stash commit

.circleci/create_circleci_config.py

@@ -151,13 +151,10 @@ 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 --junitxml=test-results/junit.xml -n {self.pytest_num_workers} " + " ".join(pytest_flags)
+        test_command += f"python -m pytest -n {self.pytest_num_workers} " + " ".join(pytest_flags)
 
         if self.parallelism == 1:
             if self.tests_to_run is None:
@@ -229,26 +226,15 @@ class CircleCIJob:
             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'if [ -s reports/{self.job_name}/failures_short.txt ]; '
+        check_test_command += 'then echo "Some test 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()'
+        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 ", "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; '
 
+        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
@@ -259,8 +245,6 @@ class CircleCIJob:
 
         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"}})
         job["steps"] = steps

.github/ISSUE_TEMPLATE/bug-report.yml

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

.github/PULL_REQUEST_TEMPLATE.md

@@ -51,7 +51,7 @@ Library:
 - pipelines: @Narsil
 - tensorflow: @gante and @Rocketknight1
 - tokenizers: @ArthurZucker
-- trainer: @muellerzr and @pacman100
+- trainer: @muellerz 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@v3
+        uses: docker/setup-buildx-action@v2
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v3
         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@v5
+        uses: docker/build-push-action@v3
         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@v3
+        uses: docker/setup-buildx-action@v2
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v3
         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@v3
+        uses: docker/setup-buildx-action@v2
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
         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@v5
+        uses: docker/build-push-action@v3
         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@v3
+        uses: docker/setup-buildx-action@v2
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v3
         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@v3
+        uses: docker/setup-buildx-action@v2
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v3
         with:
           context: ./docker/transformers-pytorch-gpu
           build-args: |
@@ -208,41 +208,6 @@ 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
@@ -251,19 +216,19 @@ jobs:
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v2
       -
         name: Check out code
         uses: actions/checkout@v3
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v3
         with:
           context: ./docker/transformers-tensorflow-gpu
           build-args: |

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

@@ -1,425 +0,0 @@
-name: Self-hosted runner AMD GPU (push)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (push-caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - ci_*
-      - ci-*
-    paths:
-      - "src/**"
-      - "tests/**"
-      - ".github/**"
-      - "templates/**"
-      - "utils/**"
-  repository_dispatch:
-
-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 ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', mi210]
-    container:
-      # --device /dev/dri/renderD128 == AMDGPU:0 (indexing for AMDGPU starts at 128 ...)
-      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
-      options: --device /dev/kfd --device /dev/dri/renderD128 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-
-  setup_gpu:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', mi210]
-    container:
-      # --device /dev/dri/renderD128 == AMDGPU:0 (indexing for AMDGPU starts at 128 ...)
-      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
-      options: --device /dev/kfd --device /dev/dri/renderD128 --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_single_gpu:
-    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: ${{ fromJson(needs.setup_gpu.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', mi210]
-    container:
-      # --device /dev/dri/renderD128 == AMDGPU:0 (indexing for AMDGPU starts at 128 ...)
-      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
-      options: --device /dev/kfd --device /dev/dri/renderD128 --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" >> $GITHUB_ENV
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all non-slow selected tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -n 2 --dist=loadfile -v --make-reports=${{ 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 }}
-
-  run_tests_multi_gpu:
-    name: Model tests
-    needs: setup
-    # `dummy` means there is no test to run
-    if: contains(fromJson(needs.setup.outputs.matrix), 'dummy') != true
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [ multi-gpu ]
-    runs-on: [ self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}' ]
-    container:
-      image: transformers-pytorch-amd-gpu-push-ci
-      options: --device /dev/kfd "$(./get_gpu.sh)" --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ 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 "${{ fromJson(needs.setup.outputs.test_map)[matrix.folders] }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all non-slow selected tests on GPU
-        env:
-          MKL_SERVICE_FORCE_INTEL: 1
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -n 2 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} ${{ fromJson(needs.setup.outputs.test_map)[matrix.folders] }}
-
-      - 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_single_gpu,
-#        run_tests_multi_gpu,
-#        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_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 }}"

.gitignore

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

README.md

@@ -310,7 +310,6 @@ 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.
@@ -433,7 +432,6 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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.
@@ -491,7 +489,6 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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.

README_es.md

@@ -287,7 +287,6 @@ 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.
@@ -410,7 +409,6 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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 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.
@@ -468,7 +466,6 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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) 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.

README_hd.md

@@ -259,7 +259,6 @@ 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।
@@ -382,7 +381,6 @@ conda install -c huggingface transformers
 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/main/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।
@@ -440,7 +438,6 @@ conda install -c huggingface transformers
 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)** (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/main/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.

README_ja.md

@@ -321,7 +321,6 @@ 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)
@@ -444,7 +443,6 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 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/main/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)
@@ -502,7 +500,6 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 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/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/main/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.

README_ko.md

@@ -236,7 +236,6 @@ 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.
@@ -359,7 +358,6 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 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/main/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.
@@ -417,7 +415,6 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 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/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/main/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.

README_zh-hans.md

@@ -260,7 +260,6 @@ 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。
@@ -383,7 +382,6 @@ conda install -c huggingface transformers
 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/main/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。
@@ -441,7 +439,6 @@ conda install -c huggingface transformers
 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/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/main/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.

README_zh-hant.md

@@ -272,7 +272,6 @@ 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.
@@ -395,7 +394,6 @@ conda install -c huggingface transformers
 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/main/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.
@@ -453,7 +451,6 @@ conda install -c huggingface transformers
 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) 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.

docker/transformers-pytorch-amd-gpu/Dockerfile

@@ -1,31 +0,0 @@
-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

docs/README.md

@@ -364,6 +364,9 @@ 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/en/_toctree.yml

@@ -88,11 +88,6 @@
     - local: generation_strategies
       title: Customize the generation strategy
     title: Generation
-  - isExpanded: false
-    sections:
-    - local: tasks/idefics
-      title: Image tasks with IDEFICS
-    title: Prompting
   title: Task Guides
 - sections:
   - local: fast_tokenizers
@@ -103,8 +98,6 @@
     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
@@ -414,8 +407,6 @@
         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
@@ -573,8 +564,6 @@
         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
@@ -645,8 +634,6 @@
         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

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 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)
+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)
 
 
 
@@ -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 into 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 in the repository
+`{your_username}/test-dynamic-pipeline`. After that anyone can use it as long as they provide the option
 `trust_remote_code=True`:
 
 ```py
@@ -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 to the list of tasks defined in `pipelines/__init__.py`.
+with the code of your pipeline, then add it in the list of tasks defined in `pipelines/__init__.py`.
 
-Then you will need to add tests. Create a new file `tests/test_pipelines_MY_PIPELINE.py` with examples of the other tests.
+Then you will need to add tests. Create a new file `tests/test_pipelines_MY_PIPELINE.py` with example with the other tests.
 
 The `run_pipeline_test` function will be very generic and run on small random models on every possible
 architecture as defined by `model_mapping` and `tf_model_mapping`.

docs/source/en/add_tensorflow_model.md

@@ -229,6 +229,7 @@ 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`

docs/source/en/chat_templating.md

@@ -1,255 +0,0 @@
-<!--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
-we remove the first newline after each block as well as any preceding whitespace before a block by default, using the 
-Jinja `trim_blocks` and `lstrip_blocks` flags. This means that you can write your templates with indentations and 
-newlines and still have them function correctly!
-
-```
-{% 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:
-
-```
-tokenizer.chat_template = "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}"
-```
-
-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.
-
-## 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!

docs/source/en/generation_strategies.md

@@ -55,10 +55,12 @@ 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
+>>> model.generation_config  # doctest: +IGNORE_RESULT
 GenerationConfig {
+    "_from_model_config": true,
     "bos_token_id": 50256,
     "eos_token_id": 50256,
+    "transformers_version": "4.26.0.dev0"
 }
 ```
 

docs/source/en/hpo_train.md

@@ -54,18 +54,6 @@ 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

@@ -76,7 +76,6 @@ The documentation is organized into five sections:
 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. **[BROS](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](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.
@@ -199,7 +198,6 @@ The documentation is organized into five sections:
 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. **[Persimmon](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](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.
@@ -257,7 +255,6 @@ The documentation is organized into five sections:
 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. **[ViTMatte](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](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](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](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.
@@ -311,7 +308,6 @@ Flax), PyTorch, and/or TensorFlow.
 |            BLIP-2             |       ✅        |         ❌         |      ❌      |
 |             BLOOM             |       ✅        |         ❌         |      ✅      |
 |          BridgeTower          |       ✅        |         ❌         |      ❌      |
-|             BROS              |       ✅        |         ❌         |      ❌      |
 |           CamemBERT           |       ✅        |         ✅         |      ❌      |
 |            CANINE             |       ✅        |         ❌         |      ❌      |
 |         Chinese-CLIP          |       ✅        |         ❌         |      ❌      |
@@ -422,7 +418,6 @@ Flax), PyTorch, and/or TensorFlow.
 |            Pegasus            |       ✅        |         ✅         |      ✅      |
 |           PEGASUS-X           |       ✅        |         ❌         |      ❌      |
 |           Perceiver           |       ✅        |         ❌         |      ❌      |
-|           Persimmon           |       ✅        |         ❌         |      ❌      |
 |          Pix2Struct           |       ✅        |         ❌         |      ❌      |
 |            PLBart             |       ✅        |         ❌         |      ❌      |
 |          PoolFormer           |       ✅        |         ❌         |      ❌      |
@@ -481,7 +476,6 @@ Flax), PyTorch, and/or TensorFlow.
 |          ViT Hybrid           |       ✅        |         ❌         |      ❌      |
 |            VitDet             |       ✅        |         ❌         |      ❌      |
 |            ViTMAE             |       ✅        |         ✅         |      ❌      |
-|           ViTMatte            |       ✅        |         ❌         |      ❌      |
 |            ViTMSN             |       ✅        |         ❌         |      ❌      |
 |             VITS              |       ✅        |         ❌         |      ❌      |
 |             ViViT             |       ✅        |         ❌         |      ❌      |

docs/source/en/main_classes/output.md

@@ -40,13 +40,6 @@ 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/tokenizer.md

@@ -55,7 +55,6 @@ to a given token).
 
 [[autodoc]] PreTrainedTokenizer
     - __call__
-    - apply_chat_template
     - batch_decode
     - decode
     - encode
@@ -69,7 +68,6 @@ 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/bros.md

@@ -1,115 +0,0 @@
-<!--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/llama2.md

@@ -78,10 +78,6 @@ 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/persimmon.md

@@ -1,96 +0,0 @@
-<!--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/vitmatte.md

@@ -1,44 +0,0 @@
-<!--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).
-
-
-## VitMatteConfig
-
-[[autodoc]] VitMatteConfig
-
-## VitMatteImageProcessor
-
-[[autodoc]] VitMatteImageProcessor
-    - preprocess
-
-## VitMatteForImageMatting
-
-[[autodoc]] VitMatteForImageMatting
-    - forward

docs/source/en/tasks/idefics.md

@@ -1,426 +0,0 @@
-<!--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.
-
--->
-
-# Image tasks with IDEFICS
-
-[[open-in-colab]]
-
-While individual tasks can be tackled by fine-tuning specialized models, an alternative approach 
-that has recently emerged and gained popularity is to use large models for a diverse set of tasks without fine-tuning. 
-For instance, large language models can handle such NLP tasks as summarization, translation, classification, and more. 
-This approach is no longer limited to a single modality, such as text, and in this guide, we will illustrate how you can 
-solve image-text tasks with a large multimodal model called IDEFICS. 
-
-[IDEFICS](../model_doc/idefics) is an open-access vision and language model based on [Flamingo](https://huggingface.co/papers/2204.14198), 
-a state-of-the-art visual language model initially developed by DeepMind. The model accepts arbitrary sequences of image 
-and text inputs and generates coherent text as output. It can answer questions about images, describe visual content, 
-create stories grounded in multiple images, and so on. IDEFICS comes in two variants - [80 billion parameters](https://huggingface.co/HuggingFaceM4/idefics-80b) 
-and [9 billion parameters](https://huggingface.co/HuggingFaceM4/idefics-9b), both of which are available on the 🤗 Hub. For each variant, you can also find fine-tuned instructed 
-versions of the model adapted for conversational use cases.
-
-This model is exceptionally versatile and can be used for a wide range of image and multimodal tasks. However, 
-being a large model means it requires significant computational resources and infrastructure. It is up to you to decide whether 
-this approach suits your use case better than fine-tuning specialized models for each individual task. 
-
-In this guide, you'll learn how to: 
-- [Load IDEFICS](#loading-the-model) and [load the quantized version of the model](#loading-the-quantized-version-of-the-model)
-- Use IDEFICS for: 
-  - [Image captioning](#image-captioning)
-  - [Prompted image captioning](#prompted-image-captioning)
-  - [Few-shot prompting](#few-shot-prompting)
-  - [Visual question answering](#visual-question-answering)
-  - [Image classificaiton](#image-classification)
-  - [Image-guided text generation](#image-guided-text-generation)
-- [Run inference in batch mode](#running-inference-in-batch-mode)
-- [Run IDEFICS instruct for conversational use](#idefics-instruct-for-conversational-use)
-
-Before you begin, make sure you have all the necessary libraries installed. 
-
-```bash
-pip install -q bitsandbytes sentencepiece accelerate transformers
-```
-
-<Tip>
-To run the following examples with a non-quantized version of the model checkpoint you will need at least 20GB of GPU memory.
-</Tip>
-
-## Loading the model
-
-Let's start by loading the model's 9 billion parameters checkpoint: 
-
-```py
->>> checkpoint = "HuggingFaceM4/idefics-9b"
-```
-
-Just like for other Transformers models, you need to load a processor and the model itself from the checkpoint. 
-The IDEFICS processor wraps a [`LlamaTokenizer`] and IDEFICS image processor into a single processor to take care of 
-preparing text and image inputs for the model.
-
-```py
->>> import torch
-
->>> from transformers import IdeficsForVisionText2Text, AutoProcessor
-
->>> processor = AutoProcessor.from_pretrained(checkpoint)
-
->>> model = IdeficsForVisionText2Text.from_pretrained(checkpoint, torch_dtype=torch.bfloat16, device_map="auto")
-```
-
-Setting `device_map` to `"auto"` will automatically determine how to load and store the model weights in the most optimized 
-manner given existing devices.
-
-### Quantized model
-
-If high-memory GPU availability is an issue, you can load the quantized version of the model. To load the model and the 
-processor in 4bit precision, pass a `BitsAndBytesConfig` to the `from_pretrained` method and the model will be compressed 
-on the fly while loading.
-
-```py
->>> import torch
->>> from transformers import IdeficsForVisionText2Text, AutoProcessor, BitsAndBytesConfig
-
->>> quantization_config = BitsAndBytesConfig(
-...     load_in_4bit=True,
-...     bnb_4bit_compute_dtype=torch.float16,
-... )
-
->>> processor = AutoProcessor.from_pretrained(checkpoint)
-
->>> model = IdeficsForVisionText2Text.from_pretrained(
-...     checkpoint,
-...     quantization_config=quantization_config,
-...     device_map="auto"
-... )
-```
-
-Now that you have the model loaded in one of the suggested ways, let's move on to exploring tasks that you can use IDEFICS for.
-
-## Image captioning
-
-Image captioning is the task of predicting a caption for a given image. A common application is to aid visually impaired 
-people navigate through different situations, for instance, explore image content online. 
-
-To illustrate the task, get an image to be captioned, e.g.:
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-im-captioning.jpg" alt="Image of a puppy in a flower bed"/>
-</div>
-
-Photo by [Hendo Wang](https://unsplash.com/@hendoo). 
-
-IDEFICS accepts text and image prompts. However, to caption an image, you do not have to provide a text prompt to the 
-model, only the preprocessed input image. Without a text prompt, the model will start generating text from the 
-BOS (beginning-of-sequence) token thus creating a caption.
-
-As image input to the model, you can use either an image object (`PIL.Image`) or a url from which the image can be retrieved.
-
-```py
->>> prompt = [
-...     "https://images.unsplash.com/photo-1583160247711-2191776b4b91?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3542&q=80",
-... ]
-
->>> inputs = processor(prompt, return_tensors="pt").to("cuda")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, max_new_tokens=10, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> print(generated_text[0])
-A puppy in a flower bed
-```
-
-<Tip>
-
-It is a good idea to include the `bad_words_ids` in the call to `generate` to avoid errors arising when increasing 
-the `max_new_tokens`: the model will want to generate a new `<image>` or `<fake_token_around_image>` token when there 
-is no image being generated by the model.
-You can set it on-the-fly as in this guide, or store in the `GenerationConfig` as described in the [Text generation strategies](../generation_strategies) guide.
-</Tip>
-
-## Prompted image captioning
-
-You can extend image captioning by providing a text prompt, which the model will continue given the image. Let's take 
-another image to illustrate:
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-prompted-im-captioning.jpg" alt="Image of the Eiffel Tower at night"/>
-</div>
-
-Photo by [Denys Nevozhai](https://unsplash.com/@dnevozhai).
-   
-Textual and image prompts can be passed to the model's processor as a single list to create appropriate inputs.
-
-```py
->>> prompt = [
-...     "https://images.unsplash.com/photo-1543349689-9a4d426bee8e?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3501&q=80",
-...     "This is an image of ",
-... ]
-
->>> inputs = processor(prompt, return_tensors="pt").to("cuda")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, max_new_tokens=10, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> print(generated_text[0])
-This is an image of the Eiffel Tower in Paris, France.
-```
-
-## Few-shot prompting
-
-While IDEFICS demonstrates great zero-shot results, your task may require a certain format of the caption, or come with 
-other restrictions or requirements that increase task's complexity. Few-shot prompting can be used to enable in-context learning.
-By providing examples in the prompt, you can steer the model to generate results that mimic the format of given examples. 
-
-Let's use the previous image of the Eiffel Tower as an example for the model and build a prompt that demonstrates to the model 
-that in addition to learning what the object in an image is, we would also like to get some interesting information about it. 
-Then, let's see, if we can get the same response format for an image of the Statue of Liberty:
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-few-shot.jpg" alt="Image of the Statue of Liberty"/>
-</div>
-
-Photo by [Juan Mayobre](https://unsplash.com/@jmayobres).
-  
-```py
->>> prompt = ["User:",
-...            "https://images.unsplash.com/photo-1543349689-9a4d426bee8e?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3501&q=80",
-...            "Describe this image.\nAssistant: An image of the Eiffel Tower at night. Fun fact: the Eiffel Tower is the same height as an 81-storey building.\n",
-...            "User:",
-...            "https://images.unsplash.com/photo-1524099163253-32b7f0256868?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3387&q=80",
-...            "Describe this image.\nAssistant:"
-...            ]
-
->>> inputs = processor(prompt, return_tensors="pt").to("cuda")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, max_new_tokens=30, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> print(generated_text[0])
-User: Describe this image.
-Assistant: An image of the Eiffel Tower at night. Fun fact: the Eiffel Tower is the same height as an 81-storey building. 
-User: Describe this image.
-Assistant: An image of the Statue of Liberty. Fun fact: the Statue of Liberty is 151 feet tall.
-```
-
-Notice that just from a single example (i.e., 1-shot) the model has learned how to perform the task. For more complex tasks, 
-feel free to experiment with a larger number of examples (e.g., 3-shot, 5-shot, etc.).
-
-## Visual question answering
-
-Visual Question Answering (VQA) is the task of answering open-ended questions based on an image. Similar to image 
-captioning it can be used in accessibility applications, but also in education (reasoning about visual materials), customer 
-service (questions about products based on images), and image retrieval.
-
-Let's get a new image for this task: 
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-vqa.jpg" alt="Image of a couple having a picnic"/>
-</div>
-
-Photo by [Jarritos Mexican Soda](https://unsplash.com/@jarritos).
-
-You can steer the model from image captioning to visual question answering by prompting it with appropriate instructions: 
-
-```py
->>> prompt = [
-...     "Instruction: Provide an answer to the question. Use the image to answer.\n",
-...     "https://images.unsplash.com/photo-1623944889288-cd147dbb517c?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3540&q=80",
-...     "Question: Where are these people and what's the weather like? Answer:"
-... ]
-
->>> inputs = processor(prompt, return_tensors="pt").to("cuda")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, max_new_tokens=20, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> print(generated_text[0])
-Instruction: Provide an answer to the question. Use the image to answer.
- Question: Where are these people and what's the weather like? Answer: They're in a park in New York City, and it's a beautiful day.
-```
-
-## Image classification
-
-IDEFICS is capable of classifying images into different categories without being explicitly trained on data containing 
-labeled examples from those specific categories. Given a list of categories and using its image and text understanding 
-capabilities, the model can infer which category the image likely belongs to. 
-
-Say, we have this image of a vegetable stand: 
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-classification.jpg" alt="Image of a vegetable stand"/>
-</div>
-
-Photo by [Peter Wendt](https://unsplash.com/@peterwendt).
-
-We can instruct the model to classify the image into one of the categories that we have:
-
-```py
->>> categories = ['animals','vegetables', 'city landscape', 'cars', 'office']
->>> prompt = [f"Instruction: Classify the following image into a single category from the following list: {categories}.\n",
-...     "https://images.unsplash.com/photo-1471193945509-9ad0617afabf?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3540&q=80",    
-...     "Category: "
-... ]
-
->>> inputs = processor(prompt, return_tensors="pt").to("cuda")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, max_new_tokens=4, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> print(generated_text[0])
-Instruction: Classify the following image into a single category from the following list: ['animals', 'vegetables', 'city landscape', 'cars', 'office'].
-Category: Vegetables
-```  
-
-In the example above we instruct the model to classify the image into a single category, however, you can also prompt the model to do rank classification.
-
-## Image-guided text generation
-
-For more creative applications, you can use image-guided text generation to generate text based on an image. This can be 
-useful to create descriptions of products, ads, descriptions of a scene, etc. 
-
-Let's prompt IDEFICS to write a story based on a simple image of a red door: 
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-story-generation.jpg" alt="Image of a red door with a pumpkin on the steps"/>
-</div>
-
-Photo by [Craig Tidball](https://unsplash.com/@devonshiremedia).
-  
-```py
->>> prompt = ["Instruction: Use the image to write a story. \n",
-...     "https://images.unsplash.com/photo-1517086822157-2b0358e7684a?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=2203&q=80",
-...     "Story: \n"]
-
->>> inputs = processor(prompt, return_tensors="pt").to("cuda")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, num_beams=2, max_new_tokens=200, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> print(generated_text[0]) 
-Instruction: Use the image to write a story. 
- Story: 
-Once upon a time, there was a little girl who lived in a house with a red door.  She loved her red door.  It was the prettiest door in the whole world.
-
-One day, the little girl was playing in her yard when she noticed a man standing on her doorstep.  He was wearing a long black coat and a top hat.
-
-The little girl ran inside and told her mother about the man.
-
-Her mother said, “Don’t worry, honey.  He’s just a friendly ghost.”
-
-The little girl wasn’t sure if she believed her mother, but she went outside anyway.
-
-When she got to the door, the man was gone.
-
-The next day, the little girl was playing in her yard again when she noticed the man standing on her doorstep.
-
-He was wearing a long black coat and a top hat.
-
-The little girl ran
-```
-
-Looks like IDEFICS noticed the pumpkin on the doorstep and went with a spooky Halloween story about a ghost.
-
-<Tip>
-
-For longer outputs like this, you will greatly benefit from tweaking the text generation strategy. This can help 
-you significantly improve the quality of the generated output. Check out [Text generation strategies](../generation_strategies) 
-to learn more. 
-</Tip>
-
-## Running inference in batch mode
-
-All of the earlier sections illustrated IDEFICS for a single example. In a very similar fashion, you can run inference 
-for a batch of examples by passing a list of prompts:
-
-```py
->>> prompts = [
-...     [   "https://images.unsplash.com/photo-1543349689-9a4d426bee8e?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3501&q=80",
-...         "This is an image of ",
-...     ],
-...     [   "https://images.unsplash.com/photo-1623944889288-cd147dbb517c?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3540&q=80",
-...         "This is an image of ",
-...     ],
-...     [   "https://images.unsplash.com/photo-1471193945509-9ad0617afabf?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=3540&q=80",
-...         "This is an image of ",
-...     ],
-... ]
-
->>> inputs = processor(prompts, return_tensors="pt")
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, max_new_tokens=10, bad_words_ids=bad_words_ids)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> for i,t in enumerate(generated_text):
-...     print(f"{i}:\n{t}\n") 
-0:
-This is an image of the Eiffel Tower in Paris, France.
-
-1:
-This is an image of a couple on a picnic blanket.
-
-2:
-This is an image of a vegetable stand.
-```
-
-## IDEFICS instruct for conversational use
-
-For conversational use cases, you can find fine-tuned instructed versions of the model on the 🤗 Hub: 
-`HuggingFaceM4/idefics-80b-instruct` and `HuggingFaceM4/idefics-9b-instruct`.
-
-These checkpoints are the result of fine-tuning the respective base models on a mixture of supervised and instruction 
-fine-tuning datasets, which boosts the downstream performance while making the models more usable in conversational settings.
-
-The use and prompting for the conversational use is very similar to using the base models: 
-
-```py
->>> import torch
->>> from transformers import IdeficsForVisionText2Text, AutoProcessor
-
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
-
->>> checkpoint = "HuggingFaceM4/idefics-9b-instruct"
->>> model = IdeficsForVisionText2Text.from_pretrained(checkpoint, torch_dtype=torch.bfloat16).to(device)
->>> processor = AutoProcessor.from_pretrained(checkpoint)
-
->>> prompts = [
-...     [
-...         "User: What is in this image?",
-...         "https://upload.wikimedia.org/wikipedia/commons/8/86/Id%C3%A9fix.JPG",
-...         "<end_of_utterance>",
-
-...         "\nAssistant: This picture depicts Idefix, the dog of Obelix in Asterix and Obelix. Idefix is running on the ground.<end_of_utterance>",
-
-...         "\nUser:",
-...         "https://static.wikia.nocookie.net/asterix/images/2/25/R22b.gif/revision/latest?cb=20110815073052",
-...         "And who is that?<end_of_utterance>",
-
-...         "\nAssistant:",
-...     ],
-... ]
-
->>> # --batched mode
->>> inputs = processor(prompts, add_end_of_utterance_token=False, return_tensors="pt").to(device)
->>> # --single sample mode
->>> # inputs = processor(prompts[0], return_tensors="pt").to(device)
-
->>> # Generation args
->>> exit_condition = processor.tokenizer("<end_of_utterance>", add_special_tokens=False).input_ids
->>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
-
->>> generated_ids = model.generate(**inputs, eos_token_id=exit_condition, bad_words_ids=bad_words_ids, max_length=100)
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
->>> for i, t in enumerate(generated_text):
-...     print(f"{i}:\n{t}\n")
-```

docs/source/en/tasks/language_modeling.md

@@ -37,7 +37,7 @@ You can finetune other architectures for causal language modeling following the
 Choose one of the following architectures:
 
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
-[BART](../model_doc/bart), [BERT](../model_doc/bert), [Bert Generation](../model_doc/bert-generation), [BigBird](../model_doc/big_bird), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [BioGpt](../model_doc/biogpt), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CodeLlama](../model_doc/code_llama), [CodeGen](../model_doc/codegen), [CPM-Ant](../model_doc/cpmant), [CTRL](../model_doc/ctrl), [Data2VecText](../model_doc/data2vec-text), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [Falcon](../model_doc/falcon), [GIT](../model_doc/git), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [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), [LLaMA](../model_doc/llama), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [MPT](../model_doc/mpt), [MusicGen](../model_doc/musicgen), [MVP](../model_doc/mvp), [OpenLlama](../model_doc/open-llama), [OpenAI GPT](../model_doc/openai-gpt), [OPT](../model_doc/opt), [Pegasus](../model_doc/pegasus), [Persimmon](../model_doc/persimmon), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [QDQBert](../model_doc/qdqbert), [Reformer](../model_doc/reformer), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [RWKV](../model_doc/rwkv), [Speech2Text2](../model_doc/speech_to_text_2), [Transformer-XL](../model_doc/transfo-xl), [TrOCR](../model_doc/trocr), [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), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod)
+[BART](../model_doc/bart), [BERT](../model_doc/bert), [Bert Generation](../model_doc/bert-generation), [BigBird](../model_doc/big_bird), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [BioGpt](../model_doc/biogpt), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CodeLlama](../model_doc/code_llama), [CodeGen](../model_doc/codegen), [CPM-Ant](../model_doc/cpmant), [CTRL](../model_doc/ctrl), [Data2VecText](../model_doc/data2vec-text), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [Falcon](../model_doc/falcon), [GIT](../model_doc/git), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [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), [LLaMA](../model_doc/llama), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [MPT](../model_doc/mpt), [MusicGen](../model_doc/musicgen), [MVP](../model_doc/mvp), [OpenLlama](../model_doc/open-llama), [OpenAI GPT](../model_doc/openai-gpt), [OPT](../model_doc/opt), [Pegasus](../model_doc/pegasus), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [QDQBert](../model_doc/qdqbert), [Reformer](../model_doc/reformer), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [RWKV](../model_doc/rwkv), [Speech2Text2](../model_doc/speech_to_text_2), [Transformer-XL](../model_doc/transfo-xl), [TrOCR](../model_doc/trocr), [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), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod)
 
 
 

docs/source/en/tasks/object_detection.md

@@ -136,8 +136,8 @@ To get an even better understanding of the data, visualize an example in the dat
 >>> label2id = {v: k for k, v in id2label.items()}
 
 >>> for i in range(len(annotations["id"])):
-...     box = annotations["bbox"][i]
-...     class_idx = annotations["category"][i]
+...     box = annotations["bbox"][i - 1]
+...     class_idx = annotations["category"][i - 1]
 ...     x, y, w, h = tuple(box)
 ...     draw.rectangle((x, y, x + w, y + h), outline="red", width=1)
 ...     draw.text((x, y), id2label[class_idx], fill="white")

docs/source/en/tasks/sequence_classification.md

@@ -33,7 +33,7 @@ The task illustrated in this tutorial is supported by the following model archit
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
 
 
-[ALBERT](../model_doc/albert), [BART](../model_doc/bart), [BERT](../model_doc/bert), [BigBird](../model_doc/big_bird), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [BioGpt](../model_doc/biogpt), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CANINE](../model_doc/canine), [CodeLlama](../model_doc/code_llama), [ConvBERT](../model_doc/convbert), [CTRL](../model_doc/ctrl), [Data2VecText](../model_doc/data2vec-text), [DeBERTa](../model_doc/deberta), [DeBERTa-v2](../model_doc/deberta-v2), [DistilBERT](../model_doc/distilbert), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [ErnieM](../model_doc/ernie_m), [ESM](../model_doc/esm), [Falcon](../model_doc/falcon), [FlauBERT](../model_doc/flaubert), [FNet](../model_doc/fnet), [Funnel Transformer](../model_doc/funnel), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [GPT Neo](../model_doc/gpt_neo), [GPT NeoX](../model_doc/gpt_neox), [GPT-J](../model_doc/gptj), [I-BERT](../model_doc/ibert), [LayoutLM](../model_doc/layoutlm), [LayoutLMv2](../model_doc/layoutlmv2), [LayoutLMv3](../model_doc/layoutlmv3), [LED](../model_doc/led), [LiLT](../model_doc/lilt), [LLaMA](../model_doc/llama), [Longformer](../model_doc/longformer), [LUKE](../model_doc/luke), [MarkupLM](../model_doc/markuplm), [mBART](../model_doc/mbart), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [MobileBERT](../model_doc/mobilebert), [MPNet](../model_doc/mpnet), [MPT](../model_doc/mpt), [MRA](../model_doc/mra), [MT5](../model_doc/mt5), [MVP](../model_doc/mvp), [Nezha](../model_doc/nezha), [Nyströmformer](../model_doc/nystromformer), [OpenLlama](../model_doc/open-llama), [OpenAI GPT](../model_doc/openai-gpt), [OPT](../model_doc/opt), [Perceiver](../model_doc/perceiver), [Persimmon](../model_doc/persimmon), [PLBart](../model_doc/plbart), [QDQBert](../model_doc/qdqbert), [Reformer](../model_doc/reformer), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [SqueezeBERT](../model_doc/squeezebert), [T5](../model_doc/t5), [TAPAS](../model_doc/tapas), [Transformer-XL](../model_doc/transfo-xl), [UMT5](../model_doc/umt5), [XLM](../model_doc/xlm), [XLM-RoBERTa](../model_doc/xlm-roberta), [XLM-RoBERTa-XL](../model_doc/xlm-roberta-xl), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod), [YOSO](../model_doc/yoso)
+[ALBERT](../model_doc/albert), [BART](../model_doc/bart), [BERT](../model_doc/bert), [BigBird](../model_doc/big_bird), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [BioGpt](../model_doc/biogpt), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CANINE](../model_doc/canine), [CodeLlama](../model_doc/code_llama), [ConvBERT](../model_doc/convbert), [CTRL](../model_doc/ctrl), [Data2VecText](../model_doc/data2vec-text), [DeBERTa](../model_doc/deberta), [DeBERTa-v2](../model_doc/deberta-v2), [DistilBERT](../model_doc/distilbert), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [ErnieM](../model_doc/ernie_m), [ESM](../model_doc/esm), [Falcon](../model_doc/falcon), [FlauBERT](../model_doc/flaubert), [FNet](../model_doc/fnet), [Funnel Transformer](../model_doc/funnel), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [GPT Neo](../model_doc/gpt_neo), [GPT NeoX](../model_doc/gpt_neox), [GPT-J](../model_doc/gptj), [I-BERT](../model_doc/ibert), [LayoutLM](../model_doc/layoutlm), [LayoutLMv2](../model_doc/layoutlmv2), [LayoutLMv3](../model_doc/layoutlmv3), [LED](../model_doc/led), [LiLT](../model_doc/lilt), [LLaMA](../model_doc/llama), [Longformer](../model_doc/longformer), [LUKE](../model_doc/luke), [MarkupLM](../model_doc/markuplm), [mBART](../model_doc/mbart), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [MobileBERT](../model_doc/mobilebert), [MPNet](../model_doc/mpnet), [MPT](../model_doc/mpt), [MRA](../model_doc/mra), [MT5](../model_doc/mt5), [MVP](../model_doc/mvp), [Nezha](../model_doc/nezha), [Nyströmformer](../model_doc/nystromformer), [OpenLlama](../model_doc/open-llama), [OpenAI GPT](../model_doc/openai-gpt), [OPT](../model_doc/opt), [Perceiver](../model_doc/perceiver), [PLBart](../model_doc/plbart), [QDQBert](../model_doc/qdqbert), [Reformer](../model_doc/reformer), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [SqueezeBERT](../model_doc/squeezebert), [T5](../model_doc/t5), [TAPAS](../model_doc/tapas), [Transformer-XL](../model_doc/transfo-xl), [UMT5](../model_doc/umt5), [XLM](../model_doc/xlm), [XLM-RoBERTa](../model_doc/xlm-roberta), [XLM-RoBERTa-XL](../model_doc/xlm-roberta-xl), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod), [YOSO](../model_doc/yoso)
 
 
 

docs/source/en/tasks/text-to-speech.md

@@ -19,40 +19,16 @@ rendered properly in your Markdown viewer.
 [[open-in-colab]]
 
 Text-to-speech (TTS) is the task of creating natural-sounding speech from text, where the speech can be generated in multiple 
-languages and for multiple speakers. Several text-to-speech models are currently available in 🤗 Transformers, such as 
-[Bark](../model_doc/bark), [MMS](../model_doc/mms), [VITS](../model_doc/vits) and [SpeechT5](../model_doc/speecht5). 
-
-You can easily generate audio using the `"text-to-audio"` pipeline (or its alias - `"text-to-speech"`). Some models, like Bark, 
-can also be conditioned to generate non-verbal communications such as laughing, sighing and crying, or even add music.
-Here's an example of how you would use the `"text-to-speech"` pipeline with Bark: 
-
-```py
->>> from transformers import pipeline
-
->>> pipe = pipeline("text-to-speech", model="suno/bark-small")
->>> text = "[clears throat] This is a test ... and I just took a long pause."
->>> output = pipe(text)
-```
-
-Here's a code snippet you can use to listen to the resulting audio in a notebook: 
-
-```python
->>> from IPython.display import Audio
->>> Audio(output["audio"], rate=output["sampling_rate"])
-```
-
-For more examples on what Bark and other pretrained TTS models can do, refer to our 
-[Audio course](https://huggingface.co/learn/audio-course/chapter6/pre-trained_models). 
-
-If you are looking to fine-tune a TTS model, you can currently fine-tune SpeechT5 only. SpeechT5 is pre-trained on a combination of 
+languages and for multiple speakers. The only text-to-speech model currently available in 🤗 Transformers 
+is [SpeechT5](model_doc/speecht5), though more will be added in the future. SpeechT5 is pre-trained on a combination of 
 speech-to-text and text-to-speech data, allowing it to learn a unified space of hidden representations shared by both text 
 and speech. This means that the same pre-trained model can be fine-tuned for different tasks. Furthermore, SpeechT5 
 supports multiple speakers through x-vector speaker embeddings. 
 
-The remainder of this guide illustrates how to:
+This guide illustrates how to:
 
-1. Fine-tune [SpeechT5](../model_doc/speecht5) that was originally trained on English speech on the Dutch (`nl`) language subset of the [VoxPopuli](https://huggingface.co/datasets/facebook/voxpopuli) dataset.
-2. Use your refined model for inference in one of two ways: using a pipeline or directly.
+1. Fine-tune [SpeechT5](model_doc/speecht5) that was originally trained on English speech on the Dutch (`nl`) language subset of the [VoxPopuli](https://huggingface.co/datasets/facebook/voxpopuli) dataset.
+2. Use your fine-tuned model for inference.
 
 Before you begin, make sure you have all the necessary libraries installed:
 
@@ -509,12 +485,6 @@ the `per_device_train_batch_size` incrementally by factors of 2 and increase `gr
 >>> trainer.train()
 ```
 
-To be able to use your checkpoint with a pipeline, make sure to save the processor with the checkpoint: 
-
-```py
->>> processor.save_pretrained("YOUR_ACCOUNT_NAME/speecht5_finetuned_voxpopuli_nl")
-```
-
 Push the final model to the 🤗 Hub:
 
 ```py
@@ -523,70 +493,29 @@ Push the final model to the 🤗 Hub:
 
 ## Inference
 
-### Inference with a pipeline
-
 Great, now that you've fine-tuned a model, you can use it for inference!
-First, let's see how you can use it with a corresponding pipeline. Let's create a `"text-to-speech"` pipeline with your 
-checkpoint: 
-
-```py
->>> from transformers import pipeline
-
->>> pipe = pipeline("text-to-speech", model="YOUR_ACCOUNT_NAME/speecht5_finetuned_voxpopuli_nl")
-```
-
-Pick a piece of text in Dutch you'd like narrated, e.g.:
-
-```py
->>> text = "hallo allemaal, ik praat nederlands. groetjes aan iedereen!"
-```
-
-To use SpeechT5 with the pipeline, you'll need a speaker embedding. Let's get it from an example in the test dataset: 
-
-```py
->>> example = dataset["test"][304]
->>> speaker_embeddings = torch.tensor(example["speaker_embeddings"]).unsqueeze(0)
-```
-
-Now you can pass the text and speaker embeddings to the pipeline, and it will take care of the rest: 
-
-```py
->>> forward_params = {"speaker_embeddings": speaker_embeddings}
->>> output = pipe(text, forward_params=forward_params)
->>> output
-{'audio': array([-6.82714235e-05, -4.26525949e-04,  1.06134125e-04, ...,
-        -1.22392643e-03, -7.76011671e-04,  3.29112721e-04], dtype=float32),
- 'sampling_rate': 16000}
-```
-
-You can then listen to the result:
-
-```py
->>> from IPython.display import Audio
->>> Audio(output['audio'], rate=output['sampling_rate']) 
-```
-
-### Run inference manually
-
-You can achieve the same inference results without using the pipeline, however, more steps will be required. 
-
-Load the model from the 🤗 Hub: 
+Load the model from the 🤗 Hub (make sure to use your account name in the following code snippet): 
 
 ```py
 >>> model = SpeechT5ForTextToSpeech.from_pretrained("YOUR_ACCOUNT/speecht5_finetuned_voxpopuli_nl")
 ```
 
-Pick an example from the test dataset obtain a speaker embedding. 
+Pick an example, here we'll take one from the test dataset. Obtain a speaker embedding. 
 
 ```py 
 >>> example = dataset["test"][304]
 >>> speaker_embeddings = torch.tensor(example["speaker_embeddings"]).unsqueeze(0)
 ```
 
-Define the input text and tokenize it.
+Define some input text and tokenize it.
 
 ```py 
 >>> text = "hallo allemaal, ik praat nederlands. groetjes aan iedereen!"
+```
+
+Preprocess the input text: 
+
+```py
 >>> inputs = processor(text=text, return_tensors="pt")
 ```
 

docs/source/en/tasks/token_classification.md

@@ -32,7 +32,8 @@ The task illustrated in this tutorial is supported by the following model archit
 
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
 
-[ALBERT](../model_doc/albert), [BERT](../model_doc/bert), [BigBird](../model_doc/big_bird), [BioGpt](../model_doc/biogpt), [BLOOM](../model_doc/bloom), [BROS](../model_doc/bros), [CamemBERT](../model_doc/camembert), [CANINE](../model_doc/canine), [ConvBERT](../model_doc/convbert), [Data2VecText](../model_doc/data2vec-text), [DeBERTa](../model_doc/deberta), [DeBERTa-v2](../model_doc/deberta-v2), [DistilBERT](../model_doc/distilbert), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [ErnieM](../model_doc/ernie_m), [ESM](../model_doc/esm), [Falcon](../model_doc/falcon), [FlauBERT](../model_doc/flaubert), [FNet](../model_doc/fnet), [Funnel Transformer](../model_doc/funnel), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [GPT Neo](../model_doc/gpt_neo), [GPT NeoX](../model_doc/gpt_neox), [I-BERT](../model_doc/ibert), [LayoutLM](../model_doc/layoutlm), [LayoutLMv2](../model_doc/layoutlmv2), [LayoutLMv3](../model_doc/layoutlmv3), [LiLT](../model_doc/lilt), [Longformer](../model_doc/longformer), [LUKE](../model_doc/luke), [MarkupLM](../model_doc/markuplm), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [MobileBERT](../model_doc/mobilebert), [MPNet](../model_doc/mpnet), [MPT](../model_doc/mpt), [MRA](../model_doc/mra), [Nezha](../model_doc/nezha), [Nyströmformer](../model_doc/nystromformer), [QDQBert](../model_doc/qdqbert), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [SqueezeBERT](../model_doc/squeezebert), [XLM](../model_doc/xlm), [XLM-RoBERTa](../model_doc/xlm-roberta), [XLM-RoBERTa-XL](../model_doc/xlm-roberta-xl), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod), [YOSO](../model_doc/yoso)
+[ALBERT](../model_doc/albert), [BERT](../model_doc/bert), [BigBird](../model_doc/big_bird), [BioGpt](../model_doc/biogpt), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CANINE](../model_doc/canine), [ConvBERT](../model_doc/convbert), [Data2VecText](../model_doc/data2vec-text), [DeBERTa](../model_doc/deberta), [DeBERTa-v2](../model_doc/deberta-v2), [DistilBERT](../model_doc/distilbert), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [ErnieM](../model_doc/ernie_m), [ESM](../model_doc/esm), [Falcon](../model_doc/falcon), [FlauBERT](../model_doc/flaubert), [FNet](../model_doc/fnet), [Funnel Transformer](../model_doc/funnel), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [GPT Neo](../model_doc/gpt_neo), [GPT NeoX](../model_doc/gpt_neox), [I-BERT](../model_doc/ibert), [LayoutLM](../model_doc/layoutlm), [LayoutLMv2](../model_doc/layoutlmv2), [LayoutLMv3](../model_doc/layoutlmv3), [LiLT](../model_doc/lilt), [Longformer](../model_doc/longformer), [LUKE](../model_doc/luke), [MarkupLM](../model_doc/markuplm), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [MobileBERT](../model_doc/mobilebert), [MPNet](../model_doc/mpnet), [MPT](../model_doc/mpt), [MRA](../model_doc/mra), [Nezha](../model_doc/nezha), [Nyströmformer](../model_doc/nystromformer), [QDQBert](../model_doc/qdqbert), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [SqueezeBERT](../model_doc/squeezebert), [XLM](../model_doc/xlm), [XLM-RoBERTa](../model_doc/xlm-roberta), [XLM-RoBERTa-XL](../model_doc/xlm-roberta-xl), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod), [YOSO](../model_doc/yoso)
+
 
 <!--End of the generated tip-->
 

docs/source/en/tasks/zero_shot_image_classification.md

@@ -75,7 +75,7 @@ include a local path to an image or an image url.
 The candidate labels can be simple words like in this example, or more descriptive.
 
 ```py
->>> predictions = detector(image, candidate_labels=["fox", "bear", "seagull", "owl"])
+>>> predictions = classifier(image, candidate_labels=["fox", "bear", "seagull", "owl"])
 >>> predictions
 [{'score': 0.9996670484542847, 'label': 'owl'},
  {'score': 0.000199399160919711, 'label': 'seagull'},

docs/source/es/tasks/language_modeling.md

@@ -122,7 +122,7 @@ Así es como puedes crear una función de preprocesamiento para convertir la lis
 ...     return tokenizer([" ".join(x) for x in examples["answers.text"]], truncation=True)
 ```
 
-Usa de 🤗 Datasets la función [`map`](https://huggingface.co/docs/datasets/process#map) para aplicar la función de preprocesamiento sobre el dataset en su totalidad. Puedes acelerar la función `map` configurando el argumento `batched=True` para procesar múltiples elementos del dataset a la vez y aumentar la cantidad de procesos con `num_proc`. Elimina las columnas que no necesitas:
+Usa de 🤗 Datasets la función [`map`](https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map) para aplicar la función de preprocesamiento sobre el dataset en su totalidad. Puedes acelerar la función `map` configurando el argumento `batched=True` para procesar múltiples elementos del dataset a la vez y aumentar la cantidad de procesos con `num_proc`. Elimina las columnas que no necesitas:
 
 ```py
 >>> tokenized_eli5 = eli5.map(

docs/source/ko/_toctree.yml

@@ -149,8 +149,8 @@
       title: TensorFlow 모델을 위한 XLA 통합
   title: (번역중) 성능 및 확장성
 - sections:
-    - local: contributing
-      title: 🤗 Transformers에 기여하는 방법
+    - local: in_translation
+      title: (번역중) How to contribute to transformers?
     - local: add_new_model
       title: 🤗 Transformers에 새로운 모델을 추가하는 방법 
     - local: add_tensorflow_model
@@ -337,10 +337,8 @@
         title: (번역중) Jukebox
       - local: in_translation
         title: (번역중) LED
-      - local: model_doc/llama
-        title: LLaMA
-      - local: model_doc/llama2
-        title: LLaMA2
+      - local: in_translation
+        title: (번역중) LLaMA
       - local: in_translation
         title: (번역중) Longformer
       - local: in_translation
@@ -567,8 +565,8 @@
         title: (번역중) Wav2Vec2Phoneme
       - local: in_translation
         title: (번역중) WavLM
-      - local: model_doc/whisper
-        title: Whisper
+      - local: in_translation
+        title: (번역중) Whisper
       - local: in_translation
         title: (번역중) XLS-R
       - local: in_translation

docs/source/ko/add_tensorflow_model.md

@@ -175,6 +175,7 @@ git push -u origin add_tf_brand_new_bert
 
 - `src/transformers/__init__.py`에 *BrandNewBert*의 모든 공개 클래스를 포함합니다.
 - `src/transformers/models/auto/modeling_tf_auto.py`에서 *BrandNewBert* 클래스를 해당 Auto 클래스에 추가합니다.
+- `utils/documentation_tests.txt`에 모델 파일을 문서화하는 테스트 파일 목록을 추가합니다.
 - `src/transformers/utils/dummy_tf_objects.py`에 *BrandNewBert*와 관련된 레이지 로딩 클래스를 추가합니다.
 - `src/transformers/models/brand_new_bert/__init__.py`에서 공개 클래스에 대한 import 구조를 업데이트합니다.
 - `docs/source/en/model_doc/brand_new_bert.md`에서 *BrandNewBert*의 공개 메서드에 대한 문서 포인터를 추가합니다.

docs/source/ko/contributing.md

@@ -1,332 +0,0 @@
-<!---
-Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
--->
-
-# 🤗 Transformers에 기여하기 [[contribute-to-transformers]]
-
-누구나 🤗 Transformers에 기여할 수 있으며, 우리는 모든 사람의 기여를 소중히 생각합니다. 코드 기여는 커뮤니티를 돕는 유일한 방법이 아닙니다. 질문에 답하거나 다른 사람을 도와 문서를 개선하는 것도 매우 가치가 있습니다.
-
-🤗 Transformers를 널리 알리는 것도 큰 도움이 됩니다! 멋진 프로젝트들을 가능하게 한 🤗 Transformers 라이브러리에 대해 블로그 게시글에 언급하거나, 도움이 되었을 때마다 Twitter에 알리거나, 저장소에 ⭐️ 를 표시하여 감사 인사를 전해주세요.
-
-어떤 방식으로 기여하든 [행동 규칙](https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md)을 숙지하고 존중해주세요.
-
-**이 안내서는 멋진 [scikit-learn 기여 안내서](https://github.com/scikit-learn/scikit-learn/blob/main/CONTRIBUTING.md)에서 큰 영감을 받았습니다.**
-
-## 기여하는 방법 [[ways-to-contribute]]
-
-여러 가지 방법으로 🤗 Transformers에 기여할 수 있습니다:
-
-* 기존 코드의 미해결된 문제를 수정합니다.
-* 버그 또는 새로 추가되길 원하는 기능과 관련된 이슈를 제출합니다.
-* 새로운 모델을 구현합니다.
-* 예제나 문서에 기여합니다.
-
-어디서부터 시작할지 모르겠다면, [Good First Issue](https://github.com/huggingface/transformers/contribute) 목록을 확인해보세요. 이 목록은 초보자도 참여하기 쉬운 오픈 이슈 목록을 제공하며, 당신이 오픈소스에 처음으로 기여하는 데 큰 도움이 될 것입니다. 그저 작업하고 싶은 이슈에 댓글만 달아주면 됩니다.
-
-조금 더 도전적인 작업을 원한다면, [Good Second Issue](https://github.com/huggingface/transformers/labels/Good%20Second%20Issue) 목록도 확인해보세요. 이미 당신이 잘 하고 있다고 생각되더라도, 한 번 시도해보세요! 우리도 여러분을 도울 것입니다. 🚀
-
-> 커뮤니티에 이루어지는 모든 기여는 똑같이 소중합니다. 🥰
-
-## 미해결된 문제 수정하기 [[fixing-outstanding-issues]]
-
-기존 코드에서 발견한 문제점에 대한 해결책이 떠오른 경우, 언제든지 [기여를 시작](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md/#create-a-pull-request)하고 Pull Request를 생성해주세요!
-
-## 버그 관련 이슈를 제기하거나 새로운 기능 요청하기 [[submitting-a-bugrelated-issue-or-feature-request]]
-
-버그 관련 이슈를 제기하거나 새로운 기능을 요청할 때는 다음 가이드라인을 최대한 준수해주세요. 이렇게 하면 좋은 피드백과 함께 빠르게 답변해 드릴 수 있습니다.
-
-### 버그를 발견하셨나요? [[did-you-find-a-bug]]
-
-🤗 Transformers 라이브러리는 사용 중에 겪는 문제를 보고해주는 사용자들 덕분에 더욱 견고해지고 신뢰할 수 있게 되었습니다.
-
-이슈를 보고하기 전에, 버그가 이미 **보고되지 않았는지** 확인해주세요. (GitHub의 이슈 탭 아래의 검색 바를 사용하세요). 이슈는 라이브러리 자체에서 발생한 버그어야 하며, 코드의 다른 부분과 관련된 것이 아니어야 합니다. 버그가 라이브러리의 문제로 발생하였는지 확실하지 않은 경우 먼저 [포럼](https://discuss.huggingface.co/)에서 질문해 주세요. 이렇게 하면 일반적인 질문보다 라이브러리와 관련된 문제를 더 빠르게 해결할 수 있습니다.
-
-버그가 이미 보고되지 않았다는 것을 확인했다면, 다음 정보를 포함하여 이슈를 제출해 주세요. 그러면 우리가 빠르게 해결할 수 있습니다:
-
-* 사용 중인 **운영체제 종류와 버전**, 그리고 **Python**, **PyTorch** 또는 **TensorFlow** 버전.
-* 버그를 30초 이내로 재현할 수 있는 간단하고 독립적인 코드 스니펫.
-* 예외가 발생한 경우 *전체* 트레이스백.
-* 스크린샷과 같이 도움이 될 것으로 생각되는 추가 정보를 첨부해 주세요.
-
-운영체제와 소프트웨어 버전을 자동으로 가져오려면 다음 명령을 실행하세요:
-
-```bash
-transformers-cli env
-```
-
-저장소의 루트 디렉터리에서도 같은 명령을 실행할 수 있습니다:
-
-```bash
-python src/transformers/commands/transformers_cli.py env
-```
-
-
-### 새로운 기능을 원하시나요? [[do-you-want-a-new-feature]]
-
-🤗 Transformers에서 사용하고 싶은 새로운 기능이 있다면, 다음 내용을 포함하여 이슈를 제출해 주세요:
-
-1. 이 기능이 필요한 *이유*는 무엇인가요? 라이브러리에 대한 문제나 불만과 관련이 있나요? 프로젝트에 필요한 기능인가요? 커뮤니티에 도움이 될 만한 기능인가요?
-
-   어떤 내용이든 여러분의 이야기를 듣고 싶습니다!
-
-2. 요청하는 기능을 최대한 자세히 설명해 주세요. 더 많은 정보를 제공할수록 더 나은 도움을 드릴 수 있습니다.
-3. 해당 기능의 사용법을 보여주는 *코드 스니펫*을 제공해 주세요.
-4. 기능과 관련된 논문이 있는 경우 링크를 포함해 주세요.
-
-이슈가 잘 작성되었다면 이슈가 생성된 순간, 이미 80% 정도의 작업이 완료된 것입니다. 
-
-이슈를 제기하는 데 도움이 될 만한 [템플릿](https://github.com/huggingface/transformers/tree/main/templates)도 준비되어 있습니다.
-
-## 새로운 모델을 구현하고 싶으신가요? [[do-you-want-to-implement-a-new-model]]
-
-새로운 모델은 계속해서 출시됩니다. 만약 여러분이 새로운 모델을 구현하고 싶다면 다음 정보를 제공해 주세요.
-
-* 모델에 대한 간단한 설명과 논문 링크.
-* 구현이 공개되어 있다면 구현 링크.
-* 모델 가중치가 사용 가능하다면 가중치 링크.
-
-만약 모델을 직접 기여하고 싶으시다면, 알려주세요. 🤗 Transformers에 추가할 수 있도록 도와드리겠습니다!
-
-새로운 모델을 추가하는 방법에 대한 [상세 안내서와 템플릿](https://github.com/huggingface/transformers/tree/main/templates)을 제공하고 있으며, [🤗 Transformers에 새로운 모델을 추가하는 방법](https://huggingface.co/docs/transformers/add_new_model)에 대한 기술적인 안내서도 있습니다.
-
-## 문서를 추가하고 싶으신가요? [[do-you-want-to-add-documentation]]
-
-우리는 언제나 더 명확하고 정확한 문서를 제공하기 위하여 개선점을 찾고 있습니다. 오탈자나 부족한 내용, 분명하지 않거나 부정확한 내용 등을 알려주시면 개선하는 데 도움이 됩니다. 관심이 있으시다면 변경하거나 기여하실 수 있도록 도와드리겠습니다!
-
-문서를 생성, 빌드 및 작성하는 방법에 대한 자세한 내용은 [README](https://github.com/huggingface/transformers/tree/main/docs) 문서를 확인해 주세요.
-
-## 풀 리퀘스트(Pull Request) 생성하기 [[create-a-pull-request]]
-
-코드를 작성하기 전에 기존의 Pull Request나 이슈를 검색하여 누군가 이미 동일한 작업을 하고 있는지 확인하는 것이 좋습니다. 확실하지 않다면 피드백을 받기 위해 이슈를 열어보는 것이 좋습니다.
-
-🤗 Transformers에 기여하기 위해서는 기본적인 `git` 사용 능력이 필요합니다. `git`은 사용하기 쉬운 도구는 아니지만, 매우 훌륭한 매뉴얼을 제공합니다. 쉘(shell)에서 `git --help`을 입력하여 확인해보세요! 만약 책을 선호한다면, [Pro Git](https://git-scm.com/book/en/v2)은 매우 좋은 참고 자료가 될 것입니다.
-
-🤗 Transformers에 기여하려면 **[Python 3.8]((https://github.com/huggingface/transformers/blob/main/setup.py#L426))** 이상의 버전이 필요합니다. 기여를 시작하려면 다음 단계를 따르세요:
-
-1. 저장소 페이지에서 **[Fork](https://github.com/huggingface/transformers/fork)** 버튼을 클릭하여 저장소를 포크하세요. 이렇게 하면 코드의 복사본이 여러분의 GitHub 사용자 계정 아래에 생성됩니다.
-
-2. 포크한 저장소를 로컬 디스크로 클론하고, 기본 저장소를 원격(remote)으로 추가하세요:
-
-   ```bash
-   git clone git@github.com:<your Github handle>/transformers.git
-   cd transformers
-   git remote add upstream https://github.com/huggingface/transformers.git
-   ```
-
-3. 개발 변경 사항을 저장할 새 브랜치를 생성하세요:
-
-   ```bash
-   git checkout -b a-descriptive-name-for-my-changes
-   ```
-
-   🚨 절대 `main` 브랜치에서 작업하지 **마세요!**
-
-4. 가상 환경에서 다음 명령을 실행하여 개발 환경을 설정하세요:
-
-   ```bash
-   pip install -e ".[dev]"
-   ```
-
-   만약 이미 가상 환경에 🤗 Transformers가 설치되어 있다면, `-e` 플래그를 사용하여 설치하기 전에 `pip uninstall transformers`로 제거해주세요.
-   
-   여러분의 운영체제에 따라서, 그리고 🤗 Transformers의 선택적 의존성의 수가 증가하면서, 이 명령이 실패할 수도 있습니다. 그럴 경우 사용하려는 딥러닝 프레임워크(PyTorch, TensorFlow, 그리고/또는 Flax)를 설치한 후 아래 명령을 실행해주세요:
-
-   ```bash
-   pip install -e ".[quality]"
-   ```
-
-   대부분의 경우 이것으로 충분할 것입니다.
-
-5. 브랜치에서 기능을 개발하세요.
-
-   코드를 작업하는 동안 테스트 스위트(test suite)가 통과하는지 확인하세요. 다음과 같이 변경 사항에 영향을 받는 테스트를 실행하세요:
-
-   ```bash
-   pytest tests/<TEST_TO_RUN>.py
-   ```
-
-   테스트에 대한 더 많은 정보는 [테스트](https://huggingface.co/docs/transformers/testing) 가이드를 확인하세요.
-
-   🤗 Transformers는 `black`과 `ruff`를 사용하여 소스 코드의 형식을 일관되게 유지합니다. 변경 사항을 적용한 후에는 다음 명령으로 자동으로 스타일 교정 및 코드 검증을 수행하세요:
-
-   ```bash
-   make fixup
-   ```
-
-   이것은 또한 작업 중인 PR에서 수정한 파일에서만 작동하도록 최적화되어 있습니다.
-
-   검사를 하나씩 실행하려는 경우, 다음 명령으로 스타일 교정을 적용할 수 있습니다:
-
-   ```bash
-   make style
-   ```
-
-   🤗 Transformers는 또한 `ruff`와 몇 가지 사용자 정의 스크립트를 사용하여 코딩 실수를 확인합니다. CI를 통해 품질 관리가 수행되지만, 다음 명령으로 동일한 검사를 실행할 수 있습니다:
-
-   ```bash
-   make quality
-   ```
-
-   마지막으로, 새 모델을 추가할 때 일부 파일을 업데이트하는 것을 잊지 않도록 하기 위한 많은 스크립트가 있습니다. 다음 명령으로 이러한 스크립트를 실행할 수 있습니다:
-
-   ```bash
-   make repo-consistency
-   ```
-
-   이러한 검사에 대해 자세히 알아보고 관련 문제를 해결하는 방법은 [Pull Request에 대한 검사](https://huggingface.co/docs/transformers/pr_checks) 가이드를 확인하세요.
-
-   만약 `docs/source` 디렉터리 아래의 문서를 수정하는 경우, 문서가 빌드될 수 있는지 확인하세요. 이 검사는 Pull Request를 열 때도 CI에서 실행됩니다. 로컬 검사를 실행하려면 문서 빌더를 설치해야 합니다:
-   
-   ```bash
-   pip install ".[docs]"
-   ```
-
-   저장소의 루트 디렉터리에서 다음 명령을 실행하세요:
-
-   ```bash
-   doc-builder build transformers docs/source/en --build_dir ~/tmp/test-build
-   ```
-
-   이 명령은 `~/tmp/test-build` 폴더에 문서를 빌드하며, 생성된 Markdown 파일을 선호하는 편집기로 확인할 수 있습니다. Pull Request를 열 때 GitHub에서 문서를 미리 볼 수도 있습니다.
-
-   변경 사항에 만족하면 `git add`로 변경된 파일을 추가하고, `git commit`으로 변경 사항을 로컬에 기록하세요:
-
-   ```bash
-   git add modified_file.py
-   git commit
-   ```
-
-   [좋은 커밋 메시지](https://chris.beams.io/posts/git-commit/)를 작성하여 변경 사항을 명확하게 전달하세요!
-
-   변경 사항을 프로젝트 원본 저장소와 동기화하려면, PR을 *열기 전에* 브랜치를 `upstream/branch`로 리베이스(rebase)하세요. 또는 관리자의 요청에 이 작업이 필요할 수 있습니다:
-
-   ```bash
-   git fetch upstream
-   git rebase upstream/main
-   ```
-   
-   변경 사항을 브랜치에 푸시하세요:
-
-   ```bash
-   git push -u origin a-descriptive-name-for-my-changes
-   ```
-
-   이미 PR을 열었다면, `--force` 플래그와 함께 강제 푸시해야 합니다. 아직 PR이 열리지 않았다면 정상적으로 변경 사항을 푸시하면 됩니다.
-
-6. 이제 GitHub에서 포크한 저장소로 이동하고 **Pull request(풀 리퀘스트)**를 클릭하여 Pull Request를 열 수 있습니다. 아래의 [체크리스트](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md/#pull-request-checklist)에서 모든 항목에 체크 표시를 하세요. 준비가 완료되면 프로젝트 관리자에게 변경 사항을 보내 검토를 요청할 수 있습니다.
-
-7. 관리자가 변경 사항을 요청해도 괜찮습니다. 핵심 기여자들도 동일한 상황을 겪습니다! 모두가 변경 사항을 Pull Request에서 볼 수 있도록, 로컬 브랜치에서 작업하고 변경 사항을 포크한 저장소로 푸시하세요. 그러면 변경 사항이 자동으로 Pull Request에 나타납니다.
-
-### Pull Request 체크리스트 [[pull-request-checklist]]
-
-☐ Pull Request 제목은 기여 내용을 요약해야 합니다.<br>
-☐ Pull Request가 이슈를 해결하는 경우, Pull Request 설명에 이슈 번호를 언급하여 연관되어 있음을 알려주세요. (이슈를 확인하는 사람들이 해당 이슈에 대한 작업이 진행 중임을 알 수 있게 합니다).<br>
-☐ 작업이 진행중이라면 제목 앞에 `[WIP]`를 붙여주세요. 중복 작업을 피하고 병합할 준비가 된 PR과 구분하기에 유용합니다.<br>
-☐ 기존 테스트를 통과하는지 확인하세요.<br>
-☐ 새로운 기능을 추가하는 경우, 해당 기능에 대한 테스트도 추가하세요.<br>
-   - 새 모델을 추가하는 경우, `ModelTester.all_model_classes = (MyModel, MyModelWithLMHead,...)`을 사용하여 일반적인 테스트를 활성화하세요.
-   - 새 `@slow` 테스트를 추가하는 경우, 다음 명령으로 테스트를 통과하는지 확인하세요: `RUN_SLOW=1 python -m pytest tests/models/my_new_model/test_my_new_model.py`.
-   - 새 토크나이저를 추가하는 경우, 테스트를 작성하고 다음 명령으로 테스트를 통과하는지 확인하세요: `RUN_SLOW=1 python -m pytest tests/models/{your_model_name}/test_tokenization_{your_model_name}.py`. 
-   - CircleCI에서는 느린 테스트를 실행하지 않지만, GitHub Actions에서는 매일 밤 실행됩니다!<br>
-
-☐ 모든 공개 메소드는 유용한 기술문서를 가져야 합니다 (예를 들어 [`modeling_bert.py`](https://github.com/huggingface/transformers/blob/main/src/transformers/models/bert/modeling_bert.py) 참조).<br>
-☐ 저장소가 빠르게 성장하고 있으므로 저장소에 상당한 부담을 주는 이미지, 동영상 및 기타 텍스트가 아닌 파일은 추가하지 마세요. 대신 [`hf-internal-testing`](https://huggingface.co/hf-internal-testing)과 같은 Hub 저장소를 사용하여 이러한 파일을 호스팅하고 URL로 참조하세요. 문서와 관련된 이미지는 다음 저장소에 배치하는 것을 권장합니다: [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images). 이 데이터셋 저장소에서 PR을 열어서 Hugging Face 멤버에게 병합을 요청할 수 있습니다.
-
-Pull Request에서 실행되는 검사에 대한 자세한 정보는 [Pull Request에 대한 검사](https://huggingface.co/docs/transformers/pr_checks) 가이드를 확인하세요.
-
-### 테스트 [[tests]]
-
-라이브러리 동작과 여러 예제를 테스트할 수 있는 광범위한 테스트 스위트가 포함되어 있습니다. 라이브러리 테스트는 [tests](https://github.com/huggingface/transformers/tree/main/tests) 폴더에, 예제 테스트는 [examples](https://github.com/huggingface/transformers/tree/main/examples) 폴더에 있습니다.
-
-속도가 빠른 `pytest`와 `pytest-xdist`를 선호합니다. 저장소의 루트 디렉터리에서 테스트를 실행할 *하위 폴더 경로 또는 테스트 파일 경로*를 지정하세요.
-
-```bash
-python -m pytest -n auto --dist=loadfile -s -v ./tests/models/my_new_model
-```
-
-마찬가지로 `examples` 디렉터리에서도 *하위 폴더 경로 또는 테스트 파일 경로*를 지정하세요. 예를 들어, 다음 명령은 PyTorch `examples` 디렉터리의 텍스트 분류 하위 폴더를 테스트합니다:
-
-```bash
-pip install -r examples/xxx/requirements.txt  # only needed the first time
-python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/text-classification
-```
-
-이것이 실제로 `make test` 및 `make test-examples` 명령이 구현되는 방식입니다 (`pip install`은 제외합니다)!
-
-또한 특정 기능만 테스트하기 위한 더 작은 테스트를 지정할 수 있습니다.
-
-기본적으로 느린 테스트는 건너뛰지만 `RUN_SLOW` 환경 변수를 `yes`로 설정하여 실행할 수 있습니다. 이렇게 하면 많은 기가바이트 단위의 모델이 다운로드되므로 충분한 디스크 공간, 좋은 인터넷 연결과 많은 인내가 필요합니다!
-
-<Tip warning={true}>
-
-테스트를 실행하려면 *하위 폴더 경로 또는 테스트 파일 경로*를 지정하세요. 그렇지 않으면 `tests` 또는 `examples` 폴더의 모든 테스트를 실행하게 되어 매우 긴 시간이 걸립니다!
-
-</Tip>
-
-```bash
-RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./tests/models/my_new_model
-RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/text-classification
-```
-
-느린 테스트와 마찬가지로, 다음과 같이 테스트 중에 기본적으로 활성화되지 않는 다른 환경 변수도 있습니다:
-- `RUN_CUSTOM_TOKENIZERS`: 사용자 정의 토크나이저 테스트를 활성화합니다.
-- `RUN_PT_FLAX_CROSS_TESTS`: PyTorch + Flax 통합 테스트를 활성화합니다.
-- `RUN_PT_TF_CROSS_TESTS`: TensorFlow + PyTorch 통합 테스트를 활성화합니다.
-
-더 많은 환경 변수와 추가 정보는 [testing_utils.py](src/transformers/testing_utils.py)에서 찾을 수 있습니다.
-
-🤗 Transformers는 테스트 실행기로 `pytest`를 사용합니다. 그러나 테스트 스위트 자체에서는 `pytest` 관련 기능을 사용하지 않습니다.
-
-이것은 `unittest`가 완전히 지원된다는 것을 의미합니다. 다음은 `unittest`로 테스트를 실행하는 방법입니다:
-
-```bash
-python -m unittest discover -s tests -t . -v
-python -m unittest discover -s examples -t examples -v
-```
-
-### 스타일 가이드 [[style-guide]]
-
-문서는 [Google Python 스타일 가이드](https://google.github.io/styleguide/pyguide.html)를 따릅니다. 자세한 정보는 [문서 작성 가이드](https://github.com/huggingface/transformers/tree/main/docs#writing-documentation---specification)를 확인하세요.
-
-### Windows에서 개발 [[develop-on-windows]]
-
-Windows에서 개발할 경우([Windows Subsystem for Linux](https://learn.microsoft.com/en-us/windows/wsl/) 또는 WSL에서 작업하지 않는 한) Windows `CRLF` 줄 바꿈을 Linux `LF` 줄 바꿈으로 변환하도록 git을 구성해야 합니다:
-
-```bash
-git config core.autocrlf input
-```
-
-Windows에서 `make` 명령을 실행하는 한 가지 방법은 MSYS2를 사용하는 것입니다:
-
-1. [MSYS2](https://www.msys2.org/)를 다운로드합니다. `C:\msys64`에 설치되었다고 가정합니다.
-2. CLI에서 `C:\msys64\msys2.exe`를 엽니다 (시작 메뉴에서 사용 가능해야 함).
-3. 쉘에서 다음을 실행하여: `pacman -Syu` 및 `pacman -S make`로 `make`를 설치합니다.
-4. 환경 변수 PATH에 `C:\msys64\usr\bin`을 추가하세요.
-
-이제 모든 터미널 (Powershell, cmd.exe 등)에서 `make`를 사용할 수 있습니다! 🎉
-
-### 포크한 저장소를 상위 원본 브랜치(main)과 동기화하기 (Hugging Face 저장소) [[sync-a-forked-repository-with-upstream-main-the-hugging-face-repository]]
-
-포크한 저장소의 main 브랜치를 업데이트할 때, 다음 단계를 따라 수행해주세요. 이렇게 하면 각 upstream PR에 참조 노트가 추가되는 것을 피하고 이러한 PR에 관여하는 개발자들에게 불필요한 알림이 전송되는 것을 방지할 수 있습니다.
-
-1. 가능하면 포크된 저장소의 브랜치 및 PR을 사용하여 upstream과 동기화하지 마세요. 대신 포크된 main 저장소에 직접 병합하세요.
-2. PR이 반드시 필요한 경우, 브랜치를 확인한 후 다음 단계를 사용하세요:
-
-```bash
-git checkout -b your-branch-for-syncing
-git pull --squash --no-commit upstream main
-git commit -m '<your message without GitHub references>'
-git push --set-upstream origin your-branch-for-syncing
-```

docs/source/ko/model_doc/llama.md

@@ -1,117 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ 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.
-
--->
-
-# LLaMA [[llama]]
-
-## 개요 [[overview]]
-
-LLaMA 모델은 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 Lample에 의해 제안된 [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971)에서 소개되었습니다. 이 모델은 7B에서 65B개의 파라미터까지 다양한 크기의 기초 언어 모델을 모아놓은 것입니다.
-
-논문의 초록은 다음과 같습니다:
-
-*"LLaMA는 7B에서 65B개의 파라미터 수를 가진 기초 언어 모델의 모음입니다. 우리는 수조 개의 토큰으로 모델을 훈련시켰고, 공개적으로 이용 가능한 데이터셋만을 사용하여 최고 수준의 모델을 훈련시킬 수 있음을 보여줍니다. 특히, LLaMA-13B 모델은 대부분의 벤치마크에서 GPT-3 (175B)를 능가하며, LLaMA-65B는 최고 수준의 모델인 Chinchilla-70B와 PaLM-540B에 버금가는 성능을 보입니다. 우리는 모든 모델을 연구 커뮤니티에 공개합니다."*
-
-팁:
-
-- LLaMA 모델의 가중치는 [이 양식](https://docs.google.com/forms/d/e/1FAIpQLSfqNECQnMkycAp2jP4Z9TFX0cGR4uf7b_fBxjY_OjhJILlKGA/viewform?usp=send_form)을 작성하여 얻을 수 있습니다.
-- 가중치를 다운로드한 후에는 이를 [변환 스크립트](https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/convert_llama_weights_to_hf.py)를 사용하여 Hugging Face Transformers 형식으로 변환해야합니다. 변환 스크립트를 실행하려면 아래의 예시 명령어를 참고하세요:
-
-```bash
-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
-```
-
-- 변환을 하였다면 모델과 토크나이저는 다음과 같이 로드할 수 있습니다:
-
-```python
-from transformers import LlamaForCausalLM, LlamaTokenizer
-
-tokenizer = LlamaTokenizer.from_pretrained("/output/path")
-model = LlamaForCausalLM.from_pretrained("/output/path")
-```
-
-스크립트를 실행하기 위해서는 모델을 float16 정밀도로 전부 로드할 수 있을 만큼의 충분한 CPU RAM이 필요합니다. (가장 큰 버전의 모델이 여러 체크포인트로 나뉘어 있더라도, 각 체크포인트는 모델의 각 가중치의 일부를 포함하고 있기 때문에 모든 체크포인트를 RAM에 로드해야 합니다) 65B 모델의 경우, 총 130GB의 RAM이 필요합니다.
-
-
-- LLaMA 토크나이저는 [sentencepiece](https://github.com/google/sentencepiece)를 기반으로 하는 BPE 모델입니다. sentencepiece의 특징 중 하나는 시퀀스를 디코딩할 때 첫 토큰이 단어의 시작이라면 (예를 들어 "Banana"), 토크나이저는 문자열 앞에 공백을 추가하지 않는다는 것입니다.
-
-이 모델은 [BlackSamorez](https://huggingface.co/BlackSamorez)의 기여와 함께, [zphang](https://huggingface.co/zphang)에 의해 제공되었습니다. Hugging Face에서의 구현 코드는 GPT-NeoX를 기반으로 하며 [여기](https://github.com/EleutherAI/gpt-neox)에서 찾을 수 있고, 저자의 코드 원본은 [여기](https://github.com/facebookresearch/llama)에서 확인할 수 있습니다.
-
-
-원래 LLaMA 모델을 기반으로 Meta AI에서 몇 가지 후속 작업을 발표했습니다:
-
-- **Llama2**: Llama2는 구조적인 몇 가지 수정(Grouped Query Attention)을 통해 개선된 버전이며, 2조 개의 토큰으로 사전 훈련이 되어 있습니다. Llama2에 대한 자세한 내용은 [이 문서](llama2)를 참고하세요.
-
-## 리소스 [[resources]]
-
-LLaMA를 시작하는 데 도움이 될 Hugging Face 및 커뮤니티(🌎로 표시)의 공식 자료 목록입니다. 여기에 자료를 제출하고 싶다면 Pull Request를 올려주세요! 추가할 자료는 기존의 자료와 중복되지 않고 새로운 내용을 보여주는 것이 좋습니다.
-
-<PipelineTag pipeline="text-classification"/>
-
-- LLaMA 모델을 텍스트 분류 작업에 적용하기 위한 프롬프트 튜닝 방법에 대한 [노트북](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-sst2.ipynb#scrollTo=f04ba4d2) 🌎
-
-<PipelineTag pipeline="question-answering"/>
-
-- [Stack Exchange](https://stackexchange.com/)에서 질문에 답하는 LLaMA를 훈련하는 방법을 위한 [StackLLaMA: RLHF로 LLaMA를 훈련하는 실전 가이드](https://huggingface.co/blog/stackllama#stackllama-a-hands-on-guide-to-train-llama-with-rlhf) 🌎
-
-⚗️ 최적화
-- 제한된 메모리를 가진 GPU에서 xturing 라이브러리를 사용하여 LLaMA 모델을 미세 조정하는 방법에 대한 [노트북](https://colab.research.google.com/drive/1SQUXq1AMZPSLD4mk3A3swUIc6Y2dclme?usp=sharing) 🌎
-
-⚡️ 추론
-- 🤗 PEFT 라이브러리의 PeftModel을 사용하여 LLaMA 모델을 실행하는 방법에 대한 [노트북](https://colab.research.google.com/github/DominguesM/alpaca-lora-ptbr-7b/blob/main/notebooks/02%20-%20Evaluate.ipynb) 🌎
-- LangChain을 사용하여 PEFT 어댑터 LLaMA 모델을 로드하는 방법에 대한 [노트북](https://colab.research.google.com/drive/1l2GiSSPbajVyp2Nk3CFT4t3uH6-5TiBe?usp=sharing) 🌎
-
-🚀 배포
-- 🤗 PEFT 라이브러리와 사용자 친화적인 UI로 LLaMA 모델을 미세 조정하는 방법에 대한 [노트북](https://colab.research.google.com/github/lxe/simple-llama-finetuner/blob/master/Simple_LLaMA_FineTuner.ipynb#scrollTo=3PM_DilAZD8T) 🌎
-- Amazon SageMaker에서 텍스트 생성을 위해 Open-LLaMA 모델을 배포하는 방법에 대한 [노트북](https://github.com/aws/amazon-sagemaker-examples/blob/main/introduction_to_amazon_algorithms/jumpstart-foundation-models/text-generation-open-llama.ipynb) 🌎
-
-## LlamaConfig [[llamaconfig]]
-
-[[autodoc]] LlamaConfig
-
-
-## LlamaTokenizer [[llamatokenizer]]
-
-[[autodoc]] LlamaTokenizer
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - save_vocabulary
-
-## LlamaTokenizerFast [[llamatokenizerfast]]
-
-[[autodoc]] LlamaTokenizerFast
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - update_post_processor
-    - save_vocabulary
-
-## LlamaModel [[llamamodel]]
-
-[[autodoc]] LlamaModel
-    - forward
-
-
-## LlamaForCausalLM [[llamaforcausallm]]
-
-[[autodoc]] LlamaForCausalLM
-    - forward
-
-## LlamaForSequenceClassification [[llamaforsequenceclassification]]
-
-[[autodoc]] LlamaForSequenceClassification
-    - forward

docs/source/ko/model_doc/llama2.md

@@ -1,129 +0,0 @@
-<!--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 contains specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Llama2 [[llama2]]
-
-## 개요 [[overview]]
-
-Llama2 모델은 Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Ya1smine 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, Pushkar Mishra, 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 Scialom의 논문 [LLaMA: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/)에서 제안되었습니다. 채팅 어플리케이션에 맞게 미세 조정된 체크포인트를 포함된 7B에서 70B 범위의 매개변수를 가진 기초 언어 모델 모음입니다!
-
-논문의 초록은 다음과 같습니다:
-
-*이 연구에서 우리는 70억에서 700억 파라미터의 범위에서 사전 훈련 및 미세 조정된 대규모 언어 모델(LLMs)의 모음인 Llama 2를 개발 및 공개합니다. Llama 2-Chat라고 불리는 미세 조정된 LLMs은 대화 사용 사례에 최적화되었습니다. 우리의 모델은 테스트한 대부분의 벤치마크에서 오픈 소스 채팅 모델보다 성능이 뛰어나며, 유용성과 안전성에 대한 인적 평가를 바탕으로 비공개 소스 모델을 대체할 수 있는 적절한 대안이 될 수 있습니다. 우리는 Llama 2-Chat의 미세 조정 및 안전성 향상의 접근 방식에 대한 자세한 설명을 제공하여 커뮤니티가 우리의 작업을 기반으로 LLMs의 책임있는 개발에 기여할 수 있도록 합니다.*
-
-[여기](https://huggingface.co/models?search=llama2)에서 모든 Llama2 모델을 확인할 수 있습니다.
-
-<Tip warning={true}>
-
-`Llama2` 모델은 `bfloat16`을 사용하여 훈련되었지만, 원래 추론은 `float16`을 사용합니다. 허브에 업로드된 체크포인트는 `torch_dtype = 'float16'`을 사용하며, 이는 `AutoModel` API에 의해 체크포인트를 `torch.float32`에서 `torch.float16`으로 캐스팅하는 데 사용됩니다. 
-
-온라인 가중치의 `dtype`은 `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`를 사용하여 모델을 초기화할 때 `torch_dtype="auto"`를 사용하지 않는 한 대부분 관련이 없습니다. 그 이유는 모델이 먼저 다운로드될 것이고 (온라인 체크포인트의 `dtype`을 사용하여) 그다음에 기본 `dtype`인 `torch`로 캐스팅하고(`torch.float32`가 됨), 마지막으로 구성(configuration)에서 제공된 `torch_dtype`이 있는 경우 이를 사용하기 때문입니다.
-
-모델을 `float16`에서 훈련하는 것은 권장되지 않으며 `nan`을 생성하는 것으로 알려져 있습니다. 따라서 모델은 `bfloat16`에서 훈련되어야 합니다.
-
-</Tip>
-
-🍯 팁:
-
-- Llama2 모델의 가중치는 [이 양식](https://ai.meta.com/resources/models-and-libraries/llama-downloads/)을 작성하여 얻을 수 있습니다.
-- 아키텍처는 처음 버전의 Llama와 매우 유사하며, [이 논문](https://arxiv.org/pdf/2305.13245.pdf)의 내용에 따라 Grouped Query Attention (GQA)이 추가되었습니다.
-- `config.pretraining_tp`를 1과 다른 값으로 설정하면 더 정확하지만 느린 선형 레이어 계산이 활성화되어 원본 로짓과 더 잘 일치하게 됩니다.
-- 원래 모델은 `pad_id = -1`을 사용하는데, 이는 패딩 토큰이 없음을 의미합니다. 동일한 로직을 사용할 수 없으므로 `tokenizer.add_special_tokens({"pad_token":"<pad>"})`를 사용하여 패딩 토큰을 추가하고 이에 따라 토큰 임베딩 크기를 조정해야 합니다. 또한 `model.config.pad_token_id`를 설정해야 합니다. 모델의 `embed_tokens` 레이어는 `self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.config.padding_idx)`로 초기화되어, 패딩 토큰 인코딩이 0을 출력하도록 합니다. 따라서 초기화 시에 전달하는 것을 권장합니다.
-- 양식을 작성하고 모델 체크포인트 접근 권한을 얻은 후에는 이미 변환된 체크포인트를 사용할 수 있습니다. 그렇지 않고 자신의 모델을 직접 변환하려는 경우, [변환 스크립트](https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/convert_llama_weights_to_hf.py)를 자유롭게 사용하세요. 스크립트는 다음과 같은 예시의 명령어로 호출할 수 있습니다:
-
-```bash
-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
-```
-
-- 변환 후 모델과 토크나이저는 다음과 같이 로드할 수 있습니다:
-
-```python
-from transformers import LlamaForCausalLM, LlamaTokenizer
-
-tokenizer = LlamaTokenizer.from_pretrained("/output/path")
-model = LlamaForCausalLM.from_pretrained("/output/path")
-```
-
-스크립트를 실행하려면 모델을 float16 정밀도로 전부 호스트할 수 있을 만큼 충분한 CPU RAM이 필요합니다 (가장 큰 버전이 여러 체크포인트로 제공되더라도 각 체크포인트는 모델 가중치의 일부만을 포함하므로 모두 RAM에 로드해야 합니다). 75B 모델의 경우, 총 145GB의 RAM이 필요합니다.
-
-- LLaMA 토크나이저는 [sentencepiece](https://github.com/google/sentencepiece)를 기반으로 한 BPE 모델입니다. sentencepiece의 특징 중 하나는 시퀀스를 디코딩할 때 첫 번째 토큰이 단어의 시작이면 (예: "Banana") 토크나이저는 문자열 앞에 접두사 공간을 추가하지 않는 것입니다.
-
-이 모델은 [Arthur Zucker](https://huggingface.co/ArthurZ)가 [Lysandre Debut](https://huggingface.co/lysandre)의 도움을 받아 제공하였습니다. Hugging Face에서의 구현 코드는 [여기](https://github.com/EleutherAI/gpt-neox)의 GPT-NeoX 를 기반으로 합니다. 저자의 원래 코드는 [여기](https://github.com/facebookresearch/llama)에서 찾을 수 있습니다.
-
-## 리소스 [[resources]]
-
-LLaMA2를 시작하는 데 도움이 될 Hugging Face의 공식 및 커뮤니티(🌎로 표시) 리소스 목록입니다. 여기에 새로운 리소스를 추가하기 위해서 Pull Request를 열어 주시면 검토하겠습니다! 리소스는 기존 리소스와 중복되지 않는 새로운 것을 보여주는 것이 이상적입니다.
-
-- [Llama 2 is here - get it on Hugging Face](https://huggingface.co/blog/llama2), Llama 2에 관한 블로그 포스트와 🤗 Transformers 및 🤗 PEFT와 함께 사용하는 방법에 대한 내용입니다.
-- [LLaMA 2 - Every Resource you need](https://www.philschmid.de/llama-2), LLaMA 2에 대해 알아보고 빠르게 시작하는 데 필요한 관련 리소스의 모음입니다.
-
-<PipelineTag pipeline="text-generation"/>
-
-- Google Colab에서 QLoRA와 4-bit 정밀도를 사용하여 Llama 2를 미세 조정하는 방법에 대한 [노트북](https://colab.research.google.com/drive/1PEQyJO1-f6j0S_XJ8DV50NkpzasXkrzd?usp=sharing)입니다. 🌎
-- "Llama-v2-7b-guanaco" 모델을 4-bit QLoRA로 미세 조정하고 PDF에서 Q&A 데이터셋을 생성하는 방법에 대한 [노트북](https://colab.research.google.com/drive/134o_cXcMe_lsvl15ZE_4Y75Kstepsntu?usp=sharing)입니다. 🌎
-
-⚗️ 최적화
-- [Llama 2를 DPO로 미세 조정하기](https://huggingface.co/blog/dpo-trl), TRL 라이브러리의 DPO 방법을 사용하여 특정 데이터셋에서 Llama 2를 미세 조정하는 방법을 안내하는 가이드입니다.
-- [확장 가이드: Llama 2 명령어 조정](https://www.philschmid.de/instruction-tune-llama-2), 입력에서 명령어를 생성하도록 Llama 2를 훈련시키는 방법을 안내하는 가이드로, 명령어를 따르는 모델에서 명령어를 주는 모델로 변환합니다.
-- 개인 컴퓨터에서 QLoRA와 TRL을 사용하여 Llama 2 모델을 미세 조정하는 방법에 대한 [노트북](https://colab.research.google.com/drive/1SYpgFpcmtIUzdE7pxqknrM4ArCASfkFQ?usp=sharing)입니다. 🌎
-
-⚡️ 추론
-- AutoGPTQ 라이브러리의 GPTQ를 사용하여 Llama 2 모델을 양자화하는 방법에 대한 [노트북](https://colab.research.google.com/drive/1TC56ArKerXUpbgRy5vM3woRsbTEVNq7h?usp=sharing)입니다. 🌎
-- 로컬 컴퓨터나 Google Colab에서 4-bit 양자화로 Llama 2 채팅 모델을 실행하는 방법에 대한 [노트북](https://colab.research.google.com/drive/1X1z9Q6domMKl2CnEM0QGHNwidLfR4dW2?usp=sharing)입니다. 🌎
-
-🚀 배포
-- [Amazon SageMaker에서 LLaMA 2 (7-70B) 미세 조정하기](https://www.philschmid.de/sagemaker-llama2-qlora), Amazon SageMaker에서 QLoRA 미세 조정 및 배포에 이르기까지의 완전한 가이드입니다.
-- [Amazon SageMaker에서 Llama 2 7B/13B/70B 배포하기](https://www.philschmid.de/sagemaker-llama-llm), 안전하고 확장 가능한 배포를 위해 Hugging Face의 LLM DLC 컨테이너를 사용하는 방법에 대한 가이드입니다.
-
-
-## LlamaConfig [[llamaconfig]]
-
-[[autodoc]] LlamaConfig
-
-
-## LlamaTokenizer [[llamatokenizer]]
-
-[[autodoc]] LlamaTokenizer
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - save_vocabulary
-
-## LlamaTokenizerFast [[llamatokenizerfast]]
-
-[[autodoc]] LlamaTokenizerFast
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - update_post_processor
-    - save_vocabulary
-
-## LlamaModel [[llamamodel]]
-
-[[autodoc]] LlamaModel
-    - forward
-
-
-## LlamaForCausalLM [[llamaforcausallm]]
-
-[[autodoc]] LlamaForCausalLM
-    - forward
-
-## LlamaForSequenceClassification [[llamaforsequenceclassification]]
-
-[[autodoc]] LlamaForSequenceClassification
-    - forward

docs/source/ko/model_doc/whisper.md

@@ -1,120 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ 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.
-
--->
-
-# Whisper [[whisper]]
-
-## 개요 [[overview]]
-
-Whisper 모델은 Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever에 의해 [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf)에서 제안되었습니다.
-
-논문의 초록은 다음과 같습니다:
-
-*우리는 인터넷에서 대량의 오디오를 글로 옮긴 것을 예측하도록 간단히 훈련된 음성 처리 시스템의 성능을 연구합니다. 68만 시간의 다국어 및 다중 작업 지도(multitask supervision)에 확장했을 때, 결과 모델은 표준 벤치마크에 잘 일반화되며, 미세 조정이 필요 없는 제로샷 전송 설정에서 이전의 완전히 지도된(fully-supervised) 결과와 경쟁할 수 있는 경우가 많습니다. 사람과 비교하면, 이 모델은 사람의 정확도와 견고성에 근접합니다. 우리는 강력한 음성 처리를 위한 추가 작업의 기반이 될 모델과 추론 코드를 공개합니다.*
-
-
-
-팁:
-
-- 이 모델은 일반적으로 별도의 미세 조정 없이도 잘 작동합니다.
-- 아키텍처는 고전적인 인코더-디코더 아키텍처를 따르기 때문에, 추론을 위해 [`~generation.GenerationMixin.generate`] 함수를 사용합니다.
-- 현재 추론은 짧은 형식에만 구현되어 있으며, 오디오는 30초 미만의 세그먼트로 미리 분할되어야 합니다. 타임스탬프를 포함한 긴 형식에 대한 추론은 향후 릴리스에서 구현될 예정입니다.
-- [`WhisperProcessor`]를 사용하여 모델에 사용할 오디오를 준비하고, 예측된 ID를 텍스트로 디코딩할 수 있습니다.
-
-이 모델은 [Arthur Zucker](https://huggingface.co/ArthurZ)에 의해 제공되었습니다. 이 모델의 Tensorflow 버전은 [amyeroberts](https://huggingface.co/amyeroberts)에 의해 제공되었습니다.
-원본 코드는 [여기](https://github.com/openai/whisper)에서 찾을 수 있습니다.
-
-
-
-## WhisperConfig [[whisperconfig]]
-
-[[autodoc]] WhisperConfig
-
-## WhisperTokenizer [[whispertokenizer]]
-
-[[autodoc]] WhisperTokenizer
-    - set_prefix_tokens
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - save_vocabulary
-
-## WhisperTokenizerFast [[whispertokenizerfast]]
-
-[[autodoc]] WhisperTokenizerFast
-    - set_prefix_tokens
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - save_vocabulary
-
-## WhisperFeatureExtractor [[whisperfeatureextractor]]
-
-[[autodoc]] WhisperFeatureExtractor
-    - __call__
-
-## WhisperProcessor [[whisperprocessor]]
-
-[[autodoc]] WhisperProcessor
-    - __call__
-    - from_pretrained
-    - save_pretrained
-    - batch_decode
-    - decode
-
-## WhisperModel [[whispermodel]]
-
-[[autodoc]] WhisperModel
-    - forward
-    - _mask_input_features
-
-## WhisperForConditionalGeneration [[whisperforconditionalgeneration]]
-
-[[autodoc]] WhisperForConditionalGeneration
-    - forward
-
-## WhisperForAudioClassification [[whisperforaudioclassification]]
-
-[[autodoc]] WhisperForAudioClassification
-    - forward
-
-
-
-## TFWhisperModel [[tfwhispermodel]]
-
-[[autodoc]] TFWhisperModel
-    - call
-
-## TFWhisperForConditionalGeneration [[tfwhisperforconditionalgeneration]]
-
-[[autodoc]] TFWhisperForConditionalGeneration
-    - call
-
-
-## FlaxWhisperModel [[flaxwhispermodel]]
-
-[[autodoc]] FlaxWhisperModel
-    - __call__
-
-## FlaxWhisperForConditionalGeneration [[flaxwhisperforconditionalgeneration]]
-
-[[autodoc]] FlaxWhisperForConditionalGeneration
-    - __call__
-
-## FlaxWhisperForAudioClassification [[flaxwhisperforaudioclassification]]
-
-[[autodoc]] FlaxWhisperForAudioClassification
-    - __call__
-

docs/source/ms/index.md

@@ -76,7 +76,6 @@ Dokumentasi disusun kepada lima bahagian:
 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. **[Bros](model_doc/bros)** (from NAVER) 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](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.
@@ -289,7 +288,6 @@ Flax), PyTorch, dan/atau TensorFlow.
 |            BLIP-2             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |             BLOOM             |       ❌       |       ✅       |       ✅        |         ❌         |      ❌      |
 |          BridgeTower          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|             Bros              |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 |           CamemBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |            CANINE             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |         Chinese-CLIP          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/pt/tasks/sequence_classification.md

@@ -70,7 +70,7 @@ Crie uma função de pré-processamento para tokenizar o campo `text` e truncar
 ...     return tokenizer(examples["text"], truncation=True)
 ```
 
-Use a função [`map`](https://huggingface.co/docs/datasets/process#map) do 🤗 Datasets para aplicar a função de pré-processamento em todo o conjunto de dados. Você pode acelerar a função `map` definindo `batched=True` para processar vários elementos do conjunto de dados de uma só vez:
+Use a função [`map`](https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map) do 🤗 Datasets para aplicar a função de pré-processamento em todo o conjunto de dados. Você pode acelerar a função `map` definindo `batched=True` para processar vários elementos do conjunto de dados de uma só vez:
 
 ```py
 tokenized_imdb = imdb.map(preprocess_function, batched=True)

docs/source/pt/tasks/token_classification.md

@@ -128,7 +128,7 @@ Aqui está como você pode criar uma função para realinhar os tokens e rótulo
 ...     return tokenized_inputs
 ```
 
-Use a função [`map`](https://huggingface.co/docs/datasets/process#map) do 🤗 Datasets para tokenizar e alinhar os rótulos em todo o conjunto de dados. Você pode acelerar a função `map` configurando `batched=True` para processar vários elementos do conjunto de dados de uma só vez:
+Use a função [`map`](https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map) do 🤗 Datasets para tokenizar e alinhar os rótulos em todo o conjunto de dados. Você pode acelerar a função `map` configurando `batched=True` para processar vários elementos do conjunto de dados de uma só vez:
 
 ```py
 >>> tokenized_wnut = wnut.map(tokenize_and_align_labels, batched=True)

examples/flax/language-modeling/run_bart_dlm_flax.py

@@ -139,7 +139,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -684,7 +684,7 @@ def main():
     # might be slower to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
     tokenized_datasets = tokenized_datasets.map(
         group_texts,
         batched=True,

examples/flax/language-modeling/run_clm_flax.py

@@ -140,7 +140,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -574,9 +574,9 @@ def main():
         if block_size > config.max_position_embeddings:
             logger.warning(
                 f"The tokenizer picked seems to have a very large `model_max_length` ({tokenizer.model_max_length}). "
-                f"Using block_size={min(1024, config.max_position_embeddings)} instead. You can change that default value by passing --block_size xxx."
+                "Picking 1024 instead. You can change that default value by passing --block_size xxx."
             )
-            block_size = min(1024, config.max_position_embeddings)
+            block_size = 1024
     else:
         if data_args.block_size > tokenizer.model_max_length:
             logger.warning(
@@ -607,7 +607,7 @@ def main():
     # to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
     lm_datasets = tokenized_datasets.map(
         group_texts,

examples/flax/language-modeling/run_mlm_flax.py

@@ -145,7 +145,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -625,7 +625,7 @@ def main():
         # might be slower to preprocess.
         #
         # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-        # https://huggingface.co/docs/datasets/process#map
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
         tokenized_datasets = tokenized_datasets.map(
             group_texts,
             batched=True,

examples/flax/language-modeling/run_t5_mlm_flax.py

@@ -139,7 +139,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -715,7 +715,7 @@ def main():
     # might be slower to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
     tokenized_datasets = tokenized_datasets.map(
         group_texts,
         batched=True,

examples/flax/summarization/run_summarization_flax.py

@@ -159,7 +159,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )

examples/flax/vision/run_image_classification.py

@@ -137,7 +137,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -477,8 +477,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_step
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -511,7 +511,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/image-pretraining/run_mae.py

@@ -110,7 +110,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )

examples/pytorch/image-pretraining/run_mim_no_trainer.py

@@ -701,8 +701,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_steps
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -735,7 +735,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/language-modeling/run_clm.py

@@ -77,7 +77,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -533,7 +533,7 @@ def main():
     # to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
     with training_args.main_process_first(desc="grouping texts together"):
         if not data_args.streaming:

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -473,7 +473,7 @@ def main():
     # to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
     with accelerator.main_process_first():
         lm_datasets = tokenized_datasets.map(
@@ -632,7 +632,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/language-modeling/run_mlm.py

@@ -547,7 +547,7 @@ def main():
         # might be slower to preprocess.
         #
         # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-        # https://huggingface.co/docs/datasets/process#map
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
         with training_args.main_process_first(desc="grouping texts together"):
             if not data_args.streaming:

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -504,7 +504,7 @@ def main():
         # might be slower to preprocess.
         #
         # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-        # https://huggingface.co/docs/datasets/process#map
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
         with accelerator.main_process_first():
             tokenized_datasets = tokenized_datasets.map(
@@ -636,8 +636,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_steps
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -670,7 +670,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/language-modeling/run_plm.py

@@ -65,7 +65,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -478,7 +478,7 @@ def main():
         # might be slower to preprocess.
         #
         # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-        # https://huggingface.co/docs/datasets/process#map
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
         with training_args.main_process_first(desc="grouping texts together"):
             tokenized_datasets = tokenized_datasets.map(

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -583,8 +583,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -617,7 +617,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -820,8 +820,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -848,11 +848,10 @@ def main():
             resume_step = None
             completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            # need to multiply `gradient_accumulation_steps` to reflect real steps
-            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
+            resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -886,7 +885,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -581,8 +581,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -615,7 +615,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -652,8 +652,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -686,7 +686,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -530,8 +530,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_step
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -562,7 +562,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -690,8 +690,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -722,7 +722,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/pytorch/translation/run_translation_no_trainer.py

@@ -633,8 +633,8 @@ def main():
             # need to multiply `gradient_accumulation_steps` to reflect real steps
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
-            completed_steps = resume_step // args.gradient_accumulation_steps
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)
@@ -665,7 +665,7 @@ def main():
 
             if isinstance(checkpointing_steps, int):
                 if completed_steps % checkpointing_steps == 0:
-                    output_dir = f"step_{completed_steps}"
+                    output_dir = f"step_{completed_steps }"
                     if args.output_dir is not None:
                         output_dir = os.path.join(args.output_dir, output_dir)
                     accelerator.save_state(output_dir)

examples/research_projects/jax-projects/dataset-streaming/run_mlm_flax_stream.py

@@ -76,7 +76,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )

examples/research_projects/jax-projects/model_parallel/run_clm_mp.py

@@ -70,7 +70,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -395,7 +395,7 @@ def main():
     # to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
     lm_datasets = tokenized_datasets.map(
         group_texts,

examples/research_projects/mlm_wwm/run_mlm_wwm.py

@@ -62,7 +62,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )

examples/research_projects/performer/run_mlm_performer.py

@@ -99,7 +99,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )

examples/tensorflow/README.md

@@ -15,7 +15,7 @@ limitations under the License.
 
 # Examples
 
-This folder contains actively maintained examples of the use of 🤗 Transformers organized into different ML tasks. All examples in this folder are **TensorFlow** examples and are written using native Keras rather than classes like `TFTrainer`, which we now consider deprecated. If you've previously only used 🤗 Transformers via `TFTrainer`, we highly recommend taking a look at the new style - we think it's a big improvement!
+This folder contains actively maintained examples of use of 🤗 Transformers organized into different ML tasks. All examples in this folder are **TensorFlow** examples, and are written using native Keras rather than classes like `TFTrainer`, which we now consider deprecated. If you've previously only used 🤗 Transformers via `TFTrainer`, we highly recommend taking a look at the new style - we think it's a big improvement!
 
 In addition, all scripts here now support the [🤗 Datasets](https://github.com/huggingface/datasets) library - you can grab entire datasets just by changing one command-line argument!
 

examples/tensorflow/language-modeling/run_clm.py

@@ -78,7 +78,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -459,7 +459,7 @@ def main():
     # to preprocess.
     #
     # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-    # https://huggingface.co/docs/datasets/process#map
+    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
     lm_datasets = tokenized_datasets.map(
         group_texts,

examples/tensorflow/language-modeling/run_mlm.py

@@ -76,7 +76,7 @@ class ModelArguments:
         default=None,
         metadata={
             "help": (
-                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
+                "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch."
             )
         },
     )
@@ -474,7 +474,7 @@ def main():
         # might be slower to preprocess.
         #
         # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
-        # https://huggingface.co/docs/datasets/process#map
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
 
         tokenized_datasets = tokenized_datasets.map(
             group_texts,

setup.py

@@ -118,7 +118,7 @@ _deps = [
     "fugashi>=1.0",
     "GitPython<3.1.19",
     "hf-doc-builder>=0.3.0",
-    "huggingface-hub>=0.16.4,<1.0",
+    "huggingface-hub>=0.15.1,<1.0",
     "importlib_metadata",
     "ipadic>=1.0.0,<2.0",
     "isort>=5.5.4",
@@ -172,7 +172,7 @@ _deps = [
     "tf2onnx",
     "timeout-decorator",
     "timm",
-    "tokenizers>=0.14,<0.15",
+    "tokenizers>=0.11.1,!=0.11.3,<0.14",
     "torch>=1.10,!=1.12.0",
     "torchaudio",
     "torchvision",

src/transformers/__init__.py

@@ -221,7 +221,6 @@ _import_structure = {
         "BridgeTowerTextConfig",
         "BridgeTowerVisionConfig",
     ],
-    "models.bros": ["BROS_PRETRAINED_CONFIG_ARCHIVE_MAP", "BrosConfig", "BrosProcessor"],
     "models.byt5": ["ByT5Tokenizer"],
     "models.camembert": ["CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "CamembertConfig"],
     "models.canine": ["CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP", "CanineConfig", "CanineTokenizer"],
@@ -470,7 +469,6 @@ _import_structure = {
     "models.pegasus": ["PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP", "PegasusConfig", "PegasusTokenizer"],
     "models.pegasus_x": ["PEGASUS_X_PRETRAINED_CONFIG_ARCHIVE_MAP", "PegasusXConfig"],
     "models.perceiver": ["PERCEIVER_PRETRAINED_CONFIG_ARCHIVE_MAP", "PerceiverConfig", "PerceiverTokenizer"],
-    "models.persimmon": ["PERSIMMON_PRETRAINED_CONFIG_ARCHIVE_MAP", "PersimmonConfig"],
     "models.phobert": ["PhobertTokenizer"],
     "models.pix2struct": [
         "PIX2STRUCT_PRETRAINED_CONFIG_ARCHIVE_MAP",
@@ -589,7 +587,6 @@ _import_structure = {
     "models.vit_mae": ["VIT_MAE_PRETRAINED_CONFIG_ARCHIVE_MAP", "ViTMAEConfig"],
     "models.vit_msn": ["VIT_MSN_PRETRAINED_CONFIG_ARCHIVE_MAP", "ViTMSNConfig"],
     "models.vitdet": ["VITDET_PRETRAINED_CONFIG_ARCHIVE_MAP", "VitDetConfig"],
-    "models.vitmatte": ["VITMATTE_PRETRAINED_CONFIG_ARCHIVE_MAP", "VitMatteConfig"],
     "models.vits": [
         "VITS_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "VitsConfig",
@@ -986,7 +983,6 @@ else:
     _import_structure["models.vilt"].extend(["ViltFeatureExtractor", "ViltImageProcessor", "ViltProcessor"])
     _import_structure["models.vit"].extend(["ViTFeatureExtractor", "ViTImageProcessor"])
     _import_structure["models.vit_hybrid"].extend(["ViTHybridImageProcessor"])
-    _import_structure["models.vitmatte"].append("VitMatteImageProcessor")
     _import_structure["models.vivit"].append("VivitImageProcessor")
     _import_structure["models.yolos"].extend(["YolosFeatureExtractor", "YolosImageProcessor"])
 
@@ -1358,17 +1354,6 @@ else:
             "BridgeTowerPreTrainedModel",
         ]
     )
-    _import_structure["models.bros"].extend(
-        [
-            "BROS_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "BrosForTokenClassification",
-            "BrosModel",
-            "BrosPreTrainedModel",
-            "BrosProcessor",
-            "BrosSpadeEEForTokenClassification",
-            "BrosSpadeELForTokenClassification",
-        ]
-    )
     _import_structure["models.camembert"].extend(
         [
             "CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -2447,9 +2432,6 @@ else:
             "PerceiverPreTrainedModel",
         ]
     )
-    _import_structure["models.persimmon"].extend(
-        ["PersimmonForCausalLM", "PersimmonForSequenceClassification", "PersimmonModel", "PersimmonPreTrainedModel"]
-    )
     _import_structure["models.pix2struct"].extend(
         [
             "PIX2STRUCT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -2959,13 +2941,6 @@ else:
             "VitDetPreTrainedModel",
         ]
     )
-    _import_structure["models.vitmatte"].extend(
-        [
-            "VITMATTE_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "VitMatteForImageMatting",
-            "VitMattePreTrainedModel",
-        ]
-    )
     _import_structure["models.vits"].extend(
         [
             "VITS_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -4345,7 +4320,6 @@ if TYPE_CHECKING:
         BridgeTowerTextConfig,
         BridgeTowerVisionConfig,
     )
-    from .models.bros import BROS_PRETRAINED_CONFIG_ARCHIVE_MAP, BrosConfig, BrosProcessor
     from .models.byt5 import ByT5Tokenizer
     from .models.camembert import CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CamembertConfig
     from .models.canine import CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP, CanineConfig, CanineTokenizer
@@ -4577,7 +4551,6 @@ if TYPE_CHECKING:
     from .models.pegasus import PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP, PegasusConfig, PegasusTokenizer
     from .models.pegasus_x import PEGASUS_X_PRETRAINED_CONFIG_ARCHIVE_MAP, PegasusXConfig
     from .models.perceiver import PERCEIVER_PRETRAINED_CONFIG_ARCHIVE_MAP, PerceiverConfig, PerceiverTokenizer
-    from .models.persimmon import PERSIMMON_PRETRAINED_CONFIG_ARCHIVE_MAP, PersimmonConfig
     from .models.phobert import PhobertTokenizer
     from .models.pix2struct import (
         PIX2STRUCT_PRETRAINED_CONFIG_ARCHIVE_MAP,
@@ -4684,7 +4657,6 @@ if TYPE_CHECKING:
     from .models.vit_mae import VIT_MAE_PRETRAINED_CONFIG_ARCHIVE_MAP, ViTMAEConfig
     from .models.vit_msn import VIT_MSN_PRETRAINED_CONFIG_ARCHIVE_MAP, ViTMSNConfig
     from .models.vitdet import VITDET_PRETRAINED_CONFIG_ARCHIVE_MAP, VitDetConfig
-    from .models.vitmatte import VITMATTE_PRETRAINED_CONFIG_ARCHIVE_MAP, VitMatteConfig
     from .models.vits import (
         VITS_PRETRAINED_CONFIG_ARCHIVE_MAP,
         VitsConfig,
@@ -5039,7 +5011,6 @@ if TYPE_CHECKING:
         from .models.vilt import ViltFeatureExtractor, ViltImageProcessor, ViltProcessor
         from .models.vit import ViTFeatureExtractor, ViTImageProcessor
         from .models.vit_hybrid import ViTHybridImageProcessor
-        from .models.vitmatte import VitMatteImageProcessor
         from .models.vivit import VivitImageProcessor
         from .models.yolos import YolosFeatureExtractor, YolosImageProcessor
 
@@ -5362,15 +5333,6 @@ if TYPE_CHECKING:
             BridgeTowerModel,
             BridgeTowerPreTrainedModel,
         )
-        from .models.bros import (
-            BROS_PRETRAINED_MODEL_ARCHIVE_LIST,
-            BrosForTokenClassification,
-            BrosModel,
-            BrosPreTrainedModel,
-            BrosProcessor,
-            BrosSpadeEEForTokenClassification,
-            BrosSpadeELForTokenClassification,
-        )
         from .models.camembert import (
             CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             CamembertForCausalLM,
@@ -6249,12 +6211,6 @@ if TYPE_CHECKING:
             PerceiverModel,
             PerceiverPreTrainedModel,
         )
-        from .models.persimmon import (
-            PersimmonForCausalLM,
-            PersimmonForSequenceClassification,
-            PersimmonModel,
-            PersimmonPreTrainedModel,
-        )
         from .models.pix2struct import (
             PIX2STRUCT_PRETRAINED_MODEL_ARCHIVE_LIST,
             Pix2StructForConditionalGeneration,
@@ -6659,11 +6615,6 @@ if TYPE_CHECKING:
             VitDetModel,
             VitDetPreTrainedModel,
         )
-        from .models.vitmatte import (
-            VITMATTE_PRETRAINED_MODEL_ARCHIVE_LIST,
-            VitMatteForImageMatting,
-            VitMattePreTrainedModel,
-        )
         from .models.vits import (
             VITS_PRETRAINED_MODEL_ARCHIVE_LIST,
             VitsModel,

src/transformers/audio_utils.py

@@ -30,19 +30,17 @@ def hertz_to_mel(freq: Union[float, np.ndarray], mel_scale: str = "htk") -> Unio
         freq (`float` or `np.ndarray`):
             The frequency, or multiple frequencies, in hertz (Hz).
         mel_scale (`str`, *optional*, defaults to `"htk"`):
-            The mel frequency scale to use, `"htk"`, `"kaldi"` or `"slaney"`.
+            The mel frequency scale to use, `"htk"` or `"slaney"`.
 
     Returns:
         `float` or `np.ndarray`: The frequencies on the mel scale.
     """
 
-    if mel_scale not in ["slaney", "htk", "kaldi"]:
-        raise ValueError('mel_scale should be one of "htk", "slaney" or "kaldi".')
+    if mel_scale not in ["slaney", "htk"]:
+        raise ValueError('mel_scale should be one of "htk" or "slaney".')
 
     if mel_scale == "htk":
         return 2595.0 * np.log10(1.0 + (freq / 700.0))
-    elif mel_scale == "kaldi":
-        return 1127.0 * np.log(1.0 + (freq / 700.0))
 
     min_log_hertz = 1000.0
     min_log_mel = 15.0
@@ -66,19 +64,17 @@ def mel_to_hertz(mels: Union[float, np.ndarray], mel_scale: str = "htk") -> Unio
         mels (`float` or `np.ndarray`):
             The frequency, or multiple frequencies, in mels.
         mel_scale (`str`, *optional*, `"htk"`):
-            The mel frequency scale to use, `"htk"`, `"kaldi"` or `"slaney"`.
+            The mel frequency scale to use, `"htk"` or `"slaney"`.
 
     Returns:
         `float` or `np.ndarray`: The frequencies in hertz.
     """
 
-    if mel_scale not in ["slaney", "htk", "kaldi"]:
-        raise ValueError('mel_scale should be one of "htk", "slaney" or "kaldi".')
+    if mel_scale not in ["slaney", "htk"]:
+        raise ValueError('mel_scale should be one of "htk" or "slaney".')
 
     if mel_scale == "htk":
-        return 700.0 * (np.power(10, mels / 2595.0) - 1.0)
-    elif mel_scale == "kaldi":
-        return 700.0 * (np.exp(mels / 1127.0) - 1.0)
+        return 700.0 * (10.0 ** (mels / 2595.0) - 1.0)
 
     min_log_hertz = 1000.0
     min_log_mel = 15.0
@@ -124,7 +120,6 @@ def mel_filter_bank(
     sampling_rate: int,
     norm: Optional[str] = None,
     mel_scale: str = "htk",
-    triangularize_in_mel_space: bool = False,
 ) -> np.ndarray:
     """
     Creates a frequency bin conversion matrix used to obtain a mel spectrogram. This is called a *mel filter bank*, and
@@ -160,10 +155,7 @@ def mel_filter_bank(
         norm (`str`, *optional*):
             If `"slaney"`, divide the triangular mel weights by the width of the mel band (area normalization).
         mel_scale (`str`, *optional*, defaults to `"htk"`):
-            The mel frequency scale to use, `"htk"`, `"kaldi"` or `"slaney"`.
-        triangularize_in_mel_space (`bool`, *optional*, defaults to `False`):
-            If this option is enabled, the triangular filter is applied in mel space rather than frequency space. This
-            should be set to `true` in order to get the same results as `torchaudio` when computing mel filters.
+            The mel frequency scale to use, `"htk"` or `"slaney"`.
 
     Returns:
         `np.ndarray` of shape (`num_frequency_bins`, `num_mel_filters`): Triangular filter bank matrix. This is a
@@ -172,21 +164,15 @@ def mel_filter_bank(
     if norm is not None and norm != "slaney":
         raise ValueError('norm must be one of None or "slaney"')
 
+    # frequencies of FFT bins in Hz
+    fft_freqs = np.linspace(0, sampling_rate // 2, num_frequency_bins)
+
     # center points of the triangular mel filters
     mel_min = hertz_to_mel(min_frequency, mel_scale=mel_scale)
     mel_max = hertz_to_mel(max_frequency, mel_scale=mel_scale)
     mel_freqs = np.linspace(mel_min, mel_max, num_mel_filters + 2)
     filter_freqs = mel_to_hertz(mel_freqs, mel_scale=mel_scale)
 
-    if triangularize_in_mel_space:
-        # frequencies of FFT bins in Hz, but filters triangularized in mel space
-        fft_bin_width = sampling_rate / (num_frequency_bins * 2)
-        fft_freqs = hertz_to_mel(fft_bin_width * np.arange(num_frequency_bins), mel_scale=mel_scale)
-        filter_freqs = mel_freqs
-    else:
-        # frequencies of FFT bins in Hz
-        fft_freqs = np.linspace(0, sampling_rate // 2, num_frequency_bins)
-
     mel_filters = _create_triangular_filter_bank(fft_freqs, filter_freqs)
 
     if norm is not None and norm == "slaney":
@@ -232,7 +218,6 @@ def window_function(
         - `"boxcar"`: a rectangular window
         - `"hamming"`: the Hamming window
         - `"hann"`: the Hann window
-        - `"povey"`: the Povey window
 
     Args:
         window_length (`int`):
@@ -258,8 +243,6 @@ def window_function(
         window = np.hamming(length)
     elif name in ["hann", "hann_window"]:
         window = np.hanning(length)
-    elif name in ["povey"]:
-        window = np.power(np.hanning(length), 0.85)
     else:
         raise ValueError(f"Unknown window function '{name}'")
 
@@ -298,7 +281,6 @@ def spectrogram(
     reference: float = 1.0,
     min_value: float = 1e-10,
     db_range: Optional[float] = None,
-    remove_dc_offset: Optional[bool] = None,
     dtype: np.dtype = np.float32,
 ) -> np.ndarray:
     """
@@ -381,9 +363,6 @@ def spectrogram(
         db_range (`float`, *optional*):
             Sets the maximum dynamic range in decibels. For example, if `db_range = 80`, the difference between the
             peak value and the smallest value will never be more than 80 dB. Must be greater than zero.
-        remove_dc_offset (`bool`, *optional*):
-            Subtract mean from waveform on each frame, applied before pre-emphasis. This should be set to `true` in
-            order to get the same results as `torchaudio.compliance.kaldi.fbank` when computing mel filters.
         dtype (`np.dtype`, *optional*, defaults to `np.float32`):
             Data type of the spectrogram tensor. If `power` is None, this argument is ignored and the dtype will be
             `np.complex64`.
@@ -435,9 +414,6 @@ def spectrogram(
     for frame_idx in range(num_frames):
         buffer[:frame_length] = waveform[timestep : timestep + frame_length]
 
-        if remove_dc_offset:
-            buffer[:frame_length] = buffer[:frame_length] - buffer[:frame_length].mean()
-
         if preemphasis is not None:
             buffer[1:frame_length] -= preemphasis * buffer[: frame_length - 1]
             buffer[0] *= 1 - preemphasis

src/transformers/convert_slow_tokenizer.py

@@ -1153,13 +1153,7 @@ class LlamaConverter(SpmConverter):
         model_type = proto.trainer_spec.model_type
         vocab_scores = self.vocab(proto)
         if model_type == 1:
-            import tokenizers
-
-            if version.parse(tokenizers.__version__) < version.parse("0.14.0"):
-                tokenizer = Tokenizer(Unigram(vocab_scores, 0))
-            else:
-                tokenizer = Tokenizer(Unigram(vocab_scores, 0, byte_fallback=True))
-
+            raise RuntimeError("Llama is supposed to be a BPE model!")
         elif model_type == 2:
             _, merges = SentencePieceExtractor(self.original_tokenizer.vocab_file).extract(vocab_scores)
             bpe_vocab = {word: i for i, (word, _score) in enumerate(vocab_scores)}
@@ -1206,8 +1200,8 @@ class LlamaConverter(SpmConverter):
             eos = self.original_tokenizer.eos_token
             eos_token_id = self.original_tokenizer.eos_token_id
 
-            single = f"{(bos+':0 ') * add_bos}$A:0{(' '+eos+':0') if add_eos else ''}"
-            pair = f"{single}{(' '+bos+':1') * add_bos} $B:1{(' '+eos+':1') if add_eos else ''}"
+            single = f"{(bos+':0 ') * add_bos}$A:0{(' '+eos+':0') * add_eos}"
+            pair = f"{single}{(' '+bos+':1') * add_bos} $B:1{(' '+eos+':1') * add_eos}"
 
             special_tokens = []
             if add_bos:

src/transformers/dependency_versions_table.py

@@ -25,7 +25,7 @@ deps = {
     "fugashi": "fugashi>=1.0",
     "GitPython": "GitPython<3.1.19",
     "hf-doc-builder": "hf-doc-builder>=0.3.0",
-    "huggingface-hub": "huggingface-hub>=0.16.4,<1.0",
+    "huggingface-hub": "huggingface-hub>=0.15.1,<1.0",
     "importlib_metadata": "importlib_metadata",
     "ipadic": "ipadic>=1.0.0,<2.0",
     "isort": "isort>=5.5.4",
@@ -78,7 +78,7 @@ deps = {
     "tf2onnx": "tf2onnx",
     "timeout-decorator": "timeout-decorator",
     "timm": "timm",
-    "tokenizers": "tokenizers>=0.14,<0.15",
+    "tokenizers": "tokenizers>=0.11.1,!=0.11.3,<0.14",
     "torch": "torch>=1.10,!=1.12.0",
     "torchaudio": "torchaudio",
     "torchvision": "torchvision",

src/transformers/dynamic_module_utils.py

@@ -575,8 +575,9 @@ def custom_object_save(obj: Any, folder: Union[str, os.PathLike], config: Option
 
 def _raise_timeout_error(signum, frame):
     raise ValueError(
-        "Loading this model requires you to execute custom code contained in the model repository on your local"
-        "machine. Please set the option `trust_remote_code=True` to permit loading of this model."
+        "Loading this model requires you to execute the configuration file in that repo on your local machine. We "
+        "asked if it was okay but did not get an answer. Make sure you have read the code there to avoid malicious "
+        "use, then set the option `trust_remote_code=True` to remove this error."
     )
 
 
@@ -593,10 +594,9 @@ def resolve_trust_remote_code(trust_remote_code, model_name, has_local_code, has
                 signal.alarm(TIME_OUT_REMOTE_CODE)
                 while trust_remote_code is None:
                     answer = input(
-                        f"The repository for {model_name} contains custom code which must be executed to correctly"
-                        f"load the model. You can inspect the repository content at https://hf.co/{model_name}.\n"
-                        f"You can avoid this prompt in future by passing the argument `trust_remote_code=True`.\n\n"
-                        f"Do you wish to run the custom code? [y/N] "
+                        f"Loading {model_name} requires to execute some code in that repo, you can inspect the content of "
+                        f"the repository at https://hf.co/{model_name}. You can dismiss this prompt by passing "
+                        "`trust_remote_code=True`.\nDo you accept? [y/N] "
                     )
                     if answer.lower() in ["yes", "y", "1"]:
                         trust_remote_code = True
@@ -606,9 +606,9 @@ def resolve_trust_remote_code(trust_remote_code, model_name, has_local_code, has
             except Exception:
                 # OS which does not support signal.SIGALRM
                 raise ValueError(
-                    f"The repository for {model_name} contains custom code which must be executed to correctly"
-                    f"load the model. You can inspect the repository content at https://hf.co/{model_name}.\n"
-                    f"Please pass the argument `trust_remote_code=True` to allow custom code to be run."
+                    "Loading this model requires you to execute execute some code in that repo on your local machine. "
+                    f"Make sure you have read the code at https://hf.co/{model_name} to avoid malicious use, then set "
+                    "the option `trust_remote_code=True` to remove this error."
                 )
         elif has_remote_code:
             # For the CI which puts the timeout at 0

src/transformers/generation/configuration_utils.py

@@ -34,7 +34,6 @@ from ..utils import (
 
 
 logger = logging.get_logger(__name__)
-METADATA_FIELDS = ("_from_model_config", "_commit_hash", "_original_object_hash", "transformers_version")
 
 
 class GenerationConfig(PushToHubMixin):
@@ -316,19 +315,20 @@ class GenerationConfig(PushToHubMixin):
         # Validate the values of the attributes
         self.validate(is_init=True)
 
-    def __hash__(self):
-        return hash(self.to_json_string(ignore_metadata=True))
-
     def __eq__(self, other):
         if not isinstance(other, GenerationConfig):
             return False
 
-        self_without_metadata = self.to_json_string(use_diff=False, ignore_metadata=True)
-        other_without_metadata = other.to_json_string(use_diff=False, ignore_metadata=True)
-        return self_without_metadata == other_without_metadata
+        self_dict = self.__dict__.copy()
+        other_dict = other.__dict__.copy()
+        # ignore metadata
+        for metadata_field in ("_from_model_config", "_commit_hash", "transformers_version"):
+            self_dict.pop(metadata_field, None)
+            other_dict.pop(metadata_field, None)
+        return self_dict == other_dict
 
     def __repr__(self):
-        return f"{self.__class__.__name__} {self.to_json_string(ignore_metadata=True)}"
+        return f"{self.__class__.__name__} {self.to_json_string()}"
 
     def validate(self, is_init=False):
         """
@@ -729,9 +729,7 @@ class GenerationConfig(PushToHubMixin):
         else:
             logger.info(f"loading configuration file {configuration_file} from cache at {resolved_config_file}")
 
-        config = cls.from_dict(config_dict, **kwargs)
-        config._original_object_hash = hash(config)  # Hash to detect whether the instance was modified
-        return config
+        return cls.from_dict(config_dict, **kwargs)
 
     @classmethod
     def _dict_from_json_file(cls, json_file: Union[str, os.PathLike]):
@@ -816,12 +814,8 @@ class GenerationConfig(PushToHubMixin):
             `Dict[str, Any]`: Dictionary of all the attributes that make up this configuration instance.
         """
         output = copy.deepcopy(self.__dict__)
-
-        # Fields to ignore at serialization time
         if "_commit_hash" in output:
             del output["_commit_hash"]
-        if "_original_object_hash" in output:
-            del output["_original_object_hash"]
 
         # Transformers version when serializing this file
         output["transformers_version"] = __version__
@@ -829,7 +823,7 @@ class GenerationConfig(PushToHubMixin):
         self.dict_torch_dtype_to_str(output)
         return output
 
-    def to_json_string(self, use_diff: bool = True, ignore_metadata: bool = False) -> str:
+    def to_json_string(self, use_diff: bool = True) -> str:
         """
         Serializes this instance to a JSON string.
 
@@ -837,8 +831,6 @@ class GenerationConfig(PushToHubMixin):
             use_diff (`bool`, *optional*, defaults to `True`):
                 If set to `True`, only the difference between the config instance and the default `GenerationConfig()`
                 is serialized to JSON string.
-            ignore_metadata (`bool`, *optional*, defaults to `False`):
-                Whether to ignore the metadata fields present in the instance
 
         Returns:
             `str`: String containing all the attributes that make up this configuration instance in JSON format.
@@ -847,11 +839,6 @@ class GenerationConfig(PushToHubMixin):
             config_dict = self.to_diff_dict()
         else:
             config_dict = self.to_dict()
-
-        if ignore_metadata:
-            for metadata_field in METADATA_FIELDS:
-                config_dict.pop(metadata_field, None)
-
         return json.dumps(config_dict, indent=2, sort_keys=True) + "\n"
 
     def to_json_file(self, json_file_path: Union[str, os.PathLike], use_diff: bool = True):
@@ -895,7 +882,6 @@ class GenerationConfig(PushToHubMixin):
                     if attr in decoder_config and getattr(config, attr) == getattr(default_generation_config, attr):
                         setattr(config, attr, decoder_config[attr])
 
-        config._original_object_hash = hash(config)  # Hash to detect whether the instance was modified
         return config
 
     def update(self, **kwargs):

src/transformers/generation/flax_utils.py

@@ -310,20 +310,16 @@ class FlaxGenerationMixin:
 
         # priority: `generation_config` argument > `model.generation_config` (the default generation config)
         if generation_config is None:
-            # legacy: users may modify the model configuration to control generation. To trigger this legacy behavior,
-            # two conditions must be met
-            # 1) the generation config must have been created from the model config (`_from_model_config` field);
-            # 2) the generation config must have seen no modification since its creation (the hash is the same).
-            if self.generation_config._from_model_config and self.generation_config._original_object_hash == hash(
-                self.generation_config
-            ):
+            # legacy: users may modify the model configuration to control generation -- update the generation config
+            # model attribute accordingly, if it was created from the model config
+            if self.generation_config._from_model_config:
                 new_generation_config = GenerationConfig.from_model_config(self.config)
                 if new_generation_config != self.generation_config:
                     warnings.warn(
                         "You have modified the pretrained model configuration to control generation. This is a"
                         " deprecated strategy to control generation and will be removed soon, in a future version."
-                        " Please use and modify the model generation configuration (see"
-                        " https://huggingface.co/docs/transformers/generation_strategies#default-text-generation-configuration )"
+                        " Please use a generation configuration file (see"
+                        " https://huggingface.co/docs/transformers/main_classes/text_generation )"
                     )
                     self.generation_config = new_generation_config
             generation_config = self.generation_config
@@ -389,7 +385,7 @@ class FlaxGenerationMixin:
                 UserWarning,
             )
         elif generation_config.max_new_tokens is not None:
-            if not has_default_max_length and generation_config.max_length is not None:
+            if not has_default_max_length:
                 logger.warning(
                     f"Both `max_new_tokens` (={generation_config.max_new_tokens}) and `max_length`(="
                     f"{generation_config.max_length}) seem to have been set. `max_new_tokens` will take precedence. "

src/transformers/generation/logits_process.py

@@ -272,7 +272,7 @@ class RepetitionPenaltyLogitsProcessor(LogitsProcessor):
     [`LogitsProcessor`] that prevents the repetition of previous tokens through an exponential penalty. This technique
     shares some similarities with coverage mechanisms and other aimed at reducing repetition. During the text
     generation process, the probability distribution for the next token is determined using a formula that incorporates
-    token scores based on their occurrence in the generated sequence. Tokens with higher scores are more likely to be
+    token scores based on their occurrence in the generated sequence. Tokens with higher scores are less likely to be
     selected. The formula can be seen in the original [paper](https://arxiv.org/pdf/1909.05858.pdf). According to the
     paper a penalty of around 1.2 yields a good balance between truthful generation and lack of repetition.
 
@@ -328,7 +328,7 @@ class EncoderRepetitionPenaltyLogitsProcessor(LogitsProcessor):
         hallucination_penalty (`float`):
             The parameter for hallucination penalty. 1.0 means no penalty.
         encoder_input_ids (`torch.LongTensor`):
-            The encoder_input_ids that should be repeated within the decoder ids.
+            The encoder_input_ids that should not be repeated within the decoder ids.
     """
 
     def __init__(self, penalty: float, encoder_input_ids: torch.LongTensor):
@@ -1297,9 +1297,8 @@ class InfNanRemoveLogitsProcessor(LogitsProcessor):
 
 class ExponentialDecayLengthPenalty(LogitsProcessor):
     r"""
-    [`LogitsProcessor`] that exponentially increases the score of the `eos_token_id` after `start_index` has been
-    reached. This allows generating shorter sequences without having a hard cutoff, allowing the `eos_token` to be
-    predicted in a meaningful position.
+    [`LogitsProcessor`] that exponentially increases the score of the eos_token_id after regulation_start has been
+    reached.
 
     Args:
         exponential_decay_length_penalty (`tuple(int, float)`):
@@ -1309,51 +1308,6 @@ class ExponentialDecayLengthPenalty(LogitsProcessor):
             The id of the *end-of-sequence* token. Optionally, use a list to set multiple *end-of-sequence* tokens.
         input_ids_seq_length (`int`):
             The length of the input sequence.
-
-    Examples:
-
-    ```python
-    >>> from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
-
-    >>> set_seed(1)
-    >>> model = AutoModelForCausalLM.from_pretrained("gpt2")
-    >>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
-
-    >>> text = "Just wanted to let you know, I"
-    >>> inputs = tokenizer(text, return_tensors="pt")
-
-    >>> # Generate sequences without exponential penalty. We want short sentences, so we limit max_length=30
-    >>> # see that the answer tends to end abruptly
-    >>> outputs = model.generate(**inputs, do_sample=True, temperature=0.9, max_length=30, pad_token_id=50256)
-    >>> print(tokenizer.batch_decode(outputs)[0])
-    Just wanted to let you know, I'm not even a lawyer. I'm a man. I have no real knowledge of politics. I'm a
-
-    >>> # Generate sequences with exponential penalty, we add the exponential_decay_length_penalty=(start_index, decay_factor)
-    >>> # We see that instead of cutting at max_tokens, the output comes to an end before (at 25 tokens) and with more meaning
-    >>> # What happens is that starting from `start_index` the EOS token score will be increased by decay_factor exponentially
-    >>> outputs = model.generate(
-    ...     **inputs,
-    ...     do_sample=True,
-    ...     temperature=0.9,
-    ...     max_length=30,
-    ...     pad_token_id=50256,
-    ...     exponential_decay_length_penalty=(15, 1.6),
-    ... )
-    >>> print(tokenizer.batch_decode(outputs)[0])
-    Just wanted to let you know, I've got a very cool t-shirt educating people on how to use the Internet<|endoftext|>
-
-    >>> # Generate sequences with smaller decay_factor, still improving the hard cutoff mid-sentence
-    >>> outputs = model.generate(
-    ...     **inputs,
-    ...     do_sample=True,
-    ...     temperature=0.9,
-    ...     max_length=30,
-    ...     pad_token_id=50256,
-    ...     exponential_decay_length_penalty=(15, 1.05),
-    ... )
-    >>> print(tokenizer.batch_decode(outputs)[0])
-    Just wanted to let you know, I've been working on it for about 6 months and now it's in Alpha.<|endoftext|>
-    ```
     """
 
     def __init__(
@@ -1373,9 +1327,7 @@ class ExponentialDecayLengthPenalty(LogitsProcessor):
         cur_len = input_ids.shape[-1]
         if cur_len > self.regulation_start:
             for i in self.eos_token_id:
-                penalty_idx = cur_len - self.regulation_start
-                # To support negative logits we compute the penalty of the absolute value and add to the original logit
-                scores[:, i] = scores[:, i] + torch.abs(scores[:, i]) * (pow(self.regulation_factor, penalty_idx) - 1)
+                scores[:, i] = scores[:, i] * pow(self.regulation_factor, cur_len - self.regulation_start)
         return scores
 
 

src/transformers/generation/tf_utils.py

@@ -716,20 +716,16 @@ class TFGenerationMixin:
 
         # priority: `generation_config` argument > `model.generation_config` (the default generation config)
         if generation_config is None:
-            # legacy: users may modify the model configuration to control generation. To trigger this legacy behavior,
-            # two conditions must be met
-            # 1) the generation config must have been created from the model config (`_from_model_config` field);
-            # 2) the generation config must have seen no modification since its creation (the hash is the same).
-            if self.generation_config._from_model_config and self.generation_config._original_object_hash == hash(
-                self.generation_config
-            ):
+            # legacy: users may modify the model configuration to control generation -- update the generation config
+            # model attribute accordingly, if it was created from the model config
+            if self.generation_config._from_model_config:
                 new_generation_config = GenerationConfig.from_model_config(self.config)
                 if new_generation_config != self.generation_config:
                     warnings.warn(
                         "You have modified the pretrained model configuration to control generation. This is a"
                         " deprecated strategy to control generation and will be removed soon, in a future version."
-                        " Please use and modify the model generation configuration (see"
-                        " https://huggingface.co/docs/transformers/generation_strategies#default-text-generation-configuration )"
+                        " Please use a generation configuration file (see"
+                        " https://huggingface.co/docs/transformers/main_classes/text_generation )"
                     )
                     self.generation_config = new_generation_config
             generation_config = self.generation_config
@@ -841,7 +837,7 @@ class TFGenerationMixin:
                 UserWarning,
             )
         elif generation_config.max_new_tokens is not None:
-            if not has_default_max_length and generation_config.max_length is not None:
+            if not has_default_max_length:
                 logger.warning(
                     f"Both `max_new_tokens` (={generation_config.max_new_tokens}) and `max_length`(="
                     f"{generation_config.max_length}) seem to have been set. `max_new_tokens` will take precedence. "

src/transformers/generation/utils.py

@@ -1258,7 +1258,7 @@ class GenerationMixin:
         if has_default_max_length and generation_config.max_new_tokens is None and generation_config.max_length == 20:
             # 20 is the default max_length of the generation config
             warnings.warn(
-                f"Using the model-agnostic default `max_length` (={generation_config.max_length}) to control the "
+                f"Using the model-agnostic default `max_length` (={generation_config.max_length}) to control the"
                 "generation length. We recommend setting `max_new_tokens` to control the maximum length of the "
                 "generation.",
                 UserWarning,
@@ -1409,20 +1409,16 @@ class GenerationMixin:
 
         # priority: `generation_config` argument > `model.generation_config` (the default generation config)
         if generation_config is None:
-            # legacy: users may modify the model configuration to control generation. To trigger this legacy behavior,
-            # two conditions must be met
-            # 1) the generation config must have been created from the model config (`_from_model_config` field);
-            # 2) the generation config must have seen no modification since its creation (the hash is the same).
-            if self.generation_config._from_model_config and self.generation_config._original_object_hash == hash(
-                self.generation_config
-            ):
+            # legacy: users may modify the model configuration to control generation -- update the generation config
+            # model attribute accordingly, if it was created from the model config
+            if self.generation_config._from_model_config:
                 new_generation_config = GenerationConfig.from_model_config(self.config)
                 if new_generation_config != self.generation_config:
                     warnings.warn(
                         "You have modified the pretrained model configuration to control generation. This is a"
                         " deprecated strategy to control generation and will be removed soon, in a future version."
-                        " Please use and modify the model generation configuration (see"
-                        " https://huggingface.co/docs/transformers/generation_strategies#default-text-generation-configuration )"
+                        " Please use a generation configuration file (see"
+                        " https://huggingface.co/docs/transformers/main_classes/text_generation )"
                     )
                     self.generation_config = new_generation_config
             generation_config = self.generation_config
@@ -1517,7 +1513,7 @@ class GenerationMixin:
         input_ids_length = input_ids.shape[-1]
         has_default_max_length = kwargs.get("max_length") is None and generation_config.max_length is not None
         if generation_config.max_new_tokens is not None:
-            if not has_default_max_length and generation_config.max_length is not None:
+            if not has_default_max_length:
                 logger.warning(
                     f"Both `max_new_tokens` (={generation_config.max_new_tokens}) and `max_length`(="
                     f"{generation_config.max_length}) seem to have been set. `max_new_tokens` will take precedence. "
@@ -3038,9 +3034,7 @@ class GenerationMixin:
             )  # (batch_size * num_beams, vocab_size)
 
             next_token_scores_processed = logits_processor(input_ids, next_token_scores)
-            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(
-                next_token_scores_processed
-            )
+            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(next_token_scores)
 
             # Store scores, attentions and hidden_states when required
             if return_dict_in_generate:
@@ -3365,9 +3359,7 @@ class GenerationMixin:
             )  # (batch_size * num_beams, vocab_size)
 
             next_token_scores_processed = logits_processor(input_ids, next_token_scores)
-            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(
-                next_token_scores_processed
-            )
+            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(next_token_scores)
             # Note: logits warpers are intentionally applied after adding running beam scores. On some logits warpers
             # (like top_p) this is indiferent, but on others (like temperature) it is not. For reference, see
             # https://github.com/huggingface/transformers/pull/5420#discussion_r449779867
@@ -4084,9 +4076,7 @@ class GenerationMixin:
 
             next_token_scores_processed = logits_processor(input_ids, next_token_scores)
 
-            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(
-                next_token_scores_processed
-            )
+            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(next_token_scores)
 
             scores_for_all_vocab = next_token_scores.clone()
 

src/transformers/integrations/__init__.py

@@ -11,132 +11,61 @@
 # 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.
-from typing import TYPE_CHECKING
-
-from ..utils import _LazyModule
-
-
-_import_structure = {
-    "bitsandbytes": [
-        "get_keys_to_not_convert",
-        "replace_8bit_linear",
-        "replace_with_bnb_linear",
-        "set_module_8bit_tensor_to_device",
-        "set_module_quantized_tensor_to_device",
-    ],
-    "deepspeed": [
-        "HfDeepSpeedConfig",
-        "HfTrainerDeepSpeedConfig",
-        "deepspeed_config",
-        "deepspeed_init",
-        "deepspeed_load_checkpoint",
-        "deepspeed_optim_sched",
-        "is_deepspeed_available",
-        "is_deepspeed_zero3_enabled",
-        "set_hf_deepspeed_config",
-        "unset_hf_deepspeed_config",
-    ],
-    "integration_utils": [
-        "INTEGRATION_TO_CALLBACK",
-        "AzureMLCallback",
-        "ClearMLCallback",
-        "CodeCarbonCallback",
-        "CometCallback",
-        "DagsHubCallback",
-        "FlyteCallback",
-        "MLflowCallback",
-        "NeptuneCallback",
-        "NeptuneMissingConfiguration",
-        "TensorBoardCallback",
-        "WandbCallback",
-        "get_available_reporting_integrations",
-        "get_reporting_integration_callbacks",
-        "hp_params",
-        "is_azureml_available",
-        "is_clearml_available",
-        "is_codecarbon_available",
-        "is_comet_available",
-        "is_dagshub_available",
-        "is_fairscale_available",
-        "is_flyte_deck_standard_available",
-        "is_flytekit_available",
-        "is_mlflow_available",
-        "is_neptune_available",
-        "is_optuna_available",
-        "is_ray_available",
-        "is_ray_tune_available",
-        "is_sigopt_available",
-        "is_tensorboard_available",
-        "is_wandb_available",
-        "rewrite_logs",
-        "run_hp_search_optuna",
-        "run_hp_search_ray",
-        "run_hp_search_sigopt",
-        "run_hp_search_wandb",
-    ],
-    "peft": ["PeftAdapterMixin"],
-}
-
-if TYPE_CHECKING:
-    from .bitsandbytes import (
-        get_keys_to_not_convert,
-        replace_8bit_linear,
-        replace_with_bnb_linear,
-        set_module_8bit_tensor_to_device,
-        set_module_quantized_tensor_to_device,
-    )
-    from .deepspeed import (
-        HfDeepSpeedConfig,
-        HfTrainerDeepSpeedConfig,
-        deepspeed_config,
-        deepspeed_init,
-        deepspeed_load_checkpoint,
-        deepspeed_optim_sched,
-        is_deepspeed_available,
-        is_deepspeed_zero3_enabled,
-        set_hf_deepspeed_config,
-        unset_hf_deepspeed_config,
-    )
-    from .integration_utils import (
-        INTEGRATION_TO_CALLBACK,
-        AzureMLCallback,
-        ClearMLCallback,
-        CodeCarbonCallback,
-        CometCallback,
-        DagsHubCallback,
-        FlyteCallback,
-        MLflowCallback,
-        NeptuneCallback,
-        NeptuneMissingConfiguration,
-        TensorBoardCallback,
-        WandbCallback,
-        get_available_reporting_integrations,
-        get_reporting_integration_callbacks,
-        hp_params,
-        is_azureml_available,
-        is_clearml_available,
-        is_codecarbon_available,
-        is_comet_available,
-        is_dagshub_available,
-        is_fairscale_available,
-        is_flyte_deck_standard_available,
-        is_flytekit_available,
-        is_mlflow_available,
-        is_neptune_available,
-        is_optuna_available,
-        is_ray_available,
-        is_ray_tune_available,
-        is_sigopt_available,
-        is_tensorboard_available,
-        is_wandb_available,
-        rewrite_logs,
-        run_hp_search_optuna,
-        run_hp_search_ray,
-        run_hp_search_sigopt,
-        run_hp_search_wandb,
-    )
-    from .peft import PeftAdapterMixin
-else:
-    import sys
-
-    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
+from .bitsandbytes import (
+    get_keys_to_not_convert,
+    replace_8bit_linear,
+    replace_with_bnb_linear,
+    set_module_8bit_tensor_to_device,
+    set_module_quantized_tensor_to_device,
+)
+from .deepspeed import (
+    HfDeepSpeedConfig,
+    HfTrainerDeepSpeedConfig,
+    deepspeed_config,
+    deepspeed_init,
+    deepspeed_load_checkpoint,
+    deepspeed_optim_sched,
+    is_deepspeed_available,
+    is_deepspeed_zero3_enabled,
+    set_hf_deepspeed_config,
+    unset_hf_deepspeed_config,
+)
+from .integration_utils import (
+    INTEGRATION_TO_CALLBACK,
+    AzureMLCallback,
+    ClearMLCallback,
+    CodeCarbonCallback,
+    CometCallback,
+    DagsHubCallback,
+    FlyteCallback,
+    MLflowCallback,
+    NeptuneCallback,
+    NeptuneMissingConfiguration,
+    TensorBoardCallback,
+    WandbCallback,
+    get_available_reporting_integrations,
+    get_reporting_integration_callbacks,
+    hp_params,
+    is_azureml_available,
+    is_clearml_available,
+    is_codecarbon_available,
+    is_comet_available,
+    is_dagshub_available,
+    is_fairscale_available,
+    is_flyte_deck_standard_available,
+    is_flytekit_available,
+    is_mlflow_available,
+    is_neptune_available,
+    is_optuna_available,
+    is_ray_available,
+    is_ray_tune_available,
+    is_sigopt_available,
+    is_tensorboard_available,
+    is_wandb_available,
+    rewrite_logs,
+    run_hp_search_optuna,
+    run_hp_search_ray,
+    run_hp_search_sigopt,
+    run_hp_search_wandb,
+)
+from .peft import PeftAdapterMixin

src/transformers/integrations/deepspeed.py

@@ -15,7 +15,6 @@
 Integration with Deepspeed
 """
 
-import importlib.metadata as importlib_metadata
 import importlib.util
 import weakref
 from functools import partialmethod
@@ -33,16 +32,7 @@ logger = logging.get_logger(__name__)
 
 
 def is_deepspeed_available():
-    package_exists = importlib.util.find_spec("deepspeed") is not None
-
-    # Check we're not importing a "deepspeed" directory somewhere but the actual library by trying to grab the version
-    # AND checking it has an author field in the metadata that is HuggingFace.
-    if package_exists:
-        try:
-            _ = importlib_metadata.metadata("deepspeed")
-            return True
-        except importlib_metadata.PackageNotFoundError:
-            return False
+    return importlib.util.find_spec("deepspeed") is not None
 
 
 if is_accelerate_available() and is_deepspeed_available():

src/transformers/integrations/integration_utils.py

@@ -205,16 +205,10 @@ def run_hp_search_optuna(trainer, n_trials: int, direction: str, **kwargs) -> Be
 
         timeout = kwargs.pop("timeout", None)
         n_jobs = kwargs.pop("n_jobs", 1)
-        directions = direction if isinstance(direction, list) else None
-        direction = None if directions is not None else direction
-        study = optuna.create_study(direction=direction, directions=directions, **kwargs)
+        study = optuna.create_study(direction=direction, **kwargs)
         study.optimize(_objective, n_trials=n_trials, timeout=timeout, n_jobs=n_jobs)
-        if not study._is_multi_objective():
-            best_trial = study.best_trial
-            return BestRun(str(best_trial.number), best_trial.value, best_trial.params)
-        else:
-            best_trials = study.best_trials
-            return [BestRun(str(best.number), best.values, best.params) for best in best_trials]
+        best_trial = study.best_trial
+        return BestRun(str(best_trial.number), best_trial.value, best_trial.params)
     else:
         for i in range(n_trials):
             trainer.objective = None

src/transformers/integrations/peft.py

@@ -12,14 +12,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import inspect
-from typing import TYPE_CHECKING, Any, Dict, Optional
+from typing import Optional
 
 from ..utils import (
     check_peft_version,
     find_adapter_config_file,
     is_accelerate_available,
     is_peft_available,
-    is_torch_available,
     logging,
 )
 
@@ -31,11 +30,6 @@ if is_accelerate_available():
 # Minimum PEFT version supported for the integration
 MIN_PEFT_VERSION = "0.5.0"
 
-if TYPE_CHECKING:
-    if is_torch_available():
-        import torch
-
-
 logger = logging.get_logger(__name__)
 
 
@@ -67,7 +61,7 @@ class PeftAdapterMixin:
 
     def load_adapter(
         self,
-        peft_model_id: Optional[str] = None,
+        peft_model_id: str,
         adapter_name: Optional[str] = None,
         revision: Optional[str] = None,
         token: Optional[str] = None,
@@ -75,8 +69,6 @@ class PeftAdapterMixin:
         max_memory: Optional[str] = None,
         offload_folder: Optional[str] = None,
         offload_index: Optional[int] = None,
-        peft_config: Dict[str, Any] = None,
-        adapter_state_dict: Optional[Dict[str, "torch.Tensor"]] = None,
     ) -> None:
         """
         Load adapter weights from file or remote Hub folder. If you are not familiar with adapters and PEFT methods, we
@@ -85,7 +77,7 @@ class PeftAdapterMixin:
         Requires peft as a backend to load the adapter weights.
 
         Args:
-            peft_model_id (`str`, *optional*):
+            peft_model_id (`str`):
                 The identifier of the model to look for on the Hub, or a local path to the saved adapter config file
                 and adapter weights.
             adapter_name (`str`, *optional*):
@@ -122,12 +114,6 @@ class PeftAdapterMixin:
                 If the `device_map` contains any value `"disk"`, the folder where we will offload weights.
             offload_index (`int`, `optional`):
                 `offload_index` argument to be passed to `accelerate.dispatch_model` method.
-            peft_config (`Dict[str, Any]`, *optional*):
-                The configuration of the adapter to add, supported adapters are non-prefix tuning and adaption prompts
-                methods. This argument is used in case users directly pass PEFT state dicts
-            adapter_state_dict (`Dict[str, torch.Tensor]`, *optional*):
-                The state dict of the adapter to load. This argument is used in case users directly pass PEFT state
-                dicts
         """
         check_peft_version(min_version=MIN_PEFT_VERSION)
 
@@ -136,41 +122,33 @@ class PeftAdapterMixin:
         from peft import PeftConfig, inject_adapter_in_model, load_peft_weights
         from peft.utils import set_peft_model_state_dict
 
-        if self._hf_peft_config_loaded and adapter_name in self.peft_config:
-            raise ValueError(f"Adapter with name {adapter_name} already exists. Please use a different name.")
-
-        if peft_model_id is None and (adapter_state_dict is None and peft_config is None):
-            raise ValueError(
-                "You should either pass a `peft_model_id` or a `peft_config` and `adapter_state_dict` to load an adapter."
-            )
-
-        if peft_config is None:
-            adapter_config_file = find_adapter_config_file(
-                peft_model_id,
-                revision=revision,
-                token=token,
-            )
-
-            if adapter_config_file is None:
-                raise ValueError(
-                    f"adapter model file not found in {peft_model_id}. Make sure you are passing the correct path to the "
-                    "adapter model."
-                )
-
-            peft_config = PeftConfig.from_pretrained(
-                peft_model_id,
-                revision=revision,
-                use_auth_token=token,
-            )
-
-        # Create and add fresh new adapters into the model.
-        inject_adapter_in_model(peft_config, self, adapter_name)
-
         if not self._hf_peft_config_loaded:
             self._hf_peft_config_loaded = True
+        elif adapter_name in self.peft_config:
+            raise ValueError(f"Adapter with name {adapter_name} already exists. Please use a different name.")
 
-        if peft_model_id is not None:
-            adapter_state_dict = load_peft_weights(peft_model_id, revision=revision, use_auth_token=token)
+        adapter_config_file = find_adapter_config_file(
+            peft_model_id,
+            revision=revision,
+            token=token,
+        )
+
+        if adapter_config_file is None:
+            raise ValueError(
+                f"adapter model file not found in {peft_model_id}. Make sure you are passing the correct path to the "
+                "adapter model."
+            )
+
+        loaded_peft_config = PeftConfig.from_pretrained(
+            peft_model_id,
+            revision=revision,
+            use_auth_token=token,
+        )
+
+        # Create and add fresh new adapters into the model.
+        inject_adapter_in_model(loaded_peft_config, self, adapter_name)
+
+        adapter_state_dict = load_peft_weights(peft_model_id, revision=revision, use_auth_token=token)
 
         # We need to pre-process the state dict to remove unneeded prefixes - for backward compatibility
         processed_adapter_state_dict = {}

src/transformers/modeling_keras_outputs.py

@@ -0,0 +1,991 @@
+# 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.
+
+from __future__ import annotations
+
+import warnings
+from dataclasses import dataclass
+from typing import List, Optional, Tuple
+
+import keras_core as keras
+
+from .utils import ModelOutput
+
+
+@dataclass
+class KerasBaseModelOutput(ModelOutput):
+    """
+    Base class for model's outputs, with potential hidden states and attentions.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        hidden_states (`tuple(tf.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasBaseModelOutputWithNoAttention(ModelOutput):
+    """
+    Base class for model's outputs, with potential hidden states.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` shape `(batch_size, num_channels, height, width)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each layer) of shape `(batch_size, num_channels, height, width)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    hidden_states: Optional[Tuple[keras.KerasVariable, ...]] = None
+
+
+@dataclass
+class KerasBaseModelOutputWithPooling(ModelOutput):
+    """
+    Base class for model's outputs that also contains a pooling of the last hidden states.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        pooler_output (`keras.KerasVariable` of shape `(batch_size, hidden_size)`):
+            Last layer hidden-state of the first token of the sequence (classification token) further processed by a
+            Linear layer and a Tanh activation function. The Linear layer weights are trained from the next sentence
+            prediction (classification) objective during pretraining.
+
+            This output is usually *not* a good summary of the semantic content of the input, you're often better with
+            averaging or pooling the sequence of hidden-states for the whole input sequence.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    pooler_output: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasBaseModelOutputWithPoolingAndNoAttention(ModelOutput):
+    """
+    Base class for model's outputs that also contains a pooling of the last hidden states.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, num_channels, height, width)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        pooler_output (`keras.KerasVariable` of shape `(batch_size, hidden_size)`):
+            Last layer hidden-state after a pooling operation on the spatial dimensions.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each layer) of shape `(batch_size, num_channels, height, width)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    pooler_output: keras.KerasVariable = None
+    hidden_states: Optional[Tuple[keras.KerasVariable, ...]] = None
+
+
+@dataclass
+class KerasBaseModelOutputWithPoolingAndCrossAttentions(ModelOutput):
+    """
+    Base class for model's outputs that also contains a pooling of the last hidden states.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        pooler_output (`keras.KerasVariable` of shape `(batch_size, hidden_size)`):
+            Last layer hidden-state of the first token of the sequence (classification token) further processed by a
+            Linear layer and a Tanh activation function. The Linear layer weights are trained from the next sentence
+            prediction (classification) objective during pretraining.
+
+            This output is usually *not* a good summary of the semantic content of the input, you're often better with
+            averaging or pooling the sequence of hidden-states for the whole input sequence.
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
+            weighted average in the cross-attention heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    pooler_output: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasBaseModelOutputWithPast(ModelOutput):
+    """
+    Base class for model's outputs that may also contain a past key/values (to speed up sequential decoding).
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+
+            If `past_key_values` is used only the last hidden-state of the sequences of shape `(batch_size, 1,
+            hidden_size)` is output.
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasBaseModelOutputWithCrossAttentions(ModelOutput):
+    """
+    Base class for model's outputs, with potential hidden states and attentions.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        hidden_states (`tuple(tf.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
+            weighted average in the cross-attention heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasBaseModelOutputWithPastAndCrossAttentions(ModelOutput):
+    """
+    Base class for model's outputs that may also contain a past key/values (to speed up sequential decoding).
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+
+            If `past_key_values` is used only the last hidden-state of the sequences of shape `(batch_size, 1,
+            hidden_size)` is output.
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(tf.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
+            weighted average in the cross-attention heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSeq2SeqModelOutput(ModelOutput):
+    """
+    Base class for model encoder's outputs that also contains : pre-computed hidden states that can speed up sequential
+    decoding.
+
+    Args:
+        last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the decoder of the model.
+
+            If `past_key_values` is used only the last hidden-state of the sequences of shape `(batch_size, 1,
+            hidden_size)` is output.
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be
+            used (see `past_key_values` input) to speed up sequential decoding.
+        decoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
+        decoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
+            weighted average in the cross-attention heads.
+        encoder_last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Sequence of hidden-states at the output of the last layer of the encoder of the model.
+        encoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
+        encoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+    """
+
+    last_hidden_state: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    decoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    decoder_attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+    encoder_last_hidden_state: keras.KerasVariable | None = None
+    encoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    encoder_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasCausalLMOutput(ModelOutput):
+    """
+    Base class for causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of non-masked labels, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`keras.KerasVariable` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasCausalLMOutputWithPast(ModelOutput):
+    """
+    Base class for causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of non-masked labels, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`keras.KerasVariable` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasCausalLMOutputWithCrossAttentions(ModelOutput):
+    """
+    Base class for causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of non-masked labels, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`keras.KerasVariable` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
+            weighted average in the cross-attention heads.
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasMaskedLMOutput(ModelOutput):
+    """
+    Base class for masked language models outputs.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of non-masked labels, returned when `labels` is provided):
+            Masked language modeling (MLM) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSeq2SeqLMOutput(ModelOutput):
+    """
+    Base class for sequence-to-sequence language models outputs.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of non-masked labels, returned when `labels` is provided):
+            Language modeling loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be
+            used (see `past_key_values` input) to speed up sequential decoding.
+        decoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
+        decoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder's cross-attention layer, after the attention softmax, used to compute the
+            weighted average in the cross-attention heads.
+        encoder_last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Sequence of hidden-states at the output of the last layer of the encoder of the model.
+        encoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
+        encoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    decoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    decoder_attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+    encoder_last_hidden_state: keras.KerasVariable | None = None
+    encoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    encoder_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasNextSentencePredictorOutput(ModelOutput):
+    """
+    Base class for outputs of models predicting if two sentences are consecutive or not.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of non-masked labels, returned when `next_sentence_label` is provided):
+            Next sentence prediction loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, 2)`):
+            Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation
+            before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSequenceClassifierOutput(ModelOutput):
+    """
+    Base class for outputs of sentence classification models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(batch_size, )`, *optional*, returned when `labels` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels)`):
+            Classification (or regression if config.num_labels==1) scores (before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSeq2SeqSequenceClassifierOutput(ModelOutput):
+    """
+    Base class for outputs of sequence-to-sequence sentence classification models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `label` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels)`):
+            Classification (or regression if config.num_labels==1) scores (before SoftMax).
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be
+            used (see `past_key_values` input) to speed up sequential decoding.
+        decoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
+        decoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+        cross_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`
+        encoder_last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Sequence of hidden-states at the output of the last layer of the encoder of the model.
+        encoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
+        encoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    decoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    decoder_attentions: Tuple[keras.KerasVariable] | None = None
+    cross_attentions: Tuple[keras.KerasVariable] | None = None
+    encoder_last_hidden_state: keras.KerasVariable | None = None
+    encoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    encoder_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSemanticSegmenterOutput(ModelOutput):
+    """
+    Base class for outputs of semantic segmentation models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels, logits_height, logits_width)`):
+            Classification scores for each pixel.
+
+            <Tip warning={true}>
+
+            The logits returned do not necessarily have the same size as the `pixel_values` passed as inputs. This is
+            to avoid doing two interpolations and lose some quality when a user needs to resize the logits to the
+            original image size as post-processing. You should always check your logits shape and resize as needed.
+
+            </Tip>
+
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each layer) of shape `(batch_size, patch_size, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, patch_size, sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSemanticSegmenterOutputWithNoAttention(ModelOutput):
+    """
+    Base class for outputs of semantic segmentation models that do not output attention scores.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels, logits_height, logits_width)`):
+            Classification scores for each pixel.
+
+            <Tip warning={true}>
+
+            The logits returned do not necessarily have the same size as the `pixel_values` passed as inputs. This is
+            to avoid doing two interpolations and lose some quality when a user needs to resize the logits to the
+            original image size as post-processing. You should always check your logits shape and resize as needed.
+
+            </Tip>
+
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each layer) of shape `(batch_size, patch_size, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasImageClassifierOutput(ModelOutput):
+    """
+    Base class for outputs of image classification models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels)`):
+            Classification (or regression if config.num_labels==1) scores (before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called
+            feature maps) of the model at the output of each stage.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, patch_size, sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasMultipleChoiceModelOutput(ModelOutput):
+    """
+    Base class for outputs of multiple choice models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape *(batch_size, )*, *optional*, returned when `labels` is provided):
+            Classification loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, num_choices)`):
+            *num_choices* is the second dimension of the input tensors. (see *input_ids* above).
+
+            Classification scores (before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasTokenClassifierOutput(ModelOutput):
+    """
+    Base class for outputs of token classification models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(n,)`, *optional*, where n is the number of unmasked labels, returned when `labels` is provided) :
+            Classification loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, sequence_length, config.num_labels)`):
+            Classification scores (before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasQuestionAnsweringModelOutput(ModelOutput):
+    """
+    Base class for outputs of question answering models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(batch_size, )`, *optional*, returned when `start_positions` and `end_positions` are provided):
+            Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
+        start_logits (`keras.KerasVariable` of shape `(batch_size, sequence_length)`):
+            Span-start scores (before SoftMax).
+        end_logits (`keras.KerasVariable` of shape `(batch_size, sequence_length)`):
+            Span-end scores (before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    start_logits: keras.KerasVariable = None
+    end_logits: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSeq2SeqQuestionAnsweringModelOutput(ModelOutput):
+    """
+    Base class for outputs of sequence-to-sequence question answering models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
+        start_logits (`keras.KerasVariable` of shape `(batch_size, sequence_length)`):
+            Span-start scores (before SoftMax).
+        end_logits (`keras.KerasVariable` of shape `(batch_size, sequence_length)`):
+            Span-end scores (before SoftMax).
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be
+            used (see `past_key_values` input) to speed up sequential decoding.
+        decoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
+        decoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+        encoder_last_hidden_state (`keras.KerasVariable` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Sequence of hidden-states at the output of the last layer of the encoder of the model.
+        encoder_hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
+        encoder_attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
+            self-attention heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    start_logits: keras.KerasVariable = None
+    end_logits: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    decoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    decoder_attentions: Tuple[keras.KerasVariable] | None = None
+    encoder_last_hidden_state: keras.KerasVariable | None = None
+    encoder_hidden_states: Tuple[keras.KerasVariable] | None = None
+    encoder_attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasSequenceClassifierOutputWithPast(ModelOutput):
+    """
+    Base class for outputs of sentence classification models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(batch_size, )`, *optional*, returned when `labels` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels)`):
+            Classification (or regression if config.num_labels==1) scores (before SoftMax).
+        past_key_values (`List[keras.KerasVariable]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            List of `keras.KerasVariable` of length `config.n_layers`, with each tensor of shape `(2, batch_size, num_heads,
+            sequence_length, embed_size_per_head)`).
+
+            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings + one for the output of each layer) of shape
+            `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    past_key_values: List[keras.KerasVariable] | None = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+
+@dataclass
+class KerasImageClassifierOutputWithNoAttention(ModelOutput):
+    """
+    Base class for outputs of image classification models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Classification (or regression if config.num_labels==1) loss.
+        logits (`keras.KerasVariable` of shape `(batch_size, config.num_labels)`):
+            Classification (or regression if config.num_labels==1) scores (before SoftMax).
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each stage) of shape `(batch_size, num_channels, height, width)`. Hidden-states (also called
+            feature maps) of the model at the output of each stage.
+    """
+
+    loss: keras.KerasVariable | None = None
+    logits: keras.KerasVariable = None
+    hidden_states: Optional[Tuple[keras.KerasVariable, ...]] = None
+
+
+@dataclass
+class KerasMaskedImageModelingOutput(ModelOutput):
+    """
+    Base class for outputs of masked image completion / in-painting models.
+
+    Args:
+        loss (`keras.KerasVariable` of shape `(1,)`, *optional*, returned when `bool_masked_pos` is provided):
+            Reconstruction loss.
+        reconstruction (`keras.KerasVariable` of shape `(batch_size, num_channels, height, width)`):
+           Reconstructed / completed images.
+        hidden_states (`tuple(keras.KerasVariable)`, *optional*, returned when `output_hidden_states=True` is passed or when
+        `config.output_hidden_states=True`):
+            Tuple of `keras.KerasVariable` (one for the output of the embeddings, if the model has an embedding layer, + one for
+            the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called
+            feature maps) of the model at the output of each stage.
+        attentions (`tuple(keras.KerasVariable)`, *optional*, returned when `output_attentions=True` is passed or when
+        `config.output_attentions=True`):
+            Tuple of `keras.KerasVariable` (one for each layer) of shape `(batch_size, num_heads, patch_size, sequence_length)`.
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: keras.KerasVariable | None = None
+    reconstruction: keras.KerasVariable = None
+    hidden_states: Tuple[keras.KerasVariable] | None = None
+    attentions: Tuple[keras.KerasVariable] | None = None
+
+    @property
+    def logits(self):
+        warnings.warn(
+            "logits attribute is deprecated and will be removed in version 5 of Transformers."
+            " Please use the reconstruction attribute to retrieve the final output instead.",
+            FutureWarning,
+        )
+        return self.reconstruction

src/transformers/modeling_utils.py

@@ -81,12 +81,7 @@ from .utils import (
     strtobool,
 )
 from .utils.hub import convert_file_size_to_int, get_checkpoint_shard_files
-from .utils.import_utils import (
-    ENV_VARS_TRUE_VALUES,
-    is_sagemaker_mp_enabled,
-    is_torch_fx_proxy,
-    is_torchdynamo_compiling,
-)
+from .utils.import_utils import ENV_VARS_TRUE_VALUES, is_sagemaker_mp_enabled, is_torch_fx_proxy
 from .utils.quantization_config import BitsAndBytesConfig, GPTQConfig, QuantizationMethod
 from .utils.versions import require_version_core
 
@@ -120,15 +115,11 @@ _init_weights = True
 
 
 def is_fsdp_enabled():
-    return (
-        torch.distributed.is_available()
-        and torch.distributed.is_initialized()
-        and strtobool(os.environ.get("ACCELERATE_USE_FSDP", "False")) == 1
-    )
+    return torch.distributed.is_initialized() and strtobool(os.environ.get("ACCELERATE_USE_FSDP", "False")) == 1
 
 
 def is_fsdp_enabled_and_dist_rank_0():
-    return is_fsdp_enabled() and int(os.environ.get("LOCAL_RANK", -1)) == 0
+    return is_fsdp_enabled() and torch.distributed.get_rank() == 0
 
 
 if is_sagemaker_mp_enabled():
@@ -989,33 +980,12 @@ class ModuleUtilsMixin:
             embedding_param_names = [
                 f"{name}.weight" for name, module_type in self.named_modules() if isinstance(module_type, nn.Embedding)
             ]
-            total_parameters = [
+            non_embedding_parameters = [
                 parameter for name, parameter in self.named_parameters() if name not in embedding_param_names
             ]
+            return sum(p.numel() for p in non_embedding_parameters if p.requires_grad or not only_trainable)
         else:
-            total_parameters = list(self.parameters())
-
-        total_numel = []
-        is_loaded_in_4bit = getattr(self, "is_loaded_in_4bit", False)
-        if is_loaded_in_4bit:
-            if is_bitsandbytes_available():
-                import bitsandbytes as bnb
-            else:
-                raise ValueError(
-                    "bitsandbytes is not installed but it seems that the model has been loaded in 4bit precision, something went wrong"
-                    " make sure to install bitsandbytes with `pip install bitsandbytes`."
-                )
-
-        for param in total_parameters:
-            if param.requires_grad or not only_trainable:
-                # For 4bit models, we need to multiply the number of parameters by 2 as half of the parameters are
-                # used for the 4bit quantization (uint8 tensors are stored)
-                if is_loaded_in_4bit and isinstance(param, bnb.nn.Params4bit):
-                    total_numel.append(param.numel() * 2)
-                else:
-                    total_numel.append(param.numel())
-
-        return sum(total_numel)
+            return sum(p.numel() for p in self.parameters() if p.requires_grad or not only_trainable)
 
     def estimate_tokens(self, input_dict: Dict[str, Union[torch.Tensor, Any]]) -> int:
         """
@@ -1436,8 +1406,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 vectors at the end. Reducing the size will remove vectors from the end. If not provided or `None`, just
                 returns a pointer to the input tokens `torch.nn.Embedding` module of the model without doing anything.
             pad_to_multiple_of (`int`, *optional*):
-                If set will pad the embedding matrix to a multiple of the provided value.If `new_num_tokens` is set to
-                `None` will just pad the embedding to a multiple of `pad_to_multiple_of`.
+                If set will pad the embedding matrix to a multiple of the provided value.
 
                 This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
                 `>= 7.5` (Volta), or on TPUs which benefit from having sequence lengths be a multiple of 128. For more
@@ -1448,12 +1417,12 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
             `torch.nn.Embedding`: Pointer to the input tokens Embeddings Module of the model.
         """
         model_embeds = self._resize_token_embeddings(new_num_tokens, pad_to_multiple_of)
-        if new_num_tokens is None and pad_to_multiple_of is None:
+        if new_num_tokens is None:
             return model_embeds
 
         # Update base model and current model config
-        self.config.vocab_size = model_embeds.weight.shape[0]
-        self.vocab_size = model_embeds.weight.shape[0]
+        self.config.vocab_size = new_num_tokens
+        self.vocab_size = new_num_tokens
 
         # Tie weights again if needed
         self.tie_weights()
@@ -1509,8 +1478,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 vectors from the end. If not provided or `None`, just returns a pointer to the input tokens
                 `torch.nn.Embedding` module of the model without doing anything.
             pad_to_multiple_of (`int`, *optional*):
-                If set will pad the embedding matrix to a multiple of the provided value. If `new_num_tokens` is set to
-                `None` will just pad the embedding to a multiple of `pad_to_multiple_of`.
+                If set will pad the embedding matrix to a multiple of the provided value.
 
                 This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
                 `>= 7.5` (Volta), or on TPUs which benefit from having sequence lengths be a multiple of 128. For more
@@ -1530,7 +1498,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 )
             if new_num_tokens is None:
                 new_num_tokens = old_embeddings.weight.shape[0]
-            new_num_tokens = ((new_num_tokens + pad_to_multiple_of - 1) // pad_to_multiple_of) * pad_to_multiple_of
+            new_num_tokens = ((new_num_tokens // pad_to_multiple_of) + 1) * pad_to_multiple_of
         else:
             logger.warning(
                 "You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding"
@@ -1550,7 +1518,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         else:
             old_num_tokens, old_embedding_dim = old_embeddings.weight.size()
 
-        if old_num_tokens == new_num_tokens and not is_deepspeed_zero3_enabled():
+        if old_num_tokens == new_num_tokens:
             return old_embeddings
 
         if not isinstance(old_embeddings, nn.Embedding):
@@ -1560,34 +1528,40 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 f" {nn.Embedding}."
             )
 
-        # Build new embeddings
-
-        # When using DeepSpeed ZeRO-3, we shouldn't create new embeddings with DeepSpeed init
-        # because the shape of the new embedding layer is used across various modeling files
-        # as well as to update config vocab size. Shape will be 0 when using DeepSpeed init leading
-        # to errors when training.
-        new_embeddings = nn.Embedding(
-            new_num_tokens,
-            old_embedding_dim,
-            device=old_embeddings.weight.device,
-            dtype=old_embeddings.weight.dtype,
-        )
-
-        # initialize all new embeddings (in particular added tokens)
-        self._init_weights(new_embeddings)
-
-        # Copy token embeddings from the previous weights
-
         # numbers of tokens to copy
         n = min(old_num_tokens, new_num_tokens)
-
         if is_deepspeed_zero3_enabled():
             import deepspeed
 
+            with deepspeed.zero.Init(config_dict_or_path=deepspeed_config()):
+                # Build new embeddings
+                new_embeddings = nn.Embedding(
+                    new_num_tokens,
+                    old_embedding_dim,
+                    device=old_embeddings.weight.device,
+                    dtype=old_embeddings.weight.dtype,
+                )
+
             params = [old_embeddings.weight, new_embeddings.weight]
             with deepspeed.zero.GatheredParameters(params, modifier_rank=0):
+                # initialize all new embeddings (in particular added tokens)
+                self._init_weights(new_embeddings)
+
+                # Copy token embeddings from the previous weights
                 new_embeddings.weight.data[:n, :] = old_embeddings.weight.data[:n, :]
         else:
+            # Build new embeddings
+            new_embeddings = nn.Embedding(
+                new_num_tokens,
+                old_embedding_dim,
+                device=old_embeddings.weight.device,
+                dtype=old_embeddings.weight.dtype,
+            )
+
+            # initialize all new embeddings (in particular added tokens)
+            self._init_weights(new_embeddings)
+
+            # Copy token embeddings from the previous weights
             new_embeddings.weight.data[:n, :] = old_embeddings.weight.data[:n, :]
 
         return new_embeddings
@@ -1630,7 +1604,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 old_lm_head.weight.size() if not transposed else old_lm_head.weight.t().size()
             )
 
-        if old_num_tokens == new_num_tokens and not is_deepspeed_zero3_enabled():
+        if old_num_tokens == new_num_tokens:
             return old_lm_head
 
         if not isinstance(old_lm_head, nn.Linear):
@@ -1644,50 +1618,51 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         new_lm_head_shape = (old_lm_head_dim, new_num_tokens) if not transposed else (new_num_tokens, old_lm_head_dim)
         has_new_lm_head_bias = old_lm_head.bias is not None
 
-        # When using DeepSpeed ZeRO-3, we shouldn't create new embeddings with DeepSpeed init
-        # because the shape of the new embedding layer is used across various modeling files
-        # as well as to update config vocab size. Shape will be 0 when using DeepSpeed init leading
-        # to errors when training.
-        new_lm_head = nn.Linear(
-            *new_lm_head_shape,
-            bias=has_new_lm_head_bias,
-            device=old_lm_head.weight.device,
-            dtype=old_lm_head.weight.dtype,
-        )
-
-        # initialize new lm head (in particular added tokens)
-        self._init_weights(new_lm_head)
-
         num_tokens_to_copy = min(old_num_tokens, new_num_tokens)
 
+        # XXX: put the long block of code in a wrapper
         if is_deepspeed_zero3_enabled():
             import deepspeed
 
+            with deepspeed.zero.Init(config_dict_or_path=deepspeed_config()):
+                new_lm_head = nn.Linear(
+                    *new_lm_head_shape,
+                    bias=has_new_lm_head_bias,
+                    device=old_lm_head.weight.device,
+                    dtype=old_lm_head.weight.dtype,
+                )
             params = [old_lm_head.weight, old_lm_head.bias, new_lm_head.weight, new_lm_head.bias]
             with deepspeed.zero.GatheredParameters(params, modifier_rank=0):
-                self._copy_lm_head_original_to_resized(
-                    new_lm_head, old_lm_head, num_tokens_to_copy, transposed, has_new_lm_head_bias
-                )
+                self._init_weights(new_lm_head)
+                # Copy old lm head weights to new lm head
+                if not transposed:
+                    new_lm_head.weight.data[:num_tokens_to_copy, :] = old_lm_head.weight.data[:num_tokens_to_copy, :]
+                else:
+                    new_lm_head.weight.data[:, :num_tokens_to_copy] = old_lm_head.weight.data[:, :num_tokens_to_copy]
+
+                # Copy bias weights to new lm head
+                if has_new_lm_head_bias:
+                    new_lm_head.bias.data[:num_tokens_to_copy] = old_lm_head.bias.data[:num_tokens_to_copy]
         else:
-            self._copy_lm_head_original_to_resized(
-                new_lm_head, old_lm_head, num_tokens_to_copy, transposed, has_new_lm_head_bias
+            new_lm_head = nn.Linear(
+                *new_lm_head_shape,
+                bias=has_new_lm_head_bias,
+                device=old_lm_head.weight.device,
+                dtype=old_lm_head.weight.dtype,
             )
+            self._init_weights(new_lm_head)
+            # Copy old lm head weights to new lm head
+            if not transposed:
+                new_lm_head.weight.data[:num_tokens_to_copy, :] = old_lm_head.weight.data[:num_tokens_to_copy, :]
+            else:
+                new_lm_head.weight.data[:, :num_tokens_to_copy] = old_lm_head.weight.data[:, :num_tokens_to_copy]
+
+            # Copy bias weights to new lm head
+            if has_new_lm_head_bias:
+                new_lm_head.bias.data[:num_tokens_to_copy] = old_lm_head.bias.data[:num_tokens_to_copy]
 
         return new_lm_head
 
-    def _copy_lm_head_original_to_resized(
-        self, new_lm_head, old_lm_head, num_tokens_to_copy, transposed, has_new_lm_head_bias
-    ):
-        # Copy old lm head weights to new lm head
-        if not transposed:
-            new_lm_head.weight.data[:num_tokens_to_copy, :] = old_lm_head.weight.data[:num_tokens_to_copy, :]
-        else:
-            new_lm_head.weight.data[:, :num_tokens_to_copy] = old_lm_head.weight.data[:, :num_tokens_to_copy]
-
-        # Copy bias weights to new lm head
-        if has_new_lm_head_bias:
-            new_lm_head.bias.data[:num_tokens_to_copy] = old_lm_head.bias.data[:num_tokens_to_copy]
-
     def resize_position_embeddings(self, new_num_position_embeddings: int):
         raise NotImplementedError(
             f"`resize_position_embeddings` is not implemented for {self.__class__}`. To implement it, you should "
@@ -1745,13 +1720,6 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
             raise ValueError(f"{self.__class__.__name__} does not support gradient checkpointing.")
         self.apply(partial(self._set_gradient_checkpointing, value=True))
 
-        if getattr(self, "_hf_peft_config_loaded", False):
-            # When using PEFT + gradient checkpointing + Trainer we need to make sure the input has requires_grad=True
-            # we do it also on PEFT: https://github.com/huggingface/peft/blob/85013987aa82aa1af3da1236b6902556ce3e483e/src/peft/peft_model.py#L334
-            # When training with PEFT, only LoRA layers will have requires grad set to True, but the output of frozen layers need to propagate
-            # the gradients to make sure the gradient flows.
-            self.enable_input_require_grads()
-
     def gradient_checkpointing_disable(self):
         """
         Deactivates gradient checkpointing for the current model.
@@ -1762,9 +1730,6 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         if self.supports_gradient_checkpointing:
             self.apply(partial(self._set_gradient_checkpointing, value=False))
 
-        if getattr(self, "_hf_peft_config_loaded", False):
-            self.disable_input_require_grads()
-
     @property
     def is_gradient_checkpointing(self) -> bool:
         """
@@ -2950,18 +2915,26 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 dtype_orig = cls._set_default_torch_dtype(torch_dtype)
 
             # Check if `_keep_in_fp32_modules` is not None
-            use_keep_in_fp32_modules = (cls._keep_in_fp32_modules is not None) and (
-                torch_dtype == torch.float16 or load_in_4bit or load_in_8bit
+            use_keep_in_fp32_modules = (
+                (cls._keep_in_fp32_modules is not None)
+                and is_accelerate_available()
+                and (torch_dtype == torch.float16 or load_in_4bit or load_in_8bit)
             )
+            if (
+                (cls._keep_in_fp32_modules is not None)
+                and not is_accelerate_available()
+                and torch_dtype == torch.float16
+            ):
+                logger.warning(
+                    "For stability purposes, it is recommended to have accelerate installed when using this model in"
+                    " torch.float16, please install it with `pip install accelerate`"
+                )
 
             if is_sharded:
                 loaded_state_dict_keys = sharded_metadata["all_checkpoint_keys"]
             else:
                 loaded_state_dict_keys = list(state_dict.keys())
-            if low_cpu_mem_usage or (use_keep_in_fp32_modules and is_accelerate_available()):
-                # In case some weights need to be kept in float32 and accelerate is not installed,
-                # we later on want to take the path where state_dict is not None, that is the one
-                # that do not require accelerate.
+            if low_cpu_mem_usage or use_keep_in_fp32_modules:
                 state_dict = None
 
         config.name_or_path = pretrained_model_name_or_path
@@ -2982,8 +2955,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
 
         # Check first if we are `from_pt`
         if use_keep_in_fp32_modules:
-            if is_accelerate_available():
-                low_cpu_mem_usage = True
+            low_cpu_mem_usage = True
             keep_in_fp32_modules = model._keep_in_fp32_modules
         else:
             keep_in_fp32_modules = []
@@ -3458,8 +3430,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         if keep_in_fp32_modules is not None:
             for name, param in model.named_parameters():
                 if any(module_to_keep_in_fp32 in name for module_to_keep_in_fp32 in keep_in_fp32_modules):
-                    # param = param.to(torch.float32) does not work here as only in the local scope.
-                    param.data = param.data.to(torch.float32)
+                    param = param.to(torch.float32)
 
         # Make sure we are able to load base models as well as derived models (with heads)
         start_prefix = ""
@@ -3586,6 +3557,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                     remove_prefix_from_model,
                     ignore_mismatched_sizes,
                 )
+
                 if low_cpu_mem_usage:
                     if not is_fsdp_enabled() or is_fsdp_enabled_and_dist_rank_0():
                         new_error_msgs, offload_index, state_dict_index = _load_state_dict_into_meta_model(
@@ -3827,7 +3799,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         """
 
         # Skip the check during tracing.
-        if is_torch_fx_proxy(input_ids) or torch.jit.is_tracing() or is_torchdynamo_compiling():
+        if is_torch_fx_proxy(input_ids) or torch.jit.is_tracing():
             return
 
         if (attention_mask is not None) or (self.config.pad_token_id is None):

src/transformers/models/__init__.py

@@ -38,7 +38,6 @@ from . import (
     blip_2,
     bloom,
     bridgetower,
-    bros,
     byt5,
     camembert,
     canine,
@@ -155,7 +154,6 @@ from . import (
     pegasus,
     pegasus_x,
     perceiver,
-    persimmon,
     phobert,
     pix2struct,
     plbart,
@@ -213,7 +211,6 @@ from . import (
     vit_mae,
     vit_msn,
     vitdet,
-    vitmatte,
     vits,
     vivit,
     wav2vec2,

src/transformers/models/albert/tokenization_albert.py

@@ -159,14 +159,6 @@ class AlbertTokenizer(PreTrainedTokenizer):
 
         self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
 
-        self.do_lower_case = do_lower_case
-        self.remove_space = remove_space
-        self.keep_accents = keep_accents
-        self.vocab_file = vocab_file
-
-        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
-        self.sp_model.Load(vocab_file)
-
         super().__init__(
             do_lower_case=do_lower_case,
             remove_space=remove_space,
@@ -182,6 +174,14 @@ class AlbertTokenizer(PreTrainedTokenizer):
             **kwargs,
         )
 
+        self.do_lower_case = do_lower_case
+        self.remove_space = remove_space
+        self.keep_accents = keep_accents
+        self.vocab_file = vocab_file
+
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(vocab_file)
+
     @property
     def vocab_size(self) -> int:
         return len(self.sp_model)
@@ -228,8 +228,6 @@ class AlbertTokenizer(PreTrainedTokenizer):
         new_pieces = []
         for piece in pieces:
             if len(piece) > 1 and piece[-1] == str(",") and piece[-2].isdigit():
-                # Logic to handle special cases see https://github.com/google-research/bert/blob/master/README.md#tokenization
-                # `9,9` -> ['▁9', ',', '9'] instead of [`_9,`, '9']
                 cur_pieces = self.sp_model.EncodeAsPieces(piece[:-1].replace(SPIECE_UNDERLINE, ""))
                 if piece[0] != SPIECE_UNDERLINE and cur_pieces[0][0] == SPIECE_UNDERLINE:
                     if len(cur_pieces[0]) == 1:

src/transformers/models/auto/configuration_auto.py

@@ -50,7 +50,6 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("blip-2", "Blip2Config"),
         ("bloom", "BloomConfig"),
         ("bridgetower", "BridgeTowerConfig"),
-        ("bros", "BrosConfig"),
         ("camembert", "CamembertConfig"),
         ("canine", "CanineConfig"),
         ("chinese_clip", "ChineseCLIPConfig"),
@@ -161,7 +160,6 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("pegasus", "PegasusConfig"),
         ("pegasus_x", "PegasusXConfig"),
         ("perceiver", "PerceiverConfig"),
-        ("persimmon", "PersimmonConfig"),
         ("pix2struct", "Pix2StructConfig"),
         ("plbart", "PLBartConfig"),
         ("poolformer", "PoolFormerConfig"),
@@ -221,7 +219,6 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("vit_mae", "ViTMAEConfig"),
         ("vit_msn", "ViTMSNConfig"),
         ("vitdet", "VitDetConfig"),
-        ("vitmatte", "VitMatteConfig"),
         ("vits", "VitsConfig"),
         ("vivit", "VivitConfig"),
         ("wav2vec2", "Wav2Vec2Config"),
@@ -263,7 +260,6 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("blip-2", "BLIP_2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("bloom", "BLOOM_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("bridgetower", "BRIDGETOWER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
-        ("bros", "BROS_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("camembert", "CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("canine", "CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("chinese_clip", "CHINESE_CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -366,7 +362,6 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("pegasus", "PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("pegasus_x", "PEGASUS_X_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("perceiver", "PERCEIVER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
-        ("persimmon", "PERSIMMON_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("pix2struct", "PIX2STRUCT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("plbart", "PLBART_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("poolformer", "POOLFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -416,7 +411,6 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("vit_mae", "VIT_MAE_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("vit_msn", "VIT_MSN_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("vitdet", "VITDET_PRETRAINED_CONFIG_ARCHIVE_MAP"),
-        ("vitmatte", "VITMATTE_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("vits", "VITS_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("vivit", "VIVIT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("wav2vec2", "WAV_2_VEC_2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -462,7 +456,6 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("bloom", "BLOOM"),
         ("bort", "BORT"),
         ("bridgetower", "BridgeTower"),
-        ("bros", "BROS"),
         ("byt5", "ByT5"),
         ("camembert", "CamemBERT"),
         ("canine", "CANINE"),
@@ -589,7 +582,6 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("pegasus", "Pegasus"),
         ("pegasus_x", "PEGASUS-X"),
         ("perceiver", "Perceiver"),
-        ("persimmon", "Persimmon"),
         ("phobert", "PhoBERT"),
         ("pix2struct", "Pix2Struct"),
         ("plbart", "PLBart"),
@@ -653,7 +645,6 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("vit_mae", "ViTMAE"),
         ("vit_msn", "ViTMSN"),
         ("vitdet", "VitDet"),
-        ("vitmatte", "ViTMatte"),
         ("vits", "VITS"),
         ("vivit", "ViViT"),
         ("wav2vec2", "Wav2Vec2"),

src/transformers/models/auto/image_processing_auto.py

@@ -107,7 +107,6 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("vit_hybrid", "ViTHybridImageProcessor"),
         ("vit_mae", "ViTImageProcessor"),
         ("vit_msn", "ViTImageProcessor"),
-        ("vitmatte", "VitMatteImageProcessor"),
         ("xclip", "CLIPImageProcessor"),
         ("yolos", "YolosImageProcessor"),
     ]

src/transformers/models/auto/modeling_auto.py

@@ -48,7 +48,6 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("blip-2", "Blip2Model"),
         ("bloom", "BloomModel"),
         ("bridgetower", "BridgeTowerModel"),
-        ("bros", "BrosModel"),
         ("camembert", "CamembertModel"),
         ("canine", "CanineModel"),
         ("chinese_clip", "ChineseCLIPModel"),
@@ -156,7 +155,6 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("pegasus", "PegasusModel"),
         ("pegasus_x", "PegasusXModel"),
         ("perceiver", "PerceiverModel"),
-        ("persimmon", "PersimmonModel"),
         ("plbart", "PLBartModel"),
         ("poolformer", "PoolFormerModel"),
         ("prophetnet", "ProphetNetModel"),
@@ -418,7 +416,6 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "OpenAIGPTLMHeadModel"),
         ("opt", "OPTForCausalLM"),
         ("pegasus", "PegasusForCausalLM"),
-        ("persimmon", "PersimmonForCausalLM"),
         ("plbart", "PLBartForCausalLM"),
         ("prophetnet", "ProphetNetForCausalLM"),
         ("qdqbert", "QDQBertLMHeadModel"),
@@ -744,7 +741,6 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "OpenAIGPTForSequenceClassification"),
         ("opt", "OPTForSequenceClassification"),
         ("perceiver", "PerceiverForSequenceClassification"),
-        ("persimmon", "PersimmonForSequenceClassification"),
         ("plbart", "PLBartForSequenceClassification"),
         ("qdqbert", "QDQBertForSequenceClassification"),
         ("reformer", "ReformerForSequenceClassification"),
@@ -865,7 +861,6 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("big_bird", "BigBirdForTokenClassification"),
         ("biogpt", "BioGptForTokenClassification"),
         ("bloom", "BloomForTokenClassification"),
-        ("bros", "BrosForTokenClassification"),
         ("camembert", "CamembertForTokenClassification"),
         ("canine", "CanineForTokenClassification"),
         ("convbert", "ConvBertForTokenClassification"),

src/transformers/models/auto/tokenization_auto.py

@@ -89,7 +89,6 @@ else:
             ("blip-2", ("GPT2Tokenizer", "GPT2TokenizerFast" if is_tokenizers_available() else None)),
             ("bloom", (None, "BloomTokenizerFast" if is_tokenizers_available() else None)),
             ("bridgetower", ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None)),
-            ("bros", ("BertTokenizer", "BertTokenizerFast" if is_tokenizers_available() else None)),
             ("byt5", ("ByT5Tokenizer", None)),
             (
                 "camembert",
@@ -284,13 +283,6 @@ else:
                     None,
                 ),
             ),
-            (
-                "persimmon",
-                (
-                    "LlamaTokenizer" if is_sentencepiece_available() else None,
-                    "LlamaTokenizerFast" if is_tokenizers_available() else None,
-                ),
-            ),
             ("phobert", ("PhobertTokenizer", None)),
             ("pix2struct", ("T5Tokenizer", "T5TokenizerFast" if is_tokenizers_available() else None)),
             ("plbart", ("PLBartTokenizer" if is_sentencepiece_available() else None, None)),
@@ -784,7 +776,7 @@ class AutoTokenizer:
                 The configuration corresponding to the model to register.
             slow_tokenizer_class ([`PretrainedTokenizer`], *optional*):
                 The slow tokenizer to register.
-            fast_tokenizer_class ([`PretrainedTokenizerFast`], *optional*):
+            slow_tokenizer_class ([`PretrainedTokenizerFast`], *optional*):
                 The fast tokenizer to register.
         """
         if slow_tokenizer_class is None and fast_tokenizer_class is None:

src/transformers/models/bark/modeling_bark.py

@@ -1086,57 +1086,21 @@ class BarkFineModel(BarkPreTrainedModel):
             ]
         )
         self.set_input_embeddings(new_embeddings_list)
-        new_num_tokens = new_embeddings_list[0].weight.shape[0]
+        new_num_tokens = [embed.weight.shape[0] for embed in new_embeddings_list]
 
         # if word embeddings are not tied, make sure that lm head is resized as well
         if self.get_output_embeddings() is not None and not self.config.tie_word_embeddings:
             old_lm_head_list = self.get_output_embeddings()
             new_lm_head_list = nn.ModuleList(
-                [self._get_resized_lm_head(old_lm_head, new_num_tokens) for old_lm_head in old_lm_head_list]
+                [
+                    self._get_resized_lm_head(old_lm_head, new_num_token)
+                    for old_lm_head, new_num_token in zip(old_lm_head_list, new_num_tokens)
+                ]
             )
             self.set_output_embeddings(new_lm_head_list)
 
         return self.get_input_embeddings()
 
-    def resize_token_embeddings(
-        self, new_num_tokens: Optional[int] = None, pad_to_multiple_of: Optional[int] = None
-    ) -> nn.Embedding:
-        """
-        Resizes input token embeddings matrix of the model if `new_num_tokens != config.vocab_size`.
-
-        Takes care of tying weights embeddings afterwards if the model class has a `tie_weights()` method.
-
-        Arguments:
-            new_num_tokens (`int`, *optional*):
-                The number of new tokens in the embedding matrix. Increasing the size will add newly initialized
-                vectors at the end. Reducing the size will remove vectors from the end. If not provided or `None`, just
-                returns a pointer to the input tokens `torch.nn.Embedding` module of the model without doing anything.
-            pad_to_multiple_of (`int`, *optional*):
-                If set will pad the embedding matrix to a multiple of the provided value.
-
-                This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
-                `>= 7.5` (Volta), or on TPUs which benefit from having sequence lengths be a multiple of 128. For more
-                details about this, or help on choosing the correct value for resizing, refer to this guide:
-                https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html#requirements-tc
-
-        Return:
-            `torch.nn.Embedding`: Pointer to the input tokens Embeddings Module of the model.
-        """
-        model_embeds = self._resize_token_embeddings(new_num_tokens, pad_to_multiple_of)
-        if new_num_tokens is None and pad_to_multiple_of is None:
-            return model_embeds
-
-        # Update base model and current model config
-        self.config.output_vocab_size = model_embeds[0].weight.shape[0]
-        self.config.vocab_size = model_embeds[0].weight.shape[0]
-        self.output_vocab_size = model_embeds[0].weight.shape[0]
-        self.vocab_size = model_embeds[0].weight.shape[0]
-
-        # Tie weights again if needed
-        self.tie_weights()
-
-        return model_embeds
-
     def tie_weights(self):
         """
         Tie the weights between the input embeddings list and the output embeddings list.