem_usage=True, pad_token_id=0)
+
+pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
+
+result = pipe(prompt, max_new_tokens=60)[0]["generated_text"][len(prompt):]
+result
+```
+
+**Output**:
+```
+Here is a Python function that transforms bytes to Giga bytes:\n\n```\ndef bytes_to_gigabytes(bytes):\n    return bytes / 1024 / 1024 / 1024\n```\n\nThis function takes a single argument
+```
+
+We're almost seeing the same output text as before - just the `python` is missing just before the code snippet. Let's see how much memory was required.
+
+```python
+bytes_to_giga_bytes(torch.cuda.max_memory_allocated())
+```
+
+**Output**:
+```
+9.543574333190918
+```
+
+Just 9.5GB! That's really not a lot for a >15 billion parameter model.
+
+While we see very little degradation in accuracy for our model here, 4-bit quantization can in practice often lead to different results compared to 8-bit quantization or full `bfloat16` inference. It is up to the user to try it out.
+
+Also note that inference here was again a bit slower compared to 8-bit quantization which is due to the more aggressive quantization method used for 4-bit quantization leading to \\( \text{quantize} \\) and \\( \text{dequantize} \\) taking longer during inference.
+
+```python
+del model
+del pipe
+```
+```python
+flush()
+```
+
+Overall, we saw that running OctoCoder in 8-bit precision reduced the required GPU VRAM from 32G GPU VRAM to only 15GB and running the model in 4-bit precision further reduces the required GPU VRAM to just a bit over 9GB.
+
+4-bit quantization allows the model to be run on GPUs such as RTX3090, V100, and T4 which are quite accessible for most people.
+
+For more information on quantization and to see how one can quantize models to require even less GPU VRAM memory than 4-bit, we recommend looking into the [`AutoGPTQ`](https://huggingface.co/docs/transformers/main/en/main_classes/quantization#autogptq-integration%60) implementation.
+
+> As a conclusion, it is important to remember that model quantization trades improved memory efficiency against accuracy and in some cases inference time.
+
+If GPU memory is not a constraint for your use case, there is often no need to look into quantization. However many GPUs simply can't run LLMs without quantization methods and in this case, 4-bit and 8-bit quantization schemes are extremely useful tools.
+
+For more in-detail usage information, we strongly recommend taking a look at the [Transformers Quantization Docs](https://huggingface.co/docs/transformers/main_classes/quantization#general-usage).
+Next, let's look into how we can improve computational and memory efficiency by using better algorithms and an improved model architecture.
+
+# 2. Flash Attention
+
+Today's top-performing LLMs share more or less the same fundamental architecture that consists of feed-forward layers, activation layers, layer normalization layers, and most crucially, self-attention layers.
+
+Self-attention layers are central to Large Language Models (LLMs) in that they enable the model to understand the contextual relationships between input tokens.
+However, the peak GPU memory consumption for self-attention layers grows *quadratically* both in compute and memory complexity with number of input tokens (also called *sequence length*) that we denote in the following by \\( N \\) .
+While this is not really noticeable for shorter input sequences (of up to 1000 input tokens), it becomes a serious problem for longer input sequences (at around 16000 input tokens).
+
+Let's take a closer look. The formula to compute the output \\( \mathbf{O} \\) of a self-attention layer for an input \\( \mathbf{X} \\) of length \\( N \\) is:
+
+$$ \textbf{O} = \text{Attn}(\mathbf{X}) = \mathbf{V} \times \text{Softmax}(\mathbf{QK}^T) \text{ with } \mathbf{Q} = \mathbf{W}_q \mathbf{X}, \mathbf{V} = \mathbf{W}_v \mathbf{X}, \mathbf{K} = \mathbf{W}_k \mathbf{X} $$
+
+\\(  \mathbf{X} = (\mathbf{x}_1, ... \mathbf{x}_{N}) \\) is thereby the input sequence to the attention layer. The projections \\( \mathbf{Q} \\) and \\( \mathbf{K} \\) will each consist of \\( N \\) vectors resulting in the \\( \mathbf{QK}^T \\) being of size \\( N^2 \\) .
+
+LLMs usually have multiple attention heads, thus doing multiple self-attention computations in parallel.
+Assuming, the LLM has 40 attention heads and runs in bfloat16 precision, we can calculate the memory requirement to store the \\( \mathbf{QK^T} \\) matrices to be \\( 40 * 2 * N^2 \\) bytes. For \\( N=1000 \\) only around 50 MB of VRAM are needed, however, for \\( N=16000 \\) we would need 19 GB of VRAM, and for \\( N=100,000 \\) we would need almost 1TB just to store the \\( \mathbf{QK}^T \\) matrices.
+
+Long story short, the default self-attention algorithm quickly becomes prohibitively memory-expensive for large input contexts.
+
+As LLMs improve in text comprehension and generation, they are applied to increasingly complex tasks. While models once handled the translation or summarization of a few sentences, they now manage entire pages, demanding the capability to process extensive input lengths.
+
+How can we get rid of the exorbitant memory requirements for large input lengths? We need a new way to compute the self-attention mechanism that gets rid of the \\( QK^T \\) matrix. [Tri Dao et al.](https://arxiv.org/abs/2205.14135) developed exactly such a new algorithm and called it **Flash Attention**.
+
+In a nutshell, Flash Attention breaks the  \\(\mathbf{V} \times \text{Softmax}(\mathbf{QK}^T\\)) computation apart and instead computes smaller chunks of the output by iterating over multiple softmax computation steps:
+
+$$ \textbf{O}_i \leftarrow s^a_{ij} * \textbf{O}_i + s^b_{ij} * \mathbf{V}_{j} \times \text{Softmax}(\mathbf{QK}^T_{i,j}) \text{ for multiple } i, j \text{ iterations} $$
+
+with \\( s^a_{ij} \\) and \\( s^b_{ij} \\) being some softmax normalization statistics that need to be recomputed for every \\( i \\) and \\( j \\) .
+
+Please note that the whole Flash Attention is a bit more complex and is greatly simplified here as going in too much depth is out of scope for this guide. The reader is invited to take a look at the well-written [Flash Attention paper](https://arxiv.org/abs/2205.14135) for more details.
+
+The main takeaway here is:
+
+> By keeping track of softmax normalization statistics and by using some smart mathematics, Flash Attention gives **numerical identical** outputs compared to the default self-attention layer at a memory cost that only increases linearly with \\( N \\) .
+
+Looking at the formula, one would intuitively say that Flash Attention must be much slower compared to the default self-attention formula as more computation needs to be done. Indeed Flash Attention requires more FLOPs compared to normal attention as the softmax normalization statistics have to constantly be recomputed (see [paper](https://arxiv.org/abs/2205.14135) for more details if interested)
+
+> However, Flash Attention is much faster in inference compared to default attention which comes from its ability to significantly reduce the demands on the slower, high-bandwidth memory of the GPU (VRAM), focusing instead on the faster on-chip memory (SRAM).
+
+Essentially, Flash Attention makes sure that all intermediate write and read operations can be done using the fast *on-chip* SRAM memory instead of having to access the slower VRAM memory to compute the output vector \\( \mathbf{O} \\) .
+
+In practice, there is currently absolutely no reason to **not** use Flash Attention if available. The algorithm gives mathematically the same outputs, and is both faster and more memory-efficient.
+
+Let's look at a practical example.
+
+Our OctoCoder model now gets a significantly longer input prompt which includes a so-called *system prompt*. System prompts are used to steer the LLM into a better assistant that is tailored to the users' task.
+In the following, we use a system prompt that will make OctoCoder a better coding assistant.
+
+```python
+system_prompt = """Below are a series of dialogues between various people and an AI technical assistant.
+The assistant tries to be helpful, polite, honest, sophisticated, emotionally aware, and humble but knowledgeable.
+The assistant is happy to help with code questions and will do their best to understand exactly what is needed.
+It also tries to avoid giving false or misleading information, and it caveats when it isn't entirely sure about the right answer.
+That said, the assistant is practical really does its best, and doesn't let caution get too much in the way of being useful.
+
+The Starcoder models are a series of 15.5B parameter models trained on 80+ programming languages from The Stack (v1.2) (excluding opt-out requests).
+The model uses Multi Query Attention, was trained using the Fill-in-the-Middle objective, and with 8,192 tokens context window for a trillion tokens of heavily deduplicated data.
+
+-----
+
+Question: Write a function that takes two lists and returns a list that has alternating elements from each input list.
+
+Answer: Sure. Here is a function that does that.
+
+def alternating(list1, list2):
+   results = []
+   for i in range(len(list1)):
+       results.append(list1[i])
+       results.append(list2[i])
+   return results
+
+Question: Can you write some test cases for this function?
+
+Answer: Sure, here are some tests.
+
+assert alternating([10, 20, 30], [1, 2, 3]) == [10, 1, 20, 2, 30, 3]
+assert alternating([True, False], [4, 5]) == [True, 4, False, 5]
+assert alternating([], []) == []
+
+Question: Modify the function so that it returns all input elements when the lists have uneven length. The elements from the longer list should be at the end.
+
+Answer: Here is the modified function.
+
+def alternating(list1, list2):
+   results = []
+   for i in range(min(len(list1), len(list2))):
+       results.append(list1[i])
+       results.append(list2[i])
+   if len(list1) > len(list2):
+       results.extend(list1[i+1:])
+   else:
+       results.extend(list2[i+1:])
+   return results
+
+-----
+"""
+```
+For demonstration purposes, we duplicate the system prompt by ten so that the input length is long enough to observe Flash Attention's memory savings.
+We append the original text prompt `"Question: Please write a function in Python that transforms bytes to Giga bytes.\n\nAnswer: Here"`
+
+```python
+long_prompt = 10 * system_prompt + prompt
+```
+
+We instantiate our model again in bfloat16 precision.
+
+```python
+model = AutoModelForCausalLM.from_pretrained("bigcode/octocoder", torch_dtype=torch.bfloat16, device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained("bigcode/octocoder")
+
+pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
+```
+
+Let's now run the model just like before *without Flash Attention* and measure the peak GPU memory requirement and inference time.
+
+```python
+import time
+
+start_time = time.time()
+result = pipe(long_prompt, max_new_tokens=60)[0]["generated_text"][len(long_prompt):]
+
+print(f"Generated in {time.time() - start_time} seconds.")
+result
+```
+
+**Output**:
+```
+Generated in 10.96854019165039 seconds.
+Sure. Here is a function that does that.\n\ndef bytes_to_giga(bytes):\n   return bytes / 1024 / 1024 / 1024\n\nAnswer: Sure. Here is a function that does that.\n\ndef
+````
+
+We're getting the same output as before, however this time, the model repeats the answer multiple times until it's 60 tokens cut-off. This is not surprising as we've repeated the system prompt ten times for demonstration purposes and thus cued the model to repeat itself.
+
+**Note** that the system prompt should not be repeated ten times in real-world applications - one time is enough!
+
+Let's measure the peak GPU memory requirement.
+
+```python
+bytes_to_giga_bytes(torch.cuda.max_memory_allocated())
+```
+
+**Output**:
+```bash
+37.668193340301514
+```
+
+As we can see the peak GPU memory requirement is now significantly higher than in the beginning, which is largely due to the longer input sequence. Also the generation takes a little over a minute now.
+
+We call `flush()` to free GPU memory for our next experiment.
+
+```python
+flush()
+```
+
+For comparison, let's run the same function, but enable Flash Attention instead.
+To do so, we convert the model to [BetterTransformers](https://huggingface.co/docs/optimum/bettertransformer/overview) and by doing so enabling PyTorch's [SDPA self-attention](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) which in turn is based on Flash Attention.
+
+```python
+model.to_bettertransformer()
+```
+
+Now we run the exact same code snippet as before and under the hood Transformers will make use of Flash Attention.
+
+```py
+start_time = time.time()
+with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
+    result = pipe(long_prompt, max_new_tokens=60)[0]["generated_text"][len(long_prompt):]
+
+print(f"Generated in {time.time() - start_time} seconds.")
+result
+```
+
+**Output**:
+```
+Generated in 3.0211617946624756 seconds.
+ Sure. Here is a function that does that.\n\ndef bytes_to_giga(bytes):\n   return bytes / 1024 / 1024 / 1024\n\nAnswer: Sure. Here is a function that does that.\n\ndef
+```
+
+We're getting the exact same result as before, but can observe a very significant speed-up thanks to Flash Attention.
+
+Let's measure the memory consumption one last time.
+
+```python
+bytes_to_giga_bytes(torch.cuda.max_memory_allocated())
+```
+
+**Output**:
+```
+32.617331981658936
+```
+
+And we're almost back to our original 29GB peak GPU memory from the beginning.
+
+We can observe that we only use roughly 100MB more GPU memory when passing a very long input sequence with Flash Attention compared to passing a short input sequence as done in the beginning.
+
+```py
+flush()
+```
+For more information on how to use Flash Attention, please have a look at [this doc page](https://huggingface.co/docs/transformers/v4.34.0/en/perf_infer_gpu_one#flash-attention-2).
+## 3. Architectural Innovations
+
+So far we have looked into improving computational and memory efficiency by:
+
+-   Casting the weights to a lower precision format
+-   Replacing the self-attention algorithm with a more memory- and compute efficient version
+
+Let's now look into how we can change the architecture of an LLM so that it is most effective and efficient for task that require long text inputs, *e.g.*:
+-   Retrieval augmented Questions Answering,
+-   Summarization,
+-   Chat
+
+Note that *chat* not only requires the LLM to handle long text inputs, but it also necessitates that the LLM is able to efficiently handle the back-and-forth dialogue between user and assistant (such as ChatGPT).
+
+Once trained, the fundamental LLM architecture is difficult to change, so it is important to make considerations about the LLM's tasks beforehand and accordingly optimize the model's architecture.
+There are two important components of the model architecture that quickly become memory and/or performance bottlenecks for large input sequences.
+
+-   The positional embeddings
+-   The key-value cache
+
+Let's go over each component in more detail
+
+### 3.1 Improving positional embeddings of LLMs
+
+Self-attention puts each token in relation to each other's tokens.
+As an example, the \\( \text{Softmax}(\mathbf{QK}^T) \\) matrix of the text input sequence *"Hello", "I", "love", "you"* could look as follows:
+
+![](/blog/assets/163_optimize_llm/self_attn_tokens.png)
+
+Each word token is given a probability mass at which it attends all other word tokens and, therefore is put into relation with all other word tokens. E.g. the word *"love"* attends to the word *"Hello"* with 5%, to *"I"* with 30%, and to itself with 65%.
+
+A LLM based on self-attention, but without position embeddings would have great difficulties in understanding the positions of the text inputs to each other.
+This is because the probability score computed by \\( \mathbf{QK}^T \\) relates each word token to each other word token in \\( O(1) \\) computations regardless of their relative positional distance to each other.
+Therefore, for the LLM without position embeddings each token appears to have the same distance to all other tokens, *e.g.* differentiating between *"Hello I love you"* and *"You love I hello"* would be very challenging.
+
+For the LLM to understand sentence order, an additional *cue* is needed and is usually applied in the form of *positional encodings* (or also called *positional embeddings*).
+Positional encodings, encode the position of each token into a numerical presentation that the LLM can leverage to better understand sentence order.
+
+The authors of the [*Attention Is All You Need*](https://arxiv.org/abs/1706.03762) paper introduced sinusoidal positional embeddings \\( \mathbf{P} = \mathbf{p}_1, \ldots, \mathbf{p}_N \\) .
+where each vector \\( \mathbf{p}_i \\) is computed as a sinusoidal function of its position \\( i \\) .
+The positional encodings are then simply added to the input sequence vectors \\( \mathbf{\hat{X}} = \mathbf{\hat{x}}_1, \ldots, \mathbf{\hat{x}}_N \\) = \\( \mathbf{x}_1 + \mathbf{p}_1, \ldots, \mathbf{x}_N + \mathbf{p}_N \\) thereby cueing the model to better learn sentence order.
+
+Instead of using fixed position embeddings, others (such as [Devlin et al.](https://arxiv.org/abs/1810.04805)) used learned positional encodings for which the positional embeddings
+\\( \mathbf{P} \\) are learned during training.
+
+Sinusoidal and learned position embeddings used to be the predominant methods to encode sentence order into LLMs, but a couple of problems related to these positional encodings were found:
+
+  1. Sinusoidal and learned position embeddings are both absolute positional embeddings, *i.e.* encoding a unique embedding for each position id: \\( 0, \ldots, N \\) . As shown by [Huang et al.](https://arxiv.org/abs/2009.13658) and [Su et al.](https://arxiv.org/abs/2104.09864), absolute positional embeddings lead to poor LLM performance for long text inputs. For long text inputs, it is advantageous if the model learns the relative positional distance input tokens have to each other instead of their absolute position.
+  2. When using learned position embeddings, the LLM has to be trained on a fixed input length \\( N \\), which makes it difficult to extrapolate to an input length longer than what it was trained on.
+
+Recently, relative positional embeddings that can tackle the above mentioned problems have become more popular, most notably:
+
+-   [Rotary Position Embedding (RoPE)](https://arxiv.org/abs/2104.09864)
+-   [ALiBi](https://arxiv.org/abs/2108.12409)
+
+Both *RoPE* and *ALiBi* argue that it's best to cue the LLM about sentence order directly in the self-attention algorithm as it's there that word tokens are put into relation with each other. More specifically, sentence order should be cued by modifying the \\( \mathbf{QK}^T \\) computation.
+
+Without going into too many details, *RoPE* notes that positional information can be encoded into query-key pairs, *e.g.* \\( \mathbf{q}_i \\) and \\( \mathbf{x}_j \\) by rotating each vector by an angle \\( \theta * i \\) and \\( \theta * j \\) respectively with \\( i, j \\) describing each vectors sentence position:
+
+$$ \mathbf{\hat{q}}_i^T \mathbf{\hat{x}}_j = \mathbf{{q}}_i^T \mathbf{R}_{\theta, i -j} \mathbf{{x}}_j. $$
+
+\\( \mathbf{R}_{\theta, i - j} \\) thereby represents a rotational matrix. \\( \theta \\) is *not* learned during training, but instead set to a pre-defined value that depends on the maximum input sequence length during training.
+
+> By doing so, the propability score between \\( \mathbf{q}_i \\) and \\( \mathbf{q}_j \\) is only affected if \\( i \ne j \\) and solely depends on the relative distance \\( i - j \\) regardless of each vector's specific positions \\( i \\) and \\( j \\) .
+
+*RoPE* is used in multiple of today's most important LLMs, such as:
+
+-   [**Falcon**](https://huggingface.co/tiiuae/falcon-40b)
+-   [**Llama**](https://arxiv.org/abs/2302.13971)
+-   [**PaLM**](https://arxiv.org/abs/2204.02311)
+
+As an alternative, *ALiBi* proposes a much simpler relative position encoding scheme. The relative distance that input tokens have to each other is added as a negative integer scaled by a pre-defined value `m` to each query-key entry of the \\( \mathbf{QK}^T \\) matrix right before the softmax computation.
+
+![](/blog/assets/163_optimize_llm/alibi.png)
+
+As shown in the [ALiBi](https://arxiv.org/abs/2108.12409) paper, this simple relative positional encoding allows the model to retain a high performance even at very long text input sequences.
+
+*ALiBi* is used in multiple of today's most important LLMs, such as:
+
+-   [**MPT**](https://huggingface.co/mosaicml/mpt-30b)
+-   [**BLOOM**](https://huggingface.co/bigscience/bloom)
+
+Both *RoPE* and *ALiBi* position encodings can extrapolate to input lengths not seen during training whereas it has been shown that extrapolation works much better out-of-the-box for *ALiBi* as compared to *RoPE*.
+For ALiBi, one simply increases the values of the lower triangular position matrix to match the length of the input sequence.
+For *RoPE*, keeping the same \\( \theta \\) that was used during training leads to poor results when passing text inputs much longer than those seen during training, *c.f* [Press et al.](https://arxiv.org/abs/2108.12409). However, the community has found a couple of effective tricks that adapt \\( \theta \\), thereby allowing *RoPE* position embeddings to work well for extrapolated text input sequences (see [here](https://github.com/huggingface/transformers/pull/24653)).
+
+> Both RoPE and ALiBi are relative positional embeddings that are *not* learned during training, but instead are based on the following intuitions:
+ -   Positional cues about the text inputs should be given directly to the \\( QK^T \\) matrix of the self-attention layer
+ -   The LLM should be incentivized to learn a constant *relative* distance positional encodings have to each other
+ -   The further text input tokens are from each other, the lower the probability of their query-value probability. Both RoPE and ALiBi lower the query-key probability of tokens far away from each other. RoPE by decreasing their vector product by increasing the angle between the query-key vectors. ALiBi by adding large negative numbers to the vector product
+
+In conclusion, LLMs that are intended to be deployed in tasks that require handling large text inputs are better trained with relative positional embeddings, such as RoPE and ALiBi. Also note that even if an LLM with RoPE and ALiBi has been trained only on a fixed length of say \\( N_1 = 2048 \\) it can still be used in practice with text inputs much larger than \\( N_1 \\), like \\( N_2 = 8192 > N_1 \\) by extrapolating the positional embeddings.
+
+### 3.2 The key-value cache
+
+Auto-regressive text generation with LLMs works by iteratively putting in an input sequence, sampling the next token, appending the next token to the input sequence, and continuing to do so until the LLM produces a token that signifies that the generation has finished.
+
+Please have a look at [Transformer's Generate Text Tutorial](https://huggingface.co/docs/transformers/llm_tutorial#generate-text) to get a more visual explanation of how auto-regressive generation works.
+
+Let's run a quick code snippet to show how auto-regressive works in practice. We will simply take the most likely next token via `torch.argmax`.
+
+```python
+input_ids = tokenizer(prompt, return_tensors="pt")["input_ids"].to("cuda")
+
+for _ in range(5):
+  next_logits = model(input_ids)["logits"][:, -1:]
+  next_token_id = torch.argmax(next_logits,dim=-1)
+
+  input_ids = torch.cat([input_ids, next_token_id], dim=-1)
+  print("shape of input_ids", input_ids.shape)
+
+generated_text = tokenizer.batch_decode(input_ids[:, -5:])
+generated_text
+```
+
+**Output**:
+```
+shape of input_ids torch.Size([1, 21])
+shape of input_ids torch.Size([1, 22])
+shape of input_ids torch.Size([1, 23])
+shape of input_ids torch.Size([1, 24])
+shape of input_ids torch.Size([1, 25])
+[' Here is a Python function']
+```
+
+As we can see every time we increase the text input tokens by the just sampled token.
+
+With very few exceptions, LLMs are trained using the [causal language modeling objective](https://huggingface.co/docs/transformers/tasks/language_modeling#causal-language-modeling) and therefore mask the upper triangle matrix of the attention score - this is why in the two diagrams above the attention scores are left blank (*a.k.a* have 0 probability). For a quick recap on causal language modeling you can refer to the [*Illustrated Self Attention blog*](https://jalammar.github.io/illustrated-gpt2/#part-2-illustrated-self-attention).
+
+As a consequence, tokens *never* depend on previous tokens, more specifically the \\( \mathbf{q}_i \\) vector is never put in relation with any key, values vectors \\( \mathbf{k}_j, \mathbf{v}_j \\) if \\( j > i \\) . Instead \\( \mathbf{q}_i \\) only attends to previous key-value vectors \\( \mathbf{k}_{m < i}, \mathbf{v}_{m < i} \text{ , for } m \in \{0, \ldots i - 1\} \\). In order to reduce unnecessary computation, one can therefore cache each layer's key-value vectors for all previous timesteps.
+
+In the following, we will tell the LLM to make use of the key-value cache by retrieving and forwarding it for each forward pass.
+In Transformers, we can retrieve the key-value cache by passing the `use_cache` flag to the `forward` call and can then pass it with the current token.
+
+```python
+past_key_values = None # past_key_values is the key-value cache
+generated_tokens = []
+next_token_id = tokenizer(prompt, return_tensors="pt")["input_ids"].to("cuda")
+
+for _ in range(5):
+  next_logits, past_key_values = model(next_token_id, past_key_values=past_key_values, use_cache=True).to_tuple()
+  next_logits = next_logits[:, -1:]
+  next_token_id = torch.argmax(next_logits, dim=-1)
+
+  print("shape of input_ids", next_token_id.shape)
+  print("length of key-value cache", len(past_key_values[0][0]))  # past_key_values are of shape [num_layers, 0 for k, 1 for v, batch_size, length, hidden_dim]
+  generated_tokens.append(next_token_id.item())
+
+generated_text = tokenizer.batch_decode(generated_tokens)
+generated_text
+```
+
+**Output**:
+```
+shape of input_ids torch.Size([1, 1])
+length of key-value cache 20
+shape of input_ids torch.Size([1, 1])
+length of key-value cache 21
+shape of input_ids torch.Size([1, 1])
+length of key-value cache 22
+shape of input_ids torch.Size([1, 1])
+length of key-value cache 23
+shape of input_ids torch.Size([1, 1])
+length of key-value cache 24
+[' Here', ' is', ' a', ' Python', ' function']
+```
+
+As one can see, when using the key-value cache the text input tokens are *not* increased in length, but remain a single input vector. The length of the key-value cache on the other hand is increased by one at every decoding step.
+
+> Making use of the key-value cache means that the \\( \mathbf{QK}^T \\) is essentially reduced to \\( \mathbf{q}_c\mathbf{K}^T \\) with \\( \mathbf{q}_c \\) being the query projection of the currently passed input token which is *always* just a single vector.
+
+Using the key-value cache has two advantages:
+-   Significant increase in computational efficiency as less computations are performed compared to computing the full \\( \mathbf{QK}^T \\) matrix. This leads to an increase in inference speed
+-   The maximum required memory is not increased quadratically with the number of generated tokens, but only increases linearly.
+
+> One should *always* make use of the key-value cache as it leads to identical results and a significant speed-up for longer input sequences. Transformers has the key-value cache enabled by default when making use of the text pipeline or the [`generate` method](https://huggingface.co/docs/transformers/main_classes/text_generation).
+
+Note that the key-value cache is especially useful for applications such as chat where multiple passes of auto-regressive decoding are required. Let's look at an example.
+
+```
+User: How many people live in France?
+Assistant: Roughly 75 million people live in France
+User: And how many are in Germany?
+Assistant: Germany has ca. 81 million inhabitants
+```
+
+In this chat, the LLM runs auto-regressive decoding twice:
+- 1. The first time, the key-value cache is empty and the input prompt is `"User: How many people live in France?"` and the model auto-regressively generates the text `"Roughly 75 million people live in France"` while increasing the key-value cache at every decoding step.
+- 2. The second time the input prompt is `"User: How many people live in France? \n Assistant: Roughly 75 million people live in France \n User: And how many in Germany?"`. Thanks to the cache, all key-value vectors for the first two sentences are already computed. Therefore the input prompt only consists of `"User: And how many in Germany?"`. While processing the shortened input prompt, it's computed key-value vectors are concatenated to the key-value cache of the first decoding. The second Assistant's answer `"Germany has ca. 81 million inhabitants"` is then auto-regressively generated with the key-value cache consisting of encoded key-value vectors of `"User: How many people live in France? \n Assistant: Roughly 75 million people live in France \n User: And how many are in Germany?"`.
+
+Two things should be noted here:
+  1. Keeping all the context is crucial for LLMs deployed in chat so that the LLM understands all the previous context of the conversation. E.g. for the example above the LLM needs to understand that the user refers to the population when asking `"And how many are in Germany"`.
+  2. The key-value cache is extremely useful for chat as it allows us to continuously grow the encoded chat history instead of having to re-encode the chat history again from scratch (as e.g. would be the case when using an encoder-decoder architecture).
+
+There is however one catch. While the required peak memory for the \\( \mathbf{QK}^T \\) matrix is significantly reduced, holding the key-value cache in memory can become very memory expensive for long input sequences or multi-turn chat. Remember that the key-value cache needs to store the key-value vectors for all previous input vectors \\( \mathbf{x}_i \text{, for } i \in \{1, \ldots, c - 1\} \\) for all self-attention layers and for all attention heads.
+
+Let's compute the number of float values that need to be stored in the key-value cache for the LLM `bigcode/octocoder` that we used before.
+The number of float values amounts to two times the sequence length times the number of attention heads times the attention head dimension and times the number of layers.
+Computing this for our LLM at a hypothetical input sequence length of 16000 gives:
+
+```python
+config = model.config
+2 * 16_000 * config.n_layer * config.n_head * config.n_embd // config.n_head
+```
+
+**Output**:
+```
+7864320000
+```
+
+Roughly 8 billion float values! Storing 8 billion float values in `float16` precision requires around 15 GB of RAM which is circa half as much as the model weights themselves!
+Researchers have proposed two methods that allow to significantly reduce the memory cost of storing the key-value cache:
+
+  1.  [Multi-Query-Attention (MQA)](https://arxiv.org/abs/1911.02150)
+
+Multi-Query-Attention was proposed in Noam Shazeer's *Fast Transformer Decoding: One Write-Head is All You Need* paper. As the title says, Noam found out that instead of using `n_head` key-value projections weights, one can use a single head-value projection weight pair that is shared across all attention heads without that the model's performance significantly degrades.
+
+> By using a single head-value projection weight pair, the key value vectors \\( \mathbf{k}_i, \mathbf{v}_i \\) have to be identical across all attention heads which in turn means that we only need to store 1 key-value projection pair in the cache instead of `n_head` ones.
+
+As most LLMs use between 20 and 100 attention heads, MQA significantly reduces the memory consumption of the key-value cache. For the LLM used in this notebook we could therefore reduce the required memory consumption from 15 GB to less than 400 MB at an input sequence length of 16000.
+
+In addition to memory savings, MQA also leads to improved computational efficiency as explained in the following.
+In auto-regressive decoding, large key-value vectors need to be reloaded, concatenated with the current key-value vector pair to be then fed into the \\( \mathbf{q}_c\mathbf{K}^T \\) computation at every step. For auto-regressive decoding, the required memory bandwidth for the constant reloading can become a serious time bottleneck. By reducing the size of the key-value vectors less memory needs to be accessed, thus reducing the memory bandwidth bottleneck. For more detail, please have a look at [Noam's paper](https://arxiv.org/abs/1911.02150).
+
+The important part to understand here is that reducing the number of key-value attention heads to 1 only makes sense if a key-value cache is used. The peak memory consumption of the model for a single forward pass without key-value cache stays unchanged as every attention head still has a unique query vector so that each attention head still has a different \\( \mathbf{QK}^T \\) matrix.
+
+MQA has seen wide adoption by the community and is now used by many of the most popular LLMs:
+
+-   [**Falcon**](https://huggingface.co/tiiuae/falcon-40b)
+-   [**PaLM**](https://arxiv.org/abs/2204.02311)
+-   [**MPT**](https://huggingface.co/mosaicml/mpt-30b)
+-   [**BLOOM**](https://huggingface.co/bigscience/bloom)
+
+Also, the checkpoint used in this notebook - `bigcode/octocoder` - makes use of MQA.
+
+  2.  [Grouped-Query-Attention (GQA)](https://arxiv.org/abs/2305.13245)
+
+Grouped-Query-Attention, as proposed by Ainslie et al. from Google, found that using MQA can often lead to quality degradation compared to using vanilla multi-key-value head projections. The paper argues that more model performance can be kept by less drastically reducing the number of query head projection weights. Instead of using just a single key-value projection weight, `n < n_head` key-value projection weights should be used. By choosing `n` to a significantly smaller value than `n_head`, such as 2,4 or 8 almost all of the memory and speed gains from MQA can be kept while sacrificing less model capacity and thus arguably less performance.
+
+Moreover, the authors of GQA found out that existing model checkpoints can be *uptrained* to have a GQA architecture with as little as 5% of the original pre-training compute. While 5% of the original pre-training compute can still be a massive amount, GQA *uptraining* allows existing checkpoints to be useful for longer input sequences.
+
+GQA was only recently proposed which is why there is less adoption at the time of writing this notebook.
+The most notable application of GQA is [Llama-v2](https://huggingface.co/meta-llama/Llama-2-70b-hf).
+
+> As a conclusion, it is strongly recommended to make use of either GQA or MQA if the LLM is deployed with auto-regressive decoding and is required to handle large input sequences as is the case for example for chat.
+
+## Conclusion
+
+The research community is constantly coming up with new, nifty ways to speed up inference time for ever-larger LLMs. As an example, one such promising research direction is [speculative decoding](https://arxiv.org/abs/2211.17192) where "easy tokens" are generated by smaller, faster language models and only "hard tokens" are generated by the LLM itself. Going into more detail is out of the scope of this notebook, but can be read upon in this [nice blog post](https://huggingface.co/blog/assisted-generation).
+
+The reason massive LLMs such as GPT3/4, Llama-2-70b, Claude, PaLM can run so quickly in chat-interfaces such as [Hugging Face Chat](https://huggingface.co/chat/) or ChatGPT is to a big part thanks to the above-mentioned improvements in precision, algorithms, and architecture.
+Going forward, accelerators such as GPUs, TPUs, etc... will only get faster and allow for more memory, but one should nevertheless always make sure to use the best available algorithms and architectures to get the most bang for your buck 🤗

docs/source/en/main_classes/agent.md

@@ -18,7 +18,7 @@ rendered properly in your Markdown viewer.
 
 <Tip warning={true}>
 
-Transformers Agent is an experimental API which is subject to change at any time. Results returned by the agents
+Transformers Agents is an experimental API which is subject to change at any time. Results returned by the agents
 can vary as the APIs or underlying models are prone to change.
 
 </Tip>

docs/source/en/main_classes/callback.md

@@ -25,7 +25,7 @@ Callbacks are "read only" pieces of code, apart from the [`TrainerControl`] obje
 cannot change anything in the training loop. For customizations that require changes in the training loop, you should
 subclass [`Trainer`] and override the methods you need (see [trainer](trainer) for examples).
 
-By default a [`Trainer`] will use the following callbacks:
+By default, `TrainingArguments.report_to` is set to `"all"`, so a [`Trainer`] will use the following callbacks.
 
 - [`DefaultFlowCallback`] which handles the default behavior for logging, saving and evaluation.
 - [`PrinterCallback`] or [`ProgressCallback`] to display progress and print the
@@ -45,6 +45,8 @@ By default a [`Trainer`] will use the following callbacks:
 - [`~integrations.DagsHubCallback`] if [dagshub](https://dagshub.com/) is installed.
 - [`~integrations.FlyteCallback`] if [flyte](https://flyte.org/) is installed.
 
+If a package is installed but you don't wish to use the accompanying integration, you can change `TrainingArguments.report_to` to a list of just those integrations you want to use (e.g. `["azure_ml", "wandb"]`). 
+
 The main class that implements callbacks is [`TrainerCallback`]. It gets the
 [`TrainingArguments`] used to instantiate the [`Trainer`], can access that
 Trainer's internal state via [`TrainerState`], and can take some actions on the training loop via

docs/source/en/main_classes/deepspeed.md

@@ -1224,6 +1224,7 @@ As long as you don't enable `offload_optimizer` you can mix and match DeepSpeed
 optimizers, with the exception of using the combination of HuggingFace scheduler and DeepSpeed optimizer:
 
 | Combos       | HF Scheduler | DS Scheduler |
+|:-------------|:-------------|:-------------|
 | HF Optimizer | Yes          | Yes          |
 | DS Optimizer | No           | Yes          |
 

docs/source/en/main_classes/feature_extractor.md

@@ -16,10 +16,7 @@ rendered properly in your Markdown viewer.
 
 # Feature Extractor
 
-A feature extractor is in charge of preparing input features for audio or vision models. This includes feature extraction
-from sequences, *e.g.*, pre-processing audio files to Log-Mel Spectrogram features, feature extraction from images
-*e.g.* cropping image image files, but also padding, normalization, and conversion to Numpy, PyTorch, and TensorFlow
-tensors.
+A feature extractor is in charge of preparing input features for audio or vision models. This includes feature extraction from sequences, e.g., pre-processing audio files to generate Log-Mel Spectrogram features, feature extraction from images, e.g., cropping image files, but also padding, normalization, and conversion to NumPy, PyTorch, and TensorFlow tensors.
 
 
 ## FeatureExtractionMixin

docs/source/en/main_classes/logging.md

@@ -71,6 +71,23 @@ verbose to the most verbose), those levels (with their corresponding int values
 
 By default, `tqdm` progress bars will be displayed during model download. [`logging.disable_progress_bar`] and [`logging.enable_progress_bar`] can be used to suppress or unsuppress this behavior.
 
+## `logging` vs `warnings`
+
+Python has two logging systems that are often used in conjunction: `logging`, which is explained above, and `warnings`,
+which allows further classification of warnings in specific buckets, e.g., `FutureWarning` for a feature or path
+that has already been deprecated and `DeprecationWarning` to indicate an upcoming deprecation.
+
+We use both in the `transformers` library. We leverage and adapt `logging`'s `captureWarning` method to allow
+management of these warning messages by the verbosity setters above.
+
+What does that mean for developers of the library? We should respect the following heuristic:
+- `warnings` should be favored for developers of the library and libraries dependent on `transformers`
+- `logging` should be used for end-users of the library using it in every-day projects
+
+See reference of the `captureWarnings` method below.
+
+[[autodoc]] logging.captureWarnings
+
 ## Base setters
 
 [[autodoc]] logging.set_verbosity_error

docs/source/en/main_classes/output.md

@@ -40,6 +40,13 @@ an optional `attentions` attribute. Here we have the `loss` since we passed alon
 `hidden_states` and `attentions` because we didn't pass `output_hidden_states=True` or
 `output_attentions=True`.
 
+<Tip>
+
+When passing `output_hidden_states=True` you may expect the `outputs.hidden_states[-1]` to match `outputs.last_hidden_states` exactly.
+However, this is not always the case. Some models apply normalization or subsequent process to the last hidden state when it's returned.
+</Tip>
+
+
 You can access each attribute as you would usually do, and if that attribute has not been returned by the model, you
 will get `None`. Here for instance `outputs.loss` is the loss computed by the model, and `outputs.attentions` is
 `None`.

docs/source/en/main_classes/pipelines.md

@@ -352,6 +352,12 @@ Pipelines available for computer vision tasks include the following.
     - __call__
     - all
 
+### ImageToImagePipeline
+
+[[autodoc]] ImageToImagePipeline
+    - __call__
+    - all
+
 ### ObjectDetectionPipeline
 
 [[autodoc]] ObjectDetectionPipeline

docs/source/en/main_classes/quantization.md

@@ -20,7 +20,7 @@ rendered properly in your Markdown viewer.
 
 🤗 Transformers has integrated `optimum` API to perform GPTQ quantization on language models. You can load and quantize your model in 8, 4, 3 or even 2 bits without a big drop of performance and faster inference speed! This is supported by most GPU hardwares.
 
-To learn more about the the quantization model, check out: 
+To learn more about the quantization model, check out: 
 - the [GPTQ](https://arxiv.org/pdf/2210.17323.pdf) paper
 - the `optimum` [guide](https://huggingface.co/docs/optimum/llm_quantization/usage_guides/quantization) on GPTQ quantization
 - the [`AutoGPTQ`](https://github.com/PanQiWei/AutoGPTQ) library used as the backend

docs/source/en/main_classes/tokenizer.md

@@ -55,6 +55,7 @@ to a given token).
 
 [[autodoc]] PreTrainedTokenizer
     - __call__
+    - apply_chat_template
     - batch_decode
     - decode
     - encode
@@ -68,6 +69,7 @@ loaded very simply into 🤗 transformers. Take a look at the [Using tokenizers
 
 [[autodoc]] PreTrainedTokenizerFast
     - __call__
+    - apply_chat_template
     - batch_decode
     - decode
     - encode

docs/source/en/model_doc/auto.md

@@ -266,6 +266,10 @@ The following auto classes are available for the following computer vision tasks
 
 [[autodoc]] AutoModelForImageSegmentation
 
+### AutoModelForImageToImage
+
+[[autodoc]] AutoModelForImageToImage
+
 ### AutoModelForSemanticSegmentation
 
 [[autodoc]] AutoModelForSemanticSegmentation

docs/source/en/model_doc/bark.md

@@ -64,7 +64,7 @@ model.enable_cpu_offload()
 
 Note that 🤗 Accelerate must be installed before using this feature. [Here's how to install it.](https://huggingface.co/docs/accelerate/basic_tutorials/install)
 
-#### Combining optimizaton techniques
+#### Combining optimization techniques
 
 You can combine optimization techniques, and use CPU offload, half-precision and 🤗 Better Transformer all at once.
 

docs/source/en/model_doc/bros.md

@@ -0,0 +1,115 @@
+<!--Copyright 2023 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.
+-->
+
+# BROS
+
+## Overview
+
+The BROS model was proposed in [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
+
+BROS stands for *BERT Relying On Spatiality*. It is an encoder-only Transformer model that takes a sequence of tokens and their bounding boxes as inputs and outputs a sequence of hidden states. BROS encode relative spatial information instead of using absolute spatial information.
+
+It is pre-trained with two objectives: a token-masked language modeling objective (TMLM) used in BERT, and a novel area-masked language modeling objective (AMLM)
+In TMLM, tokens are randomly masked, and the model predicts the masked tokens using spatial information and other unmasked tokens.
+AMLM is a 2D version of TMLM. It randomly masks text tokens and predicts with the same information as TMLM, but it masks text blocks (areas).
+
+`BrosForTokenClassification` has a simple linear layer on top of BrosModel. It predicts the label of each token.
+`BrosSpadeEEForTokenClassification` has an `initial_token_classifier` and `subsequent_token_classifier` on top of BrosModel. `initial_token_classifier` is used to predict the first token of each entity, and `subsequent_token_classifier` is used to predict the next token of within entity. `BrosSpadeELForTokenClassification` has an `entity_linker` on top of BrosModel. `entity_linker` is used to predict the relation between two entities.
+
+`BrosForTokenClassification` and `BrosSpadeEEForTokenClassification` essentially perform the same job. However, `BrosForTokenClassification` assumes input tokens are perfectly serialized (which is very challenging task since they exist in a 2D space), while `BrosSpadeEEForTokenClassification` allows for more flexibility in handling serialization errors as it predicts next connection tokens from one token.
+
+`BrosSpadeELForTokenClassification` perform the intra-entity linking task. It predicts relation from one token (of one entity) to another token (of another entity) if these two entities share some relation.
+
+BROS achieves comparable or better result on Key Information Extraction (KIE) benchmarks such as FUNSD, SROIE, CORD and SciTSR, without relying on explicit visual features.
+
+
+The abstract from the paper is the following:
+
+*Key information extraction (KIE) from document images requires understanding the contextual and spatial semantics of texts in two-dimensional (2D) space. Many recent studies try to solve the task by developing pre-trained language models focusing on combining visual features from document images with texts and their layout. On the other hand, this paper tackles the problem by going back to the basic: effective combination of text and layout. Specifically, we propose a pre-trained language model, named BROS (BERT Relying On Spatiality), that encodes relative positions of texts in 2D space and learns from unlabeled documents with area-masking strategy. With this optimized training scheme for understanding texts in 2D space, BROS shows comparable or better performance compared to previous methods on four KIE benchmarks (FUNSD, SROIE*, CORD, and SciTSR) without relying on visual features. This paper also reveals two real-world challenges in KIE tasks-(1) minimizing the error from incorrect text ordering and (2) efficient learning from fewer downstream examples-and demonstrates the superiority of BROS over previous methods.*
+
+Tips:
+
+- [`~transformers.BrosModel.forward`] requires `input_ids` and `bbox` (bounding box). Each bounding box should be in (x0, y0, x1, y1) format (top-left corner, bottom-right corner). Obtaining of Bounding boxes depends on external OCR system. The `x` coordinate should be normalized by document image width, and the `y` coordinate should be normalized by document image height.
+
+```python
+def expand_and_normalize_bbox(bboxes, doc_width, doc_height):
+    # here, bboxes are numpy array
+
+    # Normalize bbox -> 0 ~ 1
+    bboxes[:, [0, 2]] = bboxes[:, [0, 2]] / width
+    bboxes[:, [1, 3]] = bboxes[:, [1, 3]] / height
+```
+
+- [`~transformers.BrosForTokenClassification.forward`, `~transformers.BrosSpadeEEForTokenClassification.forward`, `~transformers.BrosSpadeEEForTokenClassification.forward`] require not only `input_ids` and `bbox` but also `box_first_token_mask` for loss calculation. It is a mask to filter out non-first tokens of each box. You can obtain this mask by saving start token indices of bounding boxes when creating `input_ids` from words. You can make `box_first_token_mask` with following code,
+
+
+```python
+def make_box_first_token_mask(bboxes, words, tokenizer, max_seq_length=512):
+
+    box_first_token_mask = np.zeros(max_seq_length, dtype=np.bool_)
+
+    # encode(tokenize) each word from words (List[str])
+    input_ids_list: List[List[int]] = [tokenizer.encode(e, add_special_tokens=False) for e in words]
+
+    # get the length of each box
+    tokens_length_list: List[int] = [len(l) for l in input_ids_list]
+
+    box_end_token_indices = np.array(list(itertools.accumulate(tokens_length_list)))
+    box_start_token_indices = box_end_token_indices - np.array(tokens_length_list)
+
+    # filter out the indices that are out of max_seq_length
+    box_end_token_indices = box_end_token_indices[box_end_token_indices < max_seq_length - 1]
+    if len(box_start_token_indices) > len(box_end_token_indices):
+        box_start_token_indices = box_start_token_indices[: len(box_end_token_indices)]
+
+    # set box_start_token_indices to True
+    box_first_token_mask[box_start_token_indices] = True
+
+    return box_first_token_mask
+
+```
+
+- Demo scripts can be found [here](https://github.com/clovaai/bros).
+
+This model was contributed by [jinho8345](https://huggingface.co/jinho8345). The original code can be found [here](https://github.com/clovaai/bros).
+
+## BrosConfig
+
+[[autodoc]] BrosConfig
+
+## BrosProcessor
+
+[[autodoc]] BrosProcessor
+    - __call__
+
+## BrosModel
+
+[[autodoc]] BrosModel
+    - forward
+
+
+## BrosForTokenClassification
+
+[[autodoc]] BrosForTokenClassification
+    - forward
+
+
+## BrosSpadeEEForTokenClassification
+
+[[autodoc]] BrosSpadeEEForTokenClassification
+    - forward
+
+
+## BrosSpadeELForTokenClassification
+
+[[autodoc]] BrosSpadeELForTokenClassification
+    - forward

docs/source/en/model_doc/clip.md

@@ -83,8 +83,23 @@ This model was contributed by [valhalla](https://huggingface.co/valhalla). The o
 
 A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with CLIP.
 
-- A blog post on [How to fine-tune CLIP on 10,000 image-text pairs](https://huggingface.co/blog/fine-tune-clip-rsicd).
-- CLIP is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/contrastive-image-text).
+- [Fine tuning CLIP with Remote Sensing (Satellite) images and captions](https://huggingface.co/blog/fine-tune-clip-rsicd), a blog post about how to fine-tune CLIP with [RSICD dataset](https://github.com/201528014227051/RSICD_optimal) and comparison of performance changes due to data augmentation.
+- This [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/contrastive-image-text) shows how to train a CLIP-like vision-text dual encoder model using a pre-trained vision and text encoder using [COCO dataset](https://cocodataset.org/#home).
+
+<PipelineTag pipeline="image-to-text"/>
+
+- A [notebook](https://colab.research.google.com/drive/1tuoAC5F4sC7qid56Z0ap-stR3rwdk0ZV?usp=sharing) on how to use a pretrained CLIP for inference with beam search for image captioning. 🌎
+
+**Image retrieval**
+
+- A [notebook](https://colab.research.google.com/drive/1bLVwVKpAndpEDHqjzxVPr_9nGrSbuOQd?usp=sharing) on image retrieval using pretrained CLIP and computing MRR(Mean Reciprocal Rank) score. 🌎
+- A [notebook](https://colab.research.google.com/github/deep-diver/image_search_with_natural_language/blob/main/notebooks/Image_Search_CLIP.ipynb) on image retrieval and showing the similarity score. 🌎
+- A [notebook](https://colab.research.google.com/drive/1xO-wC_m_GNzgjIBQ4a4znvQkvDoZJvH4?usp=sharing) on how to map images and texts to the same vector space using Multilingual CLIP. 🌎 
+- A [notebook](https://colab.research.google.com/github/vivien000/clip-demo/blob/master/clip.ipynb#scrollTo=uzdFhRGqiWkR) on how to run CLIP on semantic image search using [Unsplash](https://unsplash.com) and [TMBD](https://www.themoviedb.org/) datasets. 🌎
+
+**Explainability**
+
+- A [notebook](https://colab.research.google.com/github/hila-chefer/Transformer-MM-Explainability/blob/main/CLIP_explainability.ipynb) on how to visualize similarity between input token and image segment. 🌎
 
 If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we will review it.
 The resource should ideally demonstrate something new instead of duplicating an existing resource.

docs/source/en/model_doc/code_llama.md

@@ -49,6 +49,8 @@ Here is a sample usage
 python src/transformers/models/llama/convert_llama_weights_to_hf.py \
     --input_dir /path/to/downloaded/llama/weights --model_size 7B --output_dir /output/path
 ```
+Note that executing the script requires enough CPU RAM to host the whole model in float16 precision (even if the biggest versions
+come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM).
 
 - After conversion, the model and tokenizer can be loaded via:
 
@@ -90,8 +92,8 @@ If you only want the infilled part:
 >>> generator = pipeline("text-generation",model="codellama/CodeLlama-7b-hf",torch_dtype=torch.float16, device_map="auto")
 >>> generator('def remove_non_ascii(s: str) -> str:\n    """ <FILL_ME>\n    return result', max_new_tokens = 128, return_type = 1)
 ```
-Note that executing the script requires enough CPU RAM to host the whole model in float16 precision (even if the biggest versions
-come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM). For the 75B model, it's thus 145GB of RAM needed.
+
+Under the hood, the tokenizer [automatically splits by `<FILL_ME>`](https://huggingface.co/docs/transformers/main/model_doc/code_llama#transformers.CodeLlamaTokenizer.fill_token) to create a formatted input string that follows [the original training pattern](https://github.com/facebookresearch/codellama/blob/cb51c14ec761370ba2e2bc351374a79265d0465e/llama/generation.py#L402). This is more robust than preparing the pattern yourself: it avoids pitfalls, such as token glueing, that are very hard to debug.  To see how much CPU and GPU memory you need for this model or others, try [this calculator](https://huggingface.co/spaces/hf-accelerate/model-memory-usage) which can help determine that value.
 
 - The LLaMA tokenizer is a BPE model based on [sentencepiece](https://github.com/google/sentencepiece). One quirk of sentencepiece is that when decoding a sequence, if the first token is the start of the word (e.g. "Banana"), the tokenizer does not prepend the prefix space to the string.
 

docs/source/en/model_doc/deberta-v2.md

@@ -152,3 +152,8 @@ contributed by [kamalkraj](https://huggingface.co/kamalkraj). The original code
 
 [[autodoc]] TFDebertaV2ForQuestionAnswering
     - call
+
+## TFDebertaV2ForMultipleChoice
+
+[[autodoc]] TFDebertaV2ForMultipleChoice
+    - call

docs/source/en/model_doc/falcon.md

@@ -0,0 +1,84 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Falcon
+
+## Overview
+
+Falcon is a class of causal decoder-only models built by [TII](https://www.tii.ae/). The largest Falcon checkpoints
+have been trained on >=1T tokens of text, with a particular emphasis on the [RefinedWeb](https://arxiv.org/abs/2306.01116)
+corpus. They are made available under the Apache 2.0 license.
+
+
+Falcon's architecture is modern and optimized for inference, with multi-query attention and support for efficient
+attention variants like `FlashAttention`. Both 'base' models trained only as causal language models as well as
+'instruct' models that have received further fine-tuning are available.
+
+
+Falcon models are (as of 2023) some of the largest and most powerful open-source language models,
+and consistently rank highly in the [OpenLLM leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
+
+## Converting custom checkpoints 
+
+<Tip>
+
+Falcon models were initially added to the Hugging Face Hub as custom code checkpoints. However, Falcon is now fully
+supported in the Transformers library. If you fine-tuned a model from a custom code checkpoint, we recommend converting
+your checkpoint to the new in-library format, as this should give significant improvements to stability and
+performance, especially for generation, as well as removing the need to use `trust_remote_code=True`!
+
+</Tip>
+
+You can convert custom code checkpoints to full Transformers checkpoints using the `convert_custom_code_checkpoint.py` 
+script located in the
+[Falcon model directory](https://github.com/huggingface/transformers/tree/main/src/transformers/models/falcon)
+of the Transformers library. To use this script, simply call it with 
+`python convert_custom_code_checkpoint.py --checkpoint_dir my_model`. This will convert your checkpoint in-place, and
+you can immediately load it from the directory afterwards with e.g. `from_pretrained()`. If your model hasn't been
+uploaded to the Hub, we recommend making a backup before attempting the conversion, just in case!
+
+
+## FalconConfig
+
+[[autodoc]] FalconConfig
+    - all
+
+## FalconModel
+
+[[autodoc]] FalconModel
+    - forward
+
+## FalconForCausalLM
+
+[[autodoc]] FalconForCausalLM
+    - forward
+
+## FalconForSequenceClassification
+
+[[autodoc]] FalconForSequenceClassification
+    - forward
+
+## FalconForTokenClassification
+
+[[autodoc]] FalconForTokenClassification
+    - forward
+
+## FalconForQuestionAnswering
+
+[[autodoc]] FalconForQuestionAnswering
+    - forward
+
+

docs/source/en/model_doc/flan-ul2.md

@@ -19,10 +19,10 @@ rendered properly in your Markdown viewer.
 ## Overview
 
 Flan-UL2 is an encoder decoder model based on the T5 architecture. It uses the same configuration as the [UL2](ul2) model released earlier last year. 
-It was fine tuned using the "Flan" prompt tuning and dataset collection. Similiar to `Flan-T5`,  one can directly use FLAN-UL2 weights without finetuning the model:
+It was fine tuned using the "Flan" prompt tuning and dataset collection. Similar to `Flan-T5`,  one can directly use FLAN-UL2 weights without finetuning the model:
 
 
-According ot the original blog here are the notable improvements:
+According to the original blog here are the notable improvements:
 
 - The original UL2 model was only trained with receptive field of 512, which made it non-ideal for N-shot prompting where N is large.
 - The Flan-UL2 checkpoint uses a receptive field of 2048 which makes it more usable for few-shot in-context learning.
@@ -53,4 +53,4 @@ The model is pretty heavy (~40GB in half precision) so if you just want to run t
 
 ## Inference
 
-The inference protocol is exaclty the same as any `T5` model, please have a look at the [T5's documentation page](t5) for more details.
+The inference protocol is exactly the same as any `T5` model, please have a look at the [T5's documentation page](t5) for more details.

docs/source/en/model_doc/fuyu.md

@@ -0,0 +1,115 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Fuyu
+
+## Overview
+
+The Fuyu model was created by [ADEPT](https://www.adept.ai/blog/fuyu-8b), and authored by Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. 
+
+The authors introduced Fuyu-8B, a decoder-only multimodal model based on the classic transformers architecture, with query and key normalization. A linear encoder is added to create multimodal embeddings from image inputs. 
+
+By treating image tokens like text tokens and using a special image-newline character, the model knows when an image line ends. Image positional embeddings are removed. This avoids the need for different training phases for various image resolutions. With 8 billion parameters and licensed under CC-BY-NC, Fuyu-8B is notable for its ability to handle both text and images, its impressive context size of 16K, and its overall performance.
+
+<Tip warning={true}>
+
+The `Fuyu` models were trained using `bfloat16`, but the original inference uses `float16` The checkpoints uploaded on the hub use `torch_dtype = 'float16'` which will be
+used by the `AutoModel` API to cast the checkpoints from `torch.float32` to `torch.float16`. 
+
+The `dtype` of the online weights is mostly irrelevant, unless you are using `torch_dtype="auto"` when initializing a model using `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`. The reason is that the model will first be downloaded ( using the `dtype` of the checkpoints online) then it will be cast to the default `dtype` of `torch` (becomes `torch.float32`). Users should specify the `torch_dtype` they want, and if they don't it will be `torch.float32`.
+
+Finetuning the model in `float16` is not recommended and known to produce `nan`, as such the model should be fine-tuned in `bfloat16`.
+
+</Tip>
+
+
+Tips:
+
+- To convert the model, you need to clone the original repository using `git clone https://github.com/persimmon-ai-labs/adept-inference`, then get the checkpoints:
+
+```bash
+git clone https://github.com/persimmon-ai-labs/adept-inference
+wget path/to/fuyu-8b-model-weights.tar
+tar -xvf fuyu-8b-model-weights.tar
+python src/transformers/models/fuyu/convert_fuyu_weights_to_hf.py  --input_dir /path/to/downloaded/fuyu/weights/ --output_dir /output/path \
+    --pt_model_path /path/to/fuyu_8b_release/iter_0001251/mp_rank_00/model_optim_rng.pt
+    --ada_lib_path /path/to/adept-inference
+```
+
+For the chat model:
+```bash
+wget https://axtkn4xl5cip.objectstorage.us-phoenix-1.oci.customer-oci.com/n/axtkn4xl5cip/b/adept-public-data/o/8b_chat_model_release.tar
+tar -xvf 8b_base_model_release.tar
+```
+Then, model can be loaded via:
+
+```py 
+from transformers import FuyuConfig, FuyuForCausalLM
+model_config = FuyuConfig()
+model = FuyuForCausalLM(model_config).from_pretrained('/output/path')
+```
+
+Inputs need to be passed through a specific Processor to have the correct formats.
+A processor requires an image_processor and a tokenizer. Hence, inputs can be loaded via:
+
+```py
+from PIL import Image
+from transformers import AutoTokenizer
+from transformers.models.fuyu.processing_fuyu import FuyuProcessor
+from transformers.models.fuyu.image_processing_fuyu import FuyuImageProcessor
+
+
+tokenizer = AutoTokenizer.from_pretrained('adept-hf-collab/fuyu-8b')
+image_processor = FuyuImageProcessor()
+
+
+processor = FuyuProcessor(image_processor=image_processor, tokenizer=tokenizer)
+text_prompt = "Generate a coco-style caption.\\n"
+
+bus_image_url = "https://huggingface.co/datasets/hf-internal-testing/fixtures-captioning/resolve/main/bus.png"
+bus_image_pil = Image.open(io.BytesIO(requests.get(bus_image_url).content))
+inputs_to_model = processor(text=text_prompt, images=image_pil)
+
+
+```
+
+This model was contributed by [Molbap](https://huggingface.co/Molbap).
+The original code can be found [here](https://github.com/persimmon-ai-labs/adept-inference).
+
+- Fuyu uses a `sentencepiece` based tokenizer, with a `Unigram` model. It supports bytefallback, which is only available in `tokenizers==0.14.0` for the fast tokenizer.
+The `LlamaTokenizer` is used as it is a standard wrapper around sentencepiece. 
+
+- The authors suggest to use the following prompt for image captioning: `f"Generate a coco-style caption.\\n"`
+
+
+## FuyuConfig
+
+[[autodoc]] FuyuConfig
+
+## FuyuForCausalLM
+
+[[autodoc]] FuyuForCausalLM
+    - forward
+
+## FuyuImageProcessor
+
+[[autodoc]] FuyuImageProcessor
+    - __call__
+
+## FuyuProcessor
+
+[[autodoc]] FuyuProcessor
+    - __call__

docs/source/en/model_doc/jukebox.md

@@ -28,7 +28,7 @@ The abstract from the paper is the following:
 
 As shown on the following figure, Jukebox is made of 3 `priors` which are decoder only models. They follow the architecture described in [Generating Long Sequences with Sparse Transformers](https://arxiv.org/abs/1904.10509), modified to support longer context length.
 First, a autoencoder is used to encode the text lyrics. Next, the first (also called `top_prior`) prior attends to the last hidden states extracted from the lyrics encoder. The priors are linked to the previous priors respectively via an `AudioConditionner` module. The`AudioConditioner` upsamples the outputs of the previous prior to raw tokens at a certain audio frame per second resolution. 
-The metadata such as *artist, genre and timing* are passed to each prior, in the form of a start token and positionnal embedding for the timing data.  The hidden states are mapped to the closest codebook vector from the VQVAE in order to convert them to raw audio.
+The metadata such as *artist, genre and timing* are passed to each prior, in the form of a start token and positional embedding for the timing data.  The hidden states are mapped to the closest codebook vector from the VQVAE in order to convert them to raw audio.
 
 ![JukeboxModel](https://gist.githubusercontent.com/ArthurZucker/92c1acaae62ebf1b6a951710bdd8b6af/raw/c9c517bf4eff61393f6c7dec9366ef02bdd059a3/jukebox.svg)
 
@@ -36,7 +36,7 @@ Tips:
 - This model only supports inference. This is for a few reasons, mostly because it requires a crazy amount of memory to train. Feel free to open a PR and add what's missing to have a full integration with the hugging face traineer!
 - This model is very slow, and takes 8h to generate a minute long audio using the 5b top prior on a V100 GPU. In order automaticallay handle the device on which the model should execute, use `accelerate`.
 - Contrary to the paper, the order of the priors goes from `0` to `1` as it felt more intuitive : we sample starting from `0`.
-- Primed sampling (conditionning the sampling on raw audio) requires more memory than ancestral sampling and should be used with `fp16` set to `True`.
+- Primed sampling (conditioning the sampling on raw audio) requires more memory than ancestral sampling and should be used with `fp16` set to `True`.
 
 This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ).
 The original code can be found [here](https://github.com/openai/jukebox).

docs/source/en/model_doc/llama.md

@@ -55,6 +55,28 @@ Based on the original LLaMA model, Meta AI has released some follow-up works:
 
 - **Llama2**: Llama2 is an improved version of Llama with some architectural tweaks (Grouped Query Attention), and is pre-trained on 2Trillion tokens. Refer to the documentation of Llama2 which can be found [here](llama2).
 
+## Resources
+
+A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with LLaMA. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
+
+<PipelineTag pipeline="text-classification"/>
+
+- A [notebook](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-sst2.ipynb#scrollTo=f04ba4d2) on how to use prompt tuning to adapt the LLaMA model for text classification task. 🌎
+
+<PipelineTag pipeline="question-answering"/>
+
+- [StackLLaMA: A hands-on guide to train LLaMA with RLHF](https://huggingface.co/blog/stackllama#stackllama-a-hands-on-guide-to-train-llama-with-rlhf), a blog post about how to train LLaMA to answer questions on [Stack Exchange](https://stackexchange.com/) with RLHF.
+
+⚗️ Optimization
+- A [notebook](https://colab.research.google.com/drive/1SQUXq1AMZPSLD4mk3A3swUIc6Y2dclme?usp=sharing) on how to fine-tune LLaMA model using xturing library on GPU which has limited memory. 🌎 
+
+⚡️ Inference
+- A [notebook](https://colab.research.google.com/github/DominguesM/alpaca-lora-ptbr-7b/blob/main/notebooks/02%20-%20Evaluate.ipynb) on how to run the LLaMA Model using PeftModel from the 🤗 PEFT library. 🌎 
+- A [notebook](https://colab.research.google.com/drive/1l2GiSSPbajVyp2Nk3CFT4t3uH6-5TiBe?usp=sharing) on how to load a PEFT adapter LLaMA model with LangChain. 🌎
+
+🚀 Deploy
+- A [notebook](https://colab.research.google.com/github/lxe/simple-llama-finetuner/blob/master/Simple_LLaMA_FineTuner.ipynb#scrollTo=3PM_DilAZD8T) on how to fine-tune LLaMA model using LoRA method via the 🤗 PEFT library with intuitive UI. 🌎 
+- A [notebook](https://github.com/aws/amazon-sagemaker-examples/blob/main/introduction_to_amazon_algorithms/jumpstart-foundation-models/text-generation-open-llama.ipynb) on how to deploy Open-LLaMA model for text generation on Amazon SageMaker. 🌎 
 
 ## LlamaConfig
 

docs/source/en/model_doc/llama2.md

@@ -78,6 +78,10 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 - A [notebook](https://colab.research.google.com/drive/1PEQyJO1-f6j0S_XJ8DV50NkpzasXkrzd?usp=sharing) on how to fine-tune Llama 2 in Google Colab using QLoRA and 4-bit precision. 🌎
 - A [notebook](https://colab.research.google.com/drive/134o_cXcMe_lsvl15ZE_4Y75Kstepsntu?usp=sharing) on how to fine-tune the "Llama-v2-7b-guanaco" model with 4-bit QLoRA and generate Q&A datasets from PDFs. 🌎
 
+<PipelineTag pipeline="text-classification"/>
+
+- A [notebook](https://colab.research.google.com/drive/1ggaa2oRFphdBmqIjSEbnb_HGkcIRC2ZB?usp=sharing) on how to fine-tune the Llama 2 model with QLoRa, TRL, and Korean text classification dataset. 🌎🇰🇷
+
 ⚗️ Optimization
 - [Fine-tune Llama 2 with DPO](https://huggingface.co/blog/dpo-trl), a guide to using the TRL library's DPO method to fine tune Llama 2 on a specific dataset.
 - [Extended Guide: Instruction-tune Llama 2](https://www.philschmid.de/instruction-tune-llama-2), a guide to training Llama 2 to generate instructions from inputs, transforming the model from instruction-following to instruction-giving.

docs/source/en/model_doc/mistral.md

@@ -0,0 +1,151 @@
+<!--Copyright 2023 Mistral AI and 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Mistral
+
+## Overview
+
+Mistral-7B-v0.1 is Mistral AI’s first Large Language Model (LLM). 
+
+## Model Details
+
+Mistral-7B-v0.1 is a decoder-based LM with the following architectural choices:
+* Sliding Window Attention - Trained with 8k context length and fixed cache size, with a theoretical attention span of 128K tokens
+* GQA (Grouped Query Attention) - allowing faster inference and lower cache size.
+* Byte-fallback BPE tokenizer - ensures that characters are never mapped to out of vocabulary tokens.
+
+We also provide an instruction fine-tuned model: `Mistral-7B-Instruct-v0.1` which can be used for chat-based inference.
+
+For more details please read our [release blog post](https://mistral.ai/news/announcing-mistral-7b/)
+
+## License
+
+Both `Mistral-7B-v0.1` and `Mistral-7B-Instruct-v0.1` are released under the Apache 2.0 license.
+
+## Usage
+
+`Mistral-7B-v0.1` and `Mistral-7B-Instruct-v0.1` can be found on the [Huggingface Hub](https://huggingface.co/mistralai)
+
+These ready-to-use checkpoints can be downloaded and used via the HuggingFace Hub:
+
+```python
+>>> from transformers import AutoModelForCausalLM, AutoTokenizer
+>>> device = "cuda" # the device to load the model onto
+
+>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1")
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
+
+>>> prompt = "My favourite condiment is"
+
+>>> model_inputs = tokenizer([prompt], return_tensors="pt").to(device)
+>>> model.to(device)
+
+>>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
+>>> tokenizer.batch_decode(generated_ids)[0]
+"The expected output"
+```
+
+Raw weights for `Mistral-7B-v0.1` and `Mistral-7B-Instruct-v0.1` can be downloaded from:
+
+| Model Name                 | Checkpoint                                                                              |
+|----------------------------|-----------------------------------------------------------------------------------------|
+| `Mistral-7B-v0.1`          | [Raw Checkpoint](https://files.mistral-7b-v0-1.mistral.ai/mistral-7B-v0.1.tar)          |
+| `Mistral-7B-Instruct-v0.1` | [Raw Checkpoint](https://files.mistral-7b-v0-1.mistral.ai/mistral-7B-instruct-v0.1.tar) |
+
+
+To use these raw checkpoints with HuggingFace you can use the `convert_mistral_weights_to_hf.py` script to convert them to the HuggingFace format:
+
+```bash
+python src/transformers/models/mistral/convert_mistral_weights_to_hf.py \
+    --input_dir /path/to/downloaded/mistral/weights --model_size 7B --output_dir /output/path
+```
+
+You can then load the converted model from the `output/path`:
+
+```python
+from transformers import MistralForCausalLM, LlamaTokenizer
+
+tokenizer = LlamaTokenizer.from_pretrained("/output/path")
+model = MistralForCausalLM.from_pretrained("/output/path")
+```
+
+## Combining Mistral and Flash Attention 2
+
+First, make sure to install the latest version of Flash Attention 2 to include the sliding window attention feature.
+
+```bash
+pip install -U flash-attn --no-build-isolation
+```
+
+Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of [`flash-attn`](https://github.com/Dao-AILab/flash-attention) repository. Make also sure to load your model in half-precision (e.g. `torch.float16`)
+
+To load and run a model using Flash Attention 2, refer to the snippet below:
+
+```python
+>>> import torch
+>>> from transformers import AutoModelForCausalLM, AutoTokenizer
+>>> device = "cuda" # the device to load the model onto
+
+>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1", torch_dtype=torch.float16, use_flash_attention_2=True)
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
+
+>>> prompt = "My favourite condiment is"
+
+>>> model_inputs = tokenizer([prompt], return_tensors="pt").to(device)
+>>> model.to(device)
+
+>>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
+>>> tokenizer.batch_decode(generated_ids)[0]
+"The expected output"
+```
+
+### Expected speedups
+
+Below is a expected speedup diagram that compares pure inference time between the native implementation in transformers using `mistralai/Mistral-7B-v0.1` checkpoint and the Flash Attention 2 version of the model.
+
+<div style="text-align: center">
+<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/mistral-7b-inference-large-seqlen.png">
+</div>
+
+### Sliding window Attention
+
+The current implementation supports the sliding window attention mechanism and memory efficient cache management. 
+To enable sliding window attention, just make sure to have a `flash-attn` version that is compatible with sliding window attention (`>=2.3.0`). 
+
+The Flash Attention-2 model uses also a more memory efficient cache slicing mechanism - as recommended per the official implementation of Mistral model that use rolling cache mechanism we keep the cache size fixed (`self.config.sliding_window`), support batched generation only for `padding_side="left"` and use the absolute position of the current token to compute the positional embedding.
+
+## The Mistral Team
+
+Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
+
+## MistralConfig
+
+[[autodoc]] MistralConfig
+
+## MistralModel
+
+[[autodoc]] MistralModel
+    - forward
+
+## MistralForCausalLM
+
+[[autodoc]] MistralForCausalLM
+    - forward
+
+## MistralForSequenceClassification
+
+[[autodoc]] MistralForSequenceClassification
+    - forward

docs/source/en/model_doc/mms.md

@@ -165,7 +165,147 @@ To further improve performance from ASR models, language model decoding can be u
 
 ### Speech Synthesis (TTS)
 
-Individual TTS models are available for each of the 1100+ languages. The models and inference documentation can be found [here](https://huggingface.co/facebook/mms-tts).
+MMS-TTS uses the same model architecture as VITS, which was added to 🤗 Transformers in v4.33. MMS trains a separate 
+model checkpoint for each of the 1100+ languages in the project. All available checkpoints can be found on the Hugging 
+Face Hub: [facebook/mms-tts](https://huggingface.co/models?sort=trending&search=facebook%2Fmms-tts), and the inference 
+documentation under [VITS](https://huggingface.co/docs/transformers/main/en/model_doc/vits).
+
+#### Inference
+
+To use the MMS model, first update to the latest version of the Transformers library:
+
+```bash
+pip install --upgrade transformers accelerate
+```
+
+Since the flow-based model in VITS is non-deterministic, it is good practice to set a seed to ensure reproducibility of 
+the outputs. 
+
+- For languages with a Roman alphabet, such as English or French, the tokenizer can be used directly to 
+pre-process the text inputs. The following code example runs a forward pass using the MMS-TTS English checkpoint:
+
+```python
+import torch
+from transformers import VitsTokenizer, VitsModel, set_seed
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
+model = VitsModel.from_pretrained("facebook/mms-tts-eng")
+
+inputs = tokenizer(text="Hello - my dog is cute", return_tensors="pt")
+
+set_seed(555)  # make deterministic
+
+with torch.no_grad():
+   outputs = model(**inputs)
+
+waveform = outputs.waveform[0]
+```
+
+The resulting waveform can be saved as a `.wav` file:
+
+```python
+import scipy
+
+scipy.io.wavfile.write("synthesized_speech.wav", rate=model.config.sampling_rate, data=waveform)
+```
+
+Or displayed in a Jupyter Notebook / Google Colab:
+
+```python
+from IPython.display import Audio
+
+Audio(waveform, rate=model.config.sampling_rate)
+```
+
+For certain languages with non-Roman alphabets, such as Arabic, Mandarin or Hindi, the [`uroman`](https://github.com/isi-nlp/uroman) 
+perl package is required to pre-process the text inputs to the Roman alphabet.
+
+You can check whether you require the `uroman` package for your language by inspecting the `is_uroman` attribute of 
+the pre-trained `tokenizer`:
+
+```python
+from transformers import VitsTokenizer
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
+print(tokenizer.is_uroman)
+```
+
+If required, you should apply the uroman package to your text inputs **prior** to passing them to the `VitsTokenizer`, 
+since currently the tokenizer does not support performing the pre-processing itself.
+
+To do this, first clone the uroman repository to your local machine and set the bash variable `UROMAN` to the local path:
+
+```bash
+git clone https://github.com/isi-nlp/uroman.git
+cd uroman
+export UROMAN=$(pwd)
+```
+
+You can then pre-process the text input using the following code snippet. You can either rely on using the bash variable 
+`UROMAN` to point to the uroman repository, or you can pass the uroman directory as an argument to the `uromaize` function:
+
+```python
+import torch
+from transformers import VitsTokenizer, VitsModel, set_seed
+import os
+import subprocess
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-kor")
+model = VitsModel.from_pretrained("facebook/mms-tts-kor")
+
+def uromanize(input_string, uroman_path):
+    """Convert non-Roman strings to Roman using the `uroman` perl package."""
+    script_path = os.path.join(uroman_path, "bin", "uroman.pl")
+
+    command = ["perl", script_path]
+
+    process = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    # Execute the perl command
+    stdout, stderr = process.communicate(input=input_string.encode())
+
+    if process.returncode != 0:
+        raise ValueError(f"Error {process.returncode}: {stderr.decode()}")
+
+    # Return the output as a string and skip the new-line character at the end
+    return stdout.decode()[:-1]
+
+text = "이봐 무슨 일이야"
+uromaized_text = uromanize(text, uroman_path=os.environ["UROMAN"])
+
+inputs = tokenizer(text=uromaized_text, return_tensors="pt")
+
+set_seed(555)  # make deterministic
+with torch.no_grad():
+   outputs = model(inputs["input_ids"])
+
+waveform = outputs.waveform[0]
+```
+
+**Tips:**
+
+* The MMS-TTS checkpoints are trained on lower-cased, un-punctuated text. By default, the `VitsTokenizer` *normalizes* the inputs by removing any casing and punctuation, to avoid passing out-of-vocabulary characters to the model. Hence, the model is agnostic to casing and punctuation, so these should be avoided in the text prompt. You can disable normalisation by setting `noramlize=False` in the call to the tokenizer, but this will lead to un-expected behaviour and is discouraged.
+* The speaking rate can be varied by setting the attribute `model.speaking_rate` to a chosen value. Likewise, the randomness of the noise is controlled by `model.noise_scale`:
+
+```python
+import torch
+from transformers import VitsTokenizer, VitsModel, set_seed
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
+model = VitsModel.from_pretrained("facebook/mms-tts-eng")
+
+inputs = tokenizer(text="Hello - my dog is cute", return_tensors="pt")
+
+# make deterministic
+set_seed(555)  
+
+# make speech faster and more noisy
+model.speaking_rate = 1.5
+model.noise_scale = 0.8
+
+with torch.no_grad():
+   outputs = model(**inputs)
+```
+
 
 ### Language Identification (LID)
 
@@ -173,11 +313,12 @@ Different LID models are available based on the number of languages they can rec
 
 #### Inference
 First, we install transformers and some other libraries
-```
-pip install torch accelerate torchaudio datasets
+
+```bash
+pip install torch accelerate datasets[audio]
 pip install --upgrade transformers
 ````
-pip install torch datasets[audio]
+
 Next, we load a couple of audio samples via `datasets`. Make sure that the audio data is sampled to 16000 kHz.
 
 ```py

docs/source/en/model_doc/mra.md

@@ -22,7 +22,7 @@ The MRA model was proposed in [Multi Resolution Analysis (MRA) for Approximate S
 
 The abstract from the paper is the following:
 
-*Transformers have emerged as a preferred model for many tasks in natural langugage processing and vision. Recent efforts on training and deploying Transformers more efficiently have identified many strategies to approximate the self-attention matrix, a key module in a Transformer architecture. Effective ideas include various prespecified sparsity patterns, low-rank basis expansions and combinations thereof. In this paper, we revisit classical Multiresolution Analysis (MRA) concepts such as Wavelets, whose potential value in this setting remains underexplored thus far. We show that simple approximations based on empirical feedback and design choices informed by modern hardware and implementation challenges, eventually yield a MRA-based approach for self-attention with an excellent performance profile across most criteria of interest. We undertake an extensive set of experiments and demonstrate that this multi-resolution scheme outperforms most efficient self-attention proposals and is favorable for both short and long sequences. Code is available at https://github.com/mlpen/mra-attention.*
+*Transformers have emerged as a preferred model for many tasks in natural language processing and vision. Recent efforts on training and deploying Transformers more efficiently have identified many strategies to approximate the self-attention matrix, a key module in a Transformer architecture. Effective ideas include various prespecified sparsity patterns, low-rank basis expansions and combinations thereof. In this paper, we revisit classical Multiresolution Analysis (MRA) concepts such as Wavelets, whose potential value in this setting remains underexplored thus far. We show that simple approximations based on empirical feedback and design choices informed by modern hardware and implementation challenges, eventually yield a MRA-based approach for self-attention with an excellent performance profile across most criteria of interest. We undertake an extensive set of experiments and demonstrate that this multi-resolution scheme outperforms most efficient self-attention proposals and is favorable for both short and long sequences. Code is available at https://github.com/mlpen/mra-attention.*
 
 This model was contributed by [novice03](https://huggingface.co/novice03).
 The original code can be found [here](https://github.com/mlpen/mra-attention).

docs/source/en/model_doc/musicgen.md

@@ -214,28 +214,7 @@ The MusicGen model can be de-composed into three distinct stages:
 
 Thus, the MusicGen model can either be used as a standalone decoder model, corresponding to the class [`MusicgenForCausalLM`],
 or as a composite model that includes the text encoder and audio encoder/decoder, corresponding to the class
-[`MusicgenForConditionalGeneration`].
-
-Since the text encoder and audio encoder/decoder models are frozen during training, the MusicGen decoder [`MusicgenForCausalLM`]
-can be trained standalone on a dataset of encoder hidden-states and audio codes. For inference, the trained decoder can
-be combined with the frozen text encoder and audio encoder/decoders to recover the composite [`MusicgenForConditionalGeneration`]
-model.
-
-Below, we demonstrate how to construct the composite [`MusicgenForConditionalGeneration`] model from its three constituent
-parts, as would typically be done following training of the MusicGen decoder LM:
-
-```python
->>> from transformers import AutoConfig, AutoModelForTextEncoding, AutoModel, MusicgenForCausalLM, MusicgenForConditionalGeneration
-
->>> text_encoder = AutoModelForTextEncoding.from_pretrained("t5-base")
->>> audio_encoder = AutoModel.from_pretrained("facebook/encodec_32khz")
->>> decoder_config = AutoConfig.from_pretrained("facebook/musicgen-small").decoder
->>> decoder = MusicgenForCausalLM.from_pretrained("facebook/musicgen-small", **decoder_config)
-
->>> model = MusicgenForConditionalGeneration.from_sub_models_pretrained(text_encoder, audio_encoder, decoder)
-```
-
-If only the decoder needs to be loaded from the pre-trained checkpoint for the composite model, it can be loaded by first 
+[`MusicgenForConditionalGeneration`]. If only the decoder needs to be loaded from the pre-trained checkpoint, it can be loaded by first 
 specifying the correct config, or be accessed through the `.decoder` attribute of the composite model:
 
 ```python
@@ -249,6 +228,11 @@ specifying the correct config, or be accessed through the `.decoder` attribute o
 >>> decoder = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small").decoder
 ```
 
+Since the text encoder and audio encoder/decoder models are frozen during training, the MusicGen decoder [`MusicgenForCausalLM`]
+can be trained standalone on a dataset of encoder hidden-states and audio codes. For inference, the trained decoder can
+be combined with the frozen text encoder and audio encoder/decoders to recover the composite [`MusicgenForConditionalGeneration`]
+model.
+
 Tips:
 * MusicGen is trained on the 32kHz checkpoint of Encodec. You should ensure you use a compatible version of the Encodec model.
 * Sampling mode tends to deliver better results than greedy - you can toggle sampling with the variable `do_sample` in the call to [`MusicgenForConditionalGeneration.generate`]

docs/source/en/model_doc/nllb-moe.md

@@ -53,7 +53,7 @@ which means that tokens have less probability of being forwarded. Moreover, if a
 states (kind of like a residual connection) while they are masked in `NLLB`'s top-2 routing mechanism. 
 
 ## Generating with NLLB-MoE
-The avalable checkpoints requires around 350GB of storage. Make sure to use `accelerate` if you do not have enough RAM on your machine.
+The available checkpoints require around 350GB of storage. Make sure to use `accelerate` if you do not have enough RAM on your machine.
 
 While generating the target text set the `forced_bos_token_id` to the target language id. The following
 example shows how to translate English to French using the *facebook/nllb-200-distilled-600M* model.

docs/source/en/model_doc/nougat.md

@@ -0,0 +1,109 @@
+<!--Copyright 2023 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
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+specific language governing permissions and limitations under the License. -->
+
+# Nougat
+
+## Overview
+
+The Nougat model was proposed in [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by
+Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. Nougat uses the same architecture as [Donut](donut), meaning an image Transformer
+encoder and an autoregressive text Transformer decoder to translate scientific PDFs to markdown, enabling easier access to them.
+
+The abstract from the paper is the following:
+
+*Scientific knowledge is predominantly stored in books and scientific journals, often in the form of PDFs. However, the PDF format leads to a loss of semantic information, particularly for mathematical expressions. We propose Nougat (Neural Optical Understanding for Academic Documents), a Visual Transformer model that performs an Optical Character Recognition (OCR) task for processing scientific documents into a markup language, and demonstrate the effectiveness of our model on a new dataset of scientific documents. The proposed approach offers a promising solution to enhance the accessibility of scientific knowledge in the digital age, by bridging the gap between human-readable documents and machine-readable text. We release the models and code to accelerate future work on scientific text recognition.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/nougat_architecture.jpg"
+alt="drawing" width="600"/>
+
+<small> Nougat high-level overview. Taken from the <a href="https://arxiv.org/abs/2308.13418">original paper</a>. </small>
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code can be found
+[here](https://github.com/facebookresearch/nougat).
+
+Tips:
+
+- The quickest way to get started with Nougat is by checking the [tutorial
+  notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/Nougat), which show how to use the model
+  at inference time as well as fine-tuning on custom data.
+- Nougat is always used within the [VisionEncoderDecoder](vision-encoder-decoder) framework. The model is identical to [Donut](donut) in terms of architecture.
+
+## Inference
+
+Nougat's [`VisionEncoderDecoder`] model accepts images as input and makes use of
+[`~generation.GenerationMixin.generate`] to autoregressively generate text given the input image.
+
+The [`NougatImageProcessor`] class is responsible for preprocessing the input image and
+[`NougatTokenizerFast`] decodes the generated target tokens to the target string. The
+[`NougatProcessor`] wraps [`NougatImageProcessor`] and [`NougatTokenizerFast`] classes
+into a single instance to both extract the input features and decode the predicted token ids.
+
+- Step-by-step PDF transcription
+
+```py
+>>> from huggingface_hub import hf_hub_download
+>>> import re
+>>> from PIL import Image
+
+>>> from transformers import NougatProcessor, VisionEncoderDecoderModel
+>>> from datasets import load_dataset
+>>> import torch
+
+>>> processor = NougatProcessor.from_pretrained("facebook/nougat-base")
+>>> model = VisionEncoderDecoderModel.from_pretrained("facebook/nougat-base")
+
+>>> device = "cuda" if torch.cuda.is_available() else "cpu"
+>>> model.to(device)  # doctest: +IGNORE_RESULT
+
+>>> # prepare PDF image for the model
+>>> filepath = hf_hub_download(repo_id="hf-internal-testing/fixtures_docvqa", filename="nougat_paper.png", repo_type="dataset")
+>>> image = Image.open(filepath)
+>>> pixel_values = processor(image, return_tensors="pt").pixel_values
+
+>>> # generate transcription (here we only generate 30 tokens)
+>>> outputs = model.generate(
+...     pixel_values.to(device),
+...     min_length=1,
+...     max_new_tokens=30,
+...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
+... )
+
+>>> sequence = processor.batch_decode(outputs, skip_special_tokens=True)[0]
+>>> sequence = processor.post_process_generation(sequence, fix_markdown=False)
+>>> # note: we're using repr here such for the sake of printing the \n characters, feel free to just print the sequence
+>>> print(repr(sequence))
+'\n\n# Nougat: Neural Optical Understanding for Academic Documents\n\n Lukas Blecher\n\nCorrespondence to: lblecher@'
+```
+
+See the [model hub](https://huggingface.co/models?filter=nougat) to look for Nougat checkpoints.
+
+## NougatImageProcessor
+
+[[autodoc]] NougatImageProcessor
+    - preprocess
+
+## NougatTokenizerFast
+
+[[autodoc]] NougatTokenizerFast
+
+## NougatProcessor
+
+[[autodoc]] NougatProcessor
+    - __call__
+    - from_pretrained
+    - save_pretrained
+    - batch_decode
+    - decode
+    - post_process_generation

docs/source/en/model_doc/owlv2.md

@@ -0,0 +1,126 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# OWLv2
+
+## Overview
+
+OWLv2 was proposed in [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby. OWLv2 scales up [OWL-ViT](owlvit) using self-training, which uses an existing detector to generate pseudo-box annotations on image-text pairs. This results in large gains over the previous state-of-the-art for zero-shot object detection.
+
+The abstract from the paper is the following:
+
+*Open-vocabulary object detection has benefited greatly from pretrained vision-language models, but is still limited by the amount of available detection training data. While detection training data can be expanded by using Web image-text pairs as weak supervision, this has not been done at scales comparable to image-level pretraining. Here, we scale up detection data with self-training, which uses an existing detector to generate pseudo-box annotations on image-text pairs. Major challenges in scaling self-training are the choice of label space, pseudo-annotation filtering, and training efficiency. We present the OWLv2 model and OWL-ST self-training recipe, which address these challenges. OWLv2 surpasses the performance of previous state-of-the-art open-vocabulary detectors already at comparable training scales (~10M examples). However, with OWL-ST, we can scale to over 1B examples, yielding further large improvement: With an L/14 architecture, OWL-ST improves AP on LVIS rare classes, for which the model has seen no human box annotations, from 31.2% to 44.6% (43% relative improvement). OWL-ST unlocks Web-scale training for open-world localization, similar to what has been seen for image classification and language modelling.*
+
+Tips:
+
+- The architecture of OWLv2 is identical to [OWL-ViT](owlvit), however the object detection head now also includes an objectness classifier, which predicts the (query-agnostic) likelihood that a predicted box contains an object (as opposed to background). The objectness score can be used to rank or filter predictions independently of text queries.
+- Usage of OWLv2 is identical to [OWL-ViT](owlvit) with a new, updated image processor ([`Owlv2ImageProcessor`]).
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/owlv2_overview.png"
+alt="drawing" width="600"/>
+
+<small> OWLv2 high-level overview. Taken from the <a href="https://arxiv.org/abs/2306.09683">original paper</a>. </small>
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr).
+The original code can be found [here](https://github.com/google-research/scenic/tree/main/scenic/projects/owl_vit).
+
+## Usage
+
+OWLv2 is, just like its predecessor [OWL-ViT](owlvit), a zero-shot text-conditioned object detection model. OWL-ViT uses [CLIP](clip) as its multi-modal backbone, with a ViT-like Transformer to get visual features and a causal language model to get the text features. To use CLIP for detection, OWL-ViT removes the final token pooling layer of the vision model and attaches a lightweight classification and box head to each transformer output token. Open-vocabulary classification is enabled by replacing the fixed classification layer weights with the class-name embeddings obtained from the text model. The authors first train CLIP from scratch and fine-tune it end-to-end with the classification and box heads on standard detection datasets using a bipartite matching loss. One or multiple text queries per image can be used to perform zero-shot text-conditioned object detection.
+
+[`Owlv2ImageProcessor`] can be used to resize (or rescale) and normalize images for the model and [`CLIPTokenizer`] is used to encode the text. [`Owlv2Processor`] wraps [`Owlv2ImageProcessor`] and [`CLIPTokenizer`] into a single instance to both encode the text and prepare the images. The following example shows how to perform object detection using [`Owlv2Processor`] and [`Owlv2ForObjectDetection`].
+
+
+```python
+>>> import requests
+>>> from PIL import Image
+>>> import torch
+
+>>> from transformers import Owlv2Processor, Owlv2ForObjectDetection
+
+>>> processor = Owlv2Processor.from_pretrained("google/owlv2-base-patch16-ensemble")
+>>> model = Owlv2ForObjectDetection.from_pretrained("google/owlv2-base-patch16-ensemble")
+
+>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+>>> image = Image.open(requests.get(url, stream=True).raw)
+>>> texts = [["a photo of a cat", "a photo of a dog"]]
+>>> inputs = processor(text=texts, images=image, return_tensors="pt")
+>>> outputs = model(**inputs)
+
+>>> # Target image sizes (height, width) to rescale box predictions [batch_size, 2]
+>>> target_sizes = torch.Tensor([image.size[::-1]])
+>>> # Convert outputs (bounding boxes and class logits) to COCO API
+>>> results = processor.post_process_object_detection(outputs=outputs, target_sizes=target_sizes, threshold=0.1)
+>>> i = 0  # Retrieve predictions for the first image for the corresponding text queries
+>>> text = texts[i]
+>>> boxes, scores, labels = results[i]["boxes"], results[i]["scores"], results[i]["labels"]
+>>> for box, score, label in zip(boxes, scores, labels):
+...     box = [round(i, 2) for i in box.tolist()]
+...     print(f"Detected {text[label]} with confidence {round(score.item(), 3)} at location {box}")
+Detected a photo of a cat with confidence 0.614 at location [341.67, 17.54, 642.32, 278.51]
+Detected a photo of a cat with confidence 0.665 at location [6.75, 38.97, 326.62, 354.85]
+```
+
+## Resources
+
+A demo notebook on using OWLv2 for zero- and one-shot (image-guided) object detection can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/OWLv2).
+
+## Owlv2Config
+
+[[autodoc]] Owlv2Config
+    - from_text_vision_configs
+
+## Owlv2TextConfig
+
+[[autodoc]] Owlv2TextConfig
+
+## Owlv2VisionConfig
+
+[[autodoc]] Owlv2VisionConfig
+
+## Owlv2ImageProcessor
+
+[[autodoc]] Owlv2ImageProcessor
+    - preprocess
+    - post_process_object_detection
+    - post_process_image_guided_detection
+
+## Owlv2Processor
+
+[[autodoc]] Owlv2Processor
+
+## Owlv2Model
+
+[[autodoc]] Owlv2Model
+    - forward
+    - get_text_features
+    - get_image_features
+
+## Owlv2TextModel
+
+[[autodoc]] Owlv2TextModel
+    - forward
+
+## Owlv2VisionModel
+
+[[autodoc]] Owlv2VisionModel
+    - forward
+
+## Owlv2ForObjectDetection
+
+[[autodoc]] Owlv2ForObjectDetection
+    - forward
+    - image_guided_detection

docs/source/en/model_doc/owlvit.md

@@ -24,6 +24,13 @@ The abstract from the paper is the following:
 
 *Combining simple architectures with large-scale pre-training has led to massive improvements in image classification. For object detection, pre-training and scaling approaches are less well established, especially in the long-tailed and open-vocabulary setting, where training data is relatively scarce. In this paper, we propose a strong recipe for transferring image-text models to open-vocabulary object detection. We use a standard Vision Transformer architecture with minimal modifications, contrastive image-text pre-training, and end-to-end detection fine-tuning. Our analysis of the scaling properties of this setup shows that increasing image-level pre-training and model size yield consistent improvements on the downstream detection task. We provide the adaptation strategies and regularizations needed to attain very strong performance on zero-shot text-conditioned and one-shot image-conditioned object detection. Code and models are available on GitHub.*
 
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/owlvit_architecture.jpg"
+alt="drawing" width="600"/>
+
+<small> OWL-ViT architecture. Taken from the <a href="https://arxiv.org/abs/2205.06230">original paper</a>. </small>
+
+This model was contributed by [adirik](https://huggingface.co/adirik). The original code can be found [here](https://github.com/google-research/scenic/tree/main/scenic/projects/owl_vit).
+
 ## Usage
 
 OWL-ViT is a zero-shot text-conditioned object detection model. OWL-ViT uses [CLIP](clip) as its multi-modal backbone, with a ViT-like Transformer to get visual features and a causal language model to get the text features. To use CLIP for detection, OWL-ViT removes the final token pooling layer of the vision model and attaches a lightweight classification and box head to each transformer output token. Open-vocabulary classification is enabled by replacing the fixed classification layer weights with the class-name embeddings obtained from the text model. The authors first train CLIP from scratch and fine-tune it end-to-end with the classification and box heads on standard detection datasets using a bipartite matching loss. One or multiple text queries per image can be used to perform zero-shot text-conditioned object detection.
@@ -61,7 +68,9 @@ Detected a photo of a cat with confidence 0.707 at location [324.97, 20.44, 640.
 Detected a photo of a cat with confidence 0.717 at location [1.46, 55.26, 315.55, 472.17]
 ```
 
-This model was contributed by [adirik](https://huggingface.co/adirik). The original code can be found [here](https://github.com/google-research/scenic/tree/main/scenic/projects/owl_vit).
+## Resources
+
+A demo notebook on using OWL-ViT for zero- and one-shot (image-guided) object detection can be found [here](https://github.com/huggingface/notebooks/blob/main/examples/zeroshot_object_detection_with_owlvit.ipynb).
 
 ## OwlViTConfig
 

docs/source/en/model_doc/persimmon.md

@@ -0,0 +1,96 @@
+<!--Copyright 2023 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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Persimmon
+
+## Overview
+
+The Persimmon model was created by [ADEPT](https://www.adept.ai/blog/persimmon-8b), and authored by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
+
+The authors introduced Persimmon-8B, a decoder model based on the classic transformers architecture, with query and key normalization. Persimmon-8B is a fully permissively-licensed model with approximately 8 billion parameters, released under the Apache license.  Some of the key attributes of Persimmon-8B are long context size (16K), performance, and capabilities for multimodal extensions.
+
+The authors showcase their approach to model evaluation, focusing on practical text generation, mirroring how users interact with language models. The work also includes a comparative analysis, pitting Persimmon-8B against other prominent models (MPT 7B Instruct and Llama 2 Base 7B 1-Shot), across various evaluation tasks. The results demonstrate Persimmon-8B's competitive performance, even with limited training data.
+
+In terms of model details, the work outlines the architecture and training methodology of Persimmon-8B, providing insights into its design choices, sequence length, and dataset composition. The authors present a fast inference code that outperforms traditional implementations through operator fusion and CUDA graph utilization while maintaining code coherence. They express their anticipation of how the community will leverage this contribution to drive innovation, hinting at further upcoming releases as part of an ongoing series of developments.
+
+
+<Tip warning={true}>
+
+The `Persimmon` models were trained using `bfloat16`, but the original inference uses `float16` The checkpoints uploaded on the hub use `torch_dtype = 'float16'` which will be
+used by the `AutoModel` API to cast the checkpoints from `torch.float32` to `torch.float16`. 
+
+The `dtype` of the online weights is mostly irrelevant, unless you are using `torch_dtype="auto"` when initializing a model using `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`. The reason is that the model will first be downloaded ( using the `dtype` of the checkpoints online) then it will be cast to the default `dtype` of `torch` (becomes `torch.float32`). Users should specify the `torch_dtype` they want, and if they don't it will be `torch.float32`.
+
+Finetuning the model in `float16` is not recommended and known to produce `nan`, as such the model should be fine-tuned in `bfloat16`.
+
+</Tip>
+
+
+Tips:
+
+- To convert the model, you need to clone the original repository using `git clone https://github.com/persimmon-ai-labs/adept-inference`, then get the checkpoints:
+
+```bash
+git clone https://github.com/persimmon-ai-labs/adept-inference
+wget https://axtkn4xl5cip.objectstorage.us-phoenix-1.oci.customer-oci.com/n/axtkn4xl5cip/b/adept-public-data/o/8b_base_model_release.tar
+tar -xvf 8b_base_model_release.tar
+python src/transformers/models/persimmon/convert_persimmon_weights_to_hf.py  --input_dir /path/to/downloaded/persimmon/weights/ --output_dir /output/path \
+    --pt_model_path /path/to/8b_chat_model_release/iter_0001251/mp_rank_00/model_optim_rng.pt
+    --ada_lib_path /path/to/adept-inference
+```
+
+For the chat model:
+```bash
+wget https://axtkn4xl5cip.objectstorage.us-phoenix-1.oci.customer-oci.com/n/axtkn4xl5cip/b/adept-public-data/o/8b_chat_model_release.tar
+tar -xvf 8b_base_model_release.tar
+```
+
+Thereafter, models can be loaded via:
+
+```py
+from transformers import PersimmonForCausalLM, PersimmonTokenizer
+
+model = PersimmonForCausalLM.from_pretrained("/output/path")
+tokenizer = PersimmonTokenizer.from_pretrained("/output/path")
+```
+
+This model was contributed by [ArthurZ](https://huggingface.co/ArthurZ).
+The original code can be found [here](https://github.com/persimmon-ai-labs/adept-inference).
+
+- Perismmon uses a `sentencepiece` based tokenizer, with a `Unigram` model. It supports bytefallback, which is only available in `tokenizers==0.14.0` for the fast tokenizer.
+The `LlamaTokenizer` is used as it is a standard wrapper around sentencepiece. The `chat` template will be updated with the templating functions in a follow up PR!
+
+- The authors suggest to use the following prompt format for the chat mode: `f"human: {prompt}\n\nadept:"`
+
+
+## PersimmonConfig
+
+[[autodoc]] PersimmonConfig
+
+## PersimmonModel
+
+[[autodoc]] PersimmonModel
+    - forward
+
+## PersimmonForCausalLM
+
+[[autodoc]] PersimmonForCausalLM
+    - forward
+
+## PersimmonForSequenceClassification
+
+[[autodoc]] PersimmonForSequenceClassification
+    - forward

docs/source/en/model_doc/pop2piano.md

@@ -12,6 +12,12 @@ specific language governing permissions and limitations under the License.
 
 # Pop2Piano
 
+<div class="flex flex-wrap space-x-1">
+<a href="https://huggingface.co/spaces/sweetcocoa/pop2piano">
+<img alt="Spaces" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue">
+</a>
+</div>
+
 ## Overview
 
 The Pop2Piano model was proposed in [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.

docs/source/en/model_doc/videomae.md

@@ -47,7 +47,7 @@ review it! The resource should ideally demonstrate something new instead of dupl
 **Video classification**
 - [A notebook](https://github.com/huggingface/notebooks/blob/main/examples/video_classification.ipynb) that shows how
 to fine-tune a VideoMAE model on a custom dataset.
-- [Video classification task guide](../tasks/video-classification)
+- [Video classification task guide](../tasks/video_classification)
 - [A 🤗 Space](https://huggingface.co/spaces/sayakpaul/video-classification-ucf101-subset) showing how to perform inference with a video classification model.
 
 

docs/source/en/model_doc/vitmatte.md

@@ -0,0 +1,55 @@
+<!--Copyright 2023 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.
+-->
+
+# ViTMatte
+
+## Overview
+
+The ViTMatte model was proposed in [Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
+ViTMatte leverages plain [Vision Transformers](vit) for the task of image matting, which is the process of accurately estimating the foreground object in images and videos.
+
+The abstract from the paper is the following:
+
+*Recently, plain vision Transformers (ViTs) have shown impressive performance on various computer vision tasks, thanks to their strong modeling capacity and large-scale pretraining. However, they have not yet conquered the problem of image matting. We hypothesize that image matting could also be boosted by ViTs and present a new efficient and robust ViT-based matting system, named ViTMatte. Our method utilizes (i) a hybrid attention mechanism combined with a convolution neck to help ViTs achieve an excellent performance-computation trade-off in matting tasks. (ii) Additionally, we introduce the detail capture module, which just consists of simple lightweight convolutions to complement the detailed information required by matting. To the best of our knowledge, ViTMatte is the first work to unleash the potential of ViT on image matting with concise adaptation. It inherits many superior properties from ViT to matting, including various pretraining strategies, concise architecture design, and flexible inference strategies. We evaluate ViTMatte on Composition-1k and Distinctions-646, the most commonly used benchmark for image matting, our method achieves state-of-the-art performance and outperforms prior matting works by a large margin.*
+
+Tips:
+
+- The model expects both the image and trimap (concatenated) as input. One can use [`ViTMatteImageProcessor`] for this purpose.
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr).
+The original code can be found [here](https://github.com/hustvl/ViTMatte).
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/vitmatte_architecture.png"
+alt="drawing" width="600"/>
+
+<small> ViTMatte high-level overview. Taken from the <a href="https://arxiv.org/abs/2305.15272">original paper.</a> </small>
+
+## Resources
+
+A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with ViTMatte.
+
+- A demo notebook regarding inference with [`VitMatteForImageMatting`], including background replacement, can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/ViTMatte).
+
+
+## VitMatteConfig
+
+[[autodoc]] VitMatteConfig
+
+## VitMatteImageProcessor
+
+[[autodoc]] VitMatteImageProcessor
+    - preprocess
+
+## VitMatteForImageMatting
+
+[[autodoc]] VitMatteForImageMatting
+    - forward

docs/source/en/model_doc/vits.md

@@ -0,0 +1,162 @@
+<!--Copyright 2023 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.
+-->
+
+# VITS
+
+## Overview
+
+The VITS model was proposed in [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
+
+
+VITS (**V**ariational **I**nference with adversarial learning for end-to-end **T**ext-to-**S**peech) is an end-to-end 
+speech synthesis model that predicts a speech waveform conditional on an input text sequence. It is a conditional variational 
+autoencoder (VAE) comprised of a posterior encoder, decoder, and conditional prior.
+
+A set of spectrogram-based acoustic features are predicted by the flow-based module, which is formed of a Transformer-based
+text encoder and multiple coupling layers. The spectrogram is decoded using a stack of transposed convolutional layers,
+much in the same style as the HiFi-GAN vocoder. Motivated by the one-to-many nature of the TTS problem, where the same text 
+input can be spoken in multiple ways, the model also includes a stochastic duration predictor, which allows the model to 
+synthesise speech with different rhythms from the same input text. 
+
+The model is trained end-to-end with a combination of losses derived from variational lower bound and adversarial training. 
+To improve the expressiveness of the model, normalizing flows are applied to the conditional prior distribution. During 
+inference, the text encodings are up-sampled based on the duration prediction module, and then mapped into the 
+waveform using a cascade of the flow module and HiFi-GAN decoder. Due to the stochastic nature of the duration predictor,
+the model is non-deterministic, and thus requires a fixed seed to generate the same speech waveform.
+
+The abstract from the paper is the following:
+
+*Several recent end-to-end text-to-speech (TTS) models enabling single-stage training and parallel sampling have been proposed, but their sample quality does not match that of two-stage TTS systems. In this work, we present a parallel end-to-end TTS method that generates more natural sounding audio than current two-stage models. Our method adopts variational inference augmented with normalizing flows and an adversarial training process, which improves the expressive power of generative modeling. We also propose a stochastic duration predictor to synthesize speech with diverse rhythms from input text. With the uncertainty modeling over latent variables and the stochastic duration predictor, our method expresses the natural one-to-many relationship in which a text input can be spoken in multiple ways with different pitches and rhythms. A subjective human evaluation (mean opinion score, or MOS) on the LJ Speech, a single speaker dataset, shows that our method outperforms the best publicly available TTS systems and achieves a MOS comparable to ground truth.*
+
+This model can also be used with TTS checkpoints from [Massively Multilingual Speech (MMS)](https://arxiv.org/abs/2305.13516) 
+as these checkpoints use the same architecture and a slightly modified tokenizer.
+
+This model was contributed by [Matthijs](https://huggingface.co/Matthijs) and [sanchit-gandhi](https://huggingface.co/sanchit-gandhi). The original code can be found [here](https://github.com/jaywalnut310/vits).
+
+## Model Usage
+
+Both the VITS and MMS-TTS checkpoints can be used with the same API. Since the flow-based model is non-deterministic, it 
+is good practice to set a seed to ensure reproducibility of the outputs. For languages with a Roman alphabet, 
+such as English or French, the tokenizer can be used directly to pre-process the text inputs. The following code example 
+runs a forward pass using the MMS-TTS English checkpoint:
+
+```python
+import torch
+from transformers import VitsTokenizer, VitsModel, set_seed
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
+model = VitsModel.from_pretrained("facebook/mms-tts-eng")
+
+inputs = tokenizer(text="Hello - my dog is cute", return_tensors="pt")
+
+set_seed(555)  # make deterministic
+
+with torch.no_grad():
+   outputs = model(**inputs)
+
+waveform = outputs.waveform[0]
+```
+
+The resulting waveform can be saved as a `.wav` file:
+
+```python
+import scipy
+
+scipy.io.wavfile.write("techno.wav", rate=model.config.sampling_rate, data=waveform)
+```
+
+Or displayed in a Jupyter Notebook / Google Colab:
+
+```python
+from IPython.display import Audio
+
+Audio(waveform, rate=model.config.sampling_rate)
+```
+
+For certain languages with a non-Roman alphabet, such as Arabic, Mandarin or Hindi, the [`uroman`](https://github.com/isi-nlp/uroman) 
+perl package is required to pre-process the text inputs to the Roman alphabet.
+
+You can check whether you require the `uroman` package for your language by inspecting the `is_uroman` attribute of 
+the pre-trained `tokenizer`:
+
+```python
+from transformers import VitsTokenizer
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
+print(tokenizer.is_uroman)
+```
+
+If required, you should apply the uroman package to your text inputs **prior** to passing them to the `VitsTokenizer`, 
+since currently the tokenizer does not support performing the pre-processing itself.  
+
+To do this, first clone the uroman repository to your local machine and set the bash variable `UROMAN` to the local path:
+
+```bash
+git clone https://github.com/isi-nlp/uroman.git
+cd uroman
+export UROMAN=$(pwd)
+```
+
+You can then pre-process the text input using the following code snippet. You can either rely on using the bash variable 
+`UROMAN` to point to the uroman repository, or you can pass the uroman directory as an argument to the `uromaize` function:
+
+```python
+import torch
+from transformers import VitsTokenizer, VitsModel, set_seed
+import os
+import subprocess
+
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-kor")
+model = VitsModel.from_pretrained("facebook/mms-tts-kor")
+
+def uromanize(input_string, uroman_path):
+    """Convert non-Roman strings to Roman using the `uroman` perl package."""
+    script_path = os.path.join(uroman_path, "bin", "uroman.pl")
+
+    command = ["perl", script_path]
+
+    process = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    # Execute the perl command
+    stdout, stderr = process.communicate(input=input_string.encode())
+
+    if process.returncode != 0:
+        raise ValueError(f"Error {process.returncode}: {stderr.decode()}")
+
+    # Return the output as a string and skip the new-line character at the end
+    return stdout.decode()[:-1]
+
+text = "이봐 무슨 일이야"
+uromaized_text = uromanize(text, uroman_path=os.environ["UROMAN"])
+
+inputs = tokenizer(text=uromaized_text, return_tensors="pt")
+
+set_seed(555)  # make deterministic
+with torch.no_grad():
+   outputs = model(inputs["input_ids"])
+
+waveform = outputs.waveform[0]
+```
+
+## VitsConfig
+
+[[autodoc]] VitsConfig
+
+## VitsTokenizer
+
+[[autodoc]] VitsTokenizer
+    - __call__
+    - save_vocabulary
+
+## VitsModel
+
+[[autodoc]] VitsModel
+    - forward

docs/source/en/model_doc/whisper.md

@@ -48,6 +48,8 @@ The original code can be found [here](https://github.com/openai/whisper).
     - get_special_tokens_mask
     - create_token_type_ids_from_sequences
     - save_vocabulary
+    - batch_decode
+    - decode
 
 ## WhisperTokenizerFast
 
@@ -57,6 +59,8 @@ The original code can be found [here](https://github.com/openai/whisper).
     - get_special_tokens_mask
     - create_token_type_ids_from_sequences
     - save_vocabulary
+    - batch_decode
+    - decode
 
 ## WhisperFeatureExtractor
 
@@ -82,6 +86,7 @@ The original code can be found [here](https://github.com/openai/whisper).
 
 [[autodoc]] WhisperForConditionalGeneration
     - forward
+    - generate
 
 ## WhisperForAudioClassification
 

docs/source/en/model_memory_anatomy.md

@@ -76,7 +76,7 @@ GPU memory occupied: 0 MB.
 
 That looks good: the GPU memory is not occupied as we would expect before we load any models. If that's not the case on 
 your machine make sure to stop all processes that are using GPU memory. However, not all free GPU memory can be used by 
-the user. When a model is loaded to the GPU also the kernels are loaded which can take up 1-2GB of memory. To see how 
+the user. When a model is loaded to the GPU the kernels are also loaded, which can take up 1-2GB of memory. To see how 
 much it is we load a tiny tensor into the GPU which triggers the kernels to be loaded as well.
 
 ```py
@@ -105,7 +105,7 @@ how much space just the weights use.
 GPU memory occupied: 2631 MB.
 ```
 
-We can see that the model weights alone take up 1.3 GB of the GPU memory. The exact number depends on the specific 
+We can see that the model weights alone take up 1.3 GB of GPU memory. The exact number depends on the specific 
 GPU you are using. Note that on newer GPUs a model can sometimes take up more space since the weights are loaded in an 
 optimized fashion that speeds up the usage of the model. Now we can also quickly check if we get the same result 
 as with `nvidia-smi` CLI:
@@ -184,7 +184,7 @@ GPU memory occupied: 14949 MB.
 We see that already a relatively small batch size almost fills up our GPU's entire memory. However, a larger batch size 
 can often result in faster model convergence or better end performance. So ideally we want to tune the batch size to our
 model's needs and not to the GPU limitations. What's interesting is that we use much more memory than the size of the model. 
-To understand a bit better why this is the case let's have look at a model's operations and memory needs.
+To understand a bit better why this is the case let's have a look at a model's operations and memory needs.
 
 ## Anatomy of Model's Operations
 

docs/source/en/model_sharing.md

@@ -95,7 +95,7 @@ Specify `from_pt=True` to convert a checkpoint from PyTorch to TensorFlow:
 >>> tf_model = TFDistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_pt=True)
 ```
 
-Then you can save your new TensorFlow model with it's new checkpoint:
+Then you can save your new TensorFlow model with its new checkpoint:
 
 ```py
 >>> tf_model.save_pretrained("path/to/awesome-name-you-picked")
@@ -201,7 +201,7 @@ Or perhaps you'd like to add the TensorFlow version of your fine-tuned PyTorch m
 >>> tf_model.push_to_hub("my-awesome-model")
 ```
 
-Now when you navigate to the your Hugging Face profile, you should see your newly created model repository. Clicking on the **Files** tab will display all the files you've uploaded to the repository.
+Now when you navigate to your Hugging Face profile, you should see your newly created model repository. Clicking on the **Files** tab will display all the files you've uploaded to the repository.
 
 For more details on how to create and upload files to a repository, refer to the Hub documentation [here](https://huggingface.co/docs/hub/how-to-upstream).
 

docs/source/en/perf_infer_gpu_many.md

@@ -22,6 +22,10 @@ Note: A multi GPU setup can use the majority of the strategies described in the
 
 </Tip>
 
+## Flash Attention 2
+
+Flash Attention 2 integration also works in a multi-GPU setup, check out the appropriate section in the [single GPU section](./perf_infer_gpu_one#Flash-Attention-2)
+
 ## BetterTransformer
 
 [BetterTransformer](https://huggingface.co/docs/optimum/bettertransformer/overview) converts 🤗 Transformers models to use the PyTorch-native fastpath execution, which calls optimized kernels like Flash Attention under the hood.