low cpu usage default to true

fix config

.circleci/create_circleci_config.py

@@ -248,7 +248,7 @@ torch_job = CircleCIJob(
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     install_steps=["uv venv && uv pip install ."],
     parallelism=6,
-    pytest_num_workers=16
+    pytest_num_workers=4
 )
 
 tokenization_job = CircleCIJob(
@@ -256,7 +256,7 @@ tokenization_job = CircleCIJob(
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     install_steps=["uv venv && uv pip install ."],
     parallelism=6,
-    pytest_num_workers=16
+    pytest_num_workers=4
 )
 
 
@@ -265,7 +265,7 @@ tf_job = CircleCIJob(
     docker_image=[{"image":"huggingface/transformers-tf-light"}],
     install_steps=["uv venv", "uv pip install -e."],
     parallelism=6,
-    pytest_num_workers=16,
+    pytest_num_workers=4,
 )
 
 
@@ -274,7 +274,7 @@ flax_job = CircleCIJob(
     docker_image=[{"image":"huggingface/transformers-jax-light"}],
     install_steps=["uv venv && uv pip install ."],
     parallelism=6,
-    pytest_num_workers=16
+    pytest_num_workers=4
 )
 
 

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -25,7 +25,7 @@ body:
 
         Models:
 
-          - text models: @ArthurZucker and @younesbelkada
+          - text models: @ArthurZucker 
           - vision models: @amyeroberts
           - speech models: @sanchit-gandhi
           - graph models: @clefourrier
@@ -44,7 +44,7 @@ body:
           - deepspeed: HF Trainer/Accelerate: @muellerzr
           - ray/raytune: @richardliaw, @amogkam
           - Big Model Inference: @SunMarc
-          - quantization (bitsandbytes, autogpt): @SunMarc and @younesbelkada
+          - quantization (bitsandbytes, autogpt): @SunMarc
 
         Documentation: @stevhliu
 

.github/PULL_REQUEST_TEMPLATE.md

@@ -39,7 +39,7 @@ members/contributors who may be interested in your PR.
 
 Models:
 
-- text models: @ArthurZucker and @younesbelkada
+- text models: @ArthurZucker
 - vision models: @amyeroberts
 - speech models: @sanchit-gandhi
 - graph models: @clefourrier
@@ -58,7 +58,7 @@ Integrations:
 - deepspeed: HF Trainer/Accelerate: @muellerzr
 - ray/raytune: @richardliaw, @amogkam
 - Big Model Inference: @SunMarc
-- quantization (bitsandbytes, autogpt): @SunMarc and @younesbelkada
+- quantization (bitsandbytes, autogpt): @SunMarc 
 
 Documentation: @stevhliu and @MKhalusova
 

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

@@ -27,10 +27,10 @@ jobs:
     strategy:
       matrix:
         file: ["quality", "consistency", "custom-tokenizers", "torch-light", "tf-light", "exotic-models", "torch-tf-light", "torch-jax-light", "jax-light", "examples-torch",  "examples-tf"]
-    continue-on-error: true 
+    continue-on-error: true
 
     steps:
-      - 
+      -
         name: Set tag
         run: |
               if ${{contains(github.event.head_commit.message, '[build-ci-image]')}}; then
@@ -61,4 +61,17 @@ jobs:
             REF=${{ github.sha }}
           file: "./docker/${{ matrix.file }}.dockerfile"
           push: ${{ contains(github.event.head_commit.message, 'ci-image]') ||  github.event_name == 'schedule' }}
-          tags: ${{ env.TAG }}
+          tags: ${{ env.TAG }}
+
+  notify:
+    runs-on: ubuntu-22.04
+    if: ${{ contains(github.event.head_commit.message, '[build-ci-image]') || contains(github.event.head_commit.message, '[push-ci-image]') && '!cancelled()' || github.event_name == 'schedule' }}
+    steps:
+      - name: Post to Slack
+        if: ${{ contains(github.event.head_commit.message, '[push-ci-image]') && github.event_name != 'schedule' }}
+        uses: huggingface/hf-workflows/.github/actions/post-slack@main
+        with:
+          slack_channel: "#transformers-ci-circleci-images"
+          title: 🤗 New docker images for CircleCI are pushed.
+          status: ${{ job.status }}
+          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

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

@@ -15,16 +15,6 @@ jobs:
     name: "Nightly PyTorch + Stable TensorFlow"
     runs-on: [intel-cpu, 8-cpu, ci]
     steps:
-      - name: Cleanup disk
-        run: |
-          sudo ls -l /usr/local/lib/
-          sudo ls -l /usr/share/
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /usr/share/dotnet
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
       -
         name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v2
@@ -52,16 +42,6 @@ jobs:
     name: "Nightly PyTorch + DeepSpeed"
     runs-on: [intel-cpu, 8-cpu, ci]
     steps:
-      - name: Cleanup disk
-        run: |
-          sudo ls -l /usr/local/lib/
-          sudo ls -l /usr/share/
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /usr/share/dotnet
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
       -
         name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v2

.github/workflows/model_jobs.yml

@@ -12,6 +12,12 @@ on:
       slice_id:
         required: true
         type: number
+      runner:
+        required: true
+        type: string
+      docker:
+        required: true
+        type: string
 
 env:
   HF_HOME: /mnt/cache
@@ -31,12 +37,13 @@ jobs:
   run_models_gpu:
     name: " "
     strategy:
+      max-parallel: 8
       fail-fast: false
       matrix:
         folders: ${{ fromJson(inputs.folder_slices)[inputs.slice_id] }}
-    runs-on: ['${{ inputs.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ inputs.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
-      image: huggingface/transformers-all-latest-gpu
+      image: ${{ inputs.docker }}
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Echo input and matrix info
@@ -65,6 +72,18 @@ jobs:
         working-directory: /transformers
         run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
 
+      - name: Update / Install some packages (for Past CI)
+        if: ${{ contains(inputs.docker, '-past-') }}
+        working-directory: /transformers
+        run: |
+          python3 -m pip install -U datasets
+
+      - name: Update / Install some packages (for Past CI)
+        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
+        working-directory: /transformers
+        run: |
+          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
+
       - name: NVIDIA-SMI
         run: |
           nvidia-smi

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

@@ -0,0 +1,43 @@
+name: Self-hosted runner (nightly-ci)
+
+
+on:
+  repository_dispatch:
+  schedule:
+    - cron: "17 2 * * *"
+  push:
+    branches:
+      - run_nightly_ci*
+
+jobs:
+  build_nightly_ci_images:
+    name: Build Nightly CI Docker Images
+    if: (github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_nightly_ci'))
+    uses: ./.github/workflows/build-nightly-ci-docker-images.yml
+    secrets: inherit
+
+  model-ci:
+    name: Model CI
+    needs: [build_nightly_ci_images]
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_models_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: ci
+      docker: huggingface/transformers-all-latest-torch-nightly-gpu
+      ci_event: Nightly CI
+    secrets: inherit
+
+  deepspeed-ci:
+    name: DeepSpeed CI
+    needs: [build_nightly_ci_images]
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_torch_cuda_extensions_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: ci
+      # test deepspeed nightly build with the latest release torch
+      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      ci_event: Nightly CI
+      working-directory-prefix: /workspace
+    secrets: inherit

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

@@ -2,32 +2,30 @@ name: Self-hosted runner (nightly-past-ci-caller)
 
 on:
   schedule:
-    # 2:17 am on each Sunday and Thursday
-
-    - cron: "17 2 * * 0,4"
+    - cron: "17 2,14 * * *"
   push:
     branches:
-      - run_nightly_ci*
       - run_past_ci*
 
 jobs:
-  build_nightly_ci_images:
-    name: Build Nightly CI Docker Images
-    if: (github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_nightly_ci'))
-    uses: ./.github/workflows/build-nightly-ci-docker-images.yml
-    secrets: inherit
-
-  run_nightly_ci:
-    name: Nightly CI
-    needs: [build_nightly_ci_images]
-    uses: ./.github/workflows/self-nightly-scheduled.yml
-    secrets: inherit
+  get_number:
+    name: Get number
+    runs-on: ubuntu-22.04
+    outputs:
+      run_number: ${{ steps.get_number.outputs.run_number }}
+    steps:
+      - name: Get number
+        id: get_number
+        run: |
+          echo "${{ github.run_number }}"
+          echo "$(python3 -c 'print(int(${{ github.run_number }}) % 10)')"
+          echo "run_number=$(python3 -c 'print(int(${{ github.run_number }}) % 10)')" >> $GITHUB_OUTPUT
 
   run_past_ci_pytorch_1-13:
     name: PyTorch 1.13
-    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    needs: [run_nightly_ci]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 0 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: pytorch
       version: "1.13"
@@ -36,9 +34,9 @@ jobs:
 
   run_past_ci_pytorch_1-12:
     name: PyTorch 1.12
-    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    needs: [run_past_ci_pytorch_1-13]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 1 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: pytorch
       version: "1.12"
@@ -47,9 +45,9 @@ jobs:
 
   run_past_ci_pytorch_1-11:
     name: PyTorch 1.11
-    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    needs: [run_past_ci_pytorch_1-12]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 2 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: pytorch
       version: "1.11"
@@ -58,9 +56,9 @@ jobs:
 
   run_past_ci_tensorflow_2-11:
     name: TensorFlow 2.11
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_pytorch_1-11]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 3 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.11"
@@ -69,9 +67,9 @@ jobs:
 
   run_past_ci_tensorflow_2-10:
     name: TensorFlow 2.10
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_tensorflow_2-11]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 4 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.10"
@@ -80,9 +78,9 @@ jobs:
 
   run_past_ci_tensorflow_2-9:
     name: TensorFlow 2.9
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_tensorflow_2-10]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 5 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.9"
@@ -91,9 +89,9 @@ jobs:
 
   run_past_ci_tensorflow_2-8:
     name: TensorFlow 2.8
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_tensorflow_2-9]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 6 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.8"
@@ -102,9 +100,9 @@ jobs:
 
   run_past_ci_tensorflow_2-7:
     name: TensorFlow 2.7
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_tensorflow_2-8]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 7 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.7"
@@ -113,9 +111,9 @@ jobs:
 
   run_past_ci_tensorflow_2-6:
     name: TensorFlow 2.6
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_tensorflow_2-7]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 8 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.6"
@@ -124,9 +122,9 @@ jobs:
 
   run_past_ci_tensorflow_2-5:
     name: TensorFlow 2.5
-    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    needs: [run_past_ci_tensorflow_2-6]
-    uses: ./.github/workflows/self-past.yml
+    needs: get_number
+    if: needs.get_number.outputs.run_number == 9 &&  (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.5"

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

@@ -1,290 +0,0 @@
-name: Self-hosted runner (nightly-ci)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  repository_dispatch:
-  workflow_call:
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-
-jobs:
-  setup:
-    name: Setup
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - id: set-matrix
-        name: Identify models to test
-        working-directory: /transformers/tests
-        run: |
-          echo "matrix=$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')" >> $GITHUB_OUTPUT
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_tests_multi_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-all-latest-torch-nightly-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_torch_cuda_extensions_gpu:
-    name: Torch CUDA extension tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    needs: setup
-    container:
-      image: huggingface/transformers-pytorch-deepspeed-nightly-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /workspace/transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /workspace/transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /workspace/transformers
-        run: |
-          python utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /workspace/transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /workspace/transformers
-        run: |
-          python -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_nightly"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_nightly
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-      setup,
-      run_tests_single_gpu,
-      run_tests_multi_gpu,
-      run_torch_cuda_extensions_gpu
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: Nightly CI
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
-
-
-      # delete-artifact
-      - uses: geekyeggo/delete-artifact@v2
-        with:
-          name: |
-              single-*
-              multi-*

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

@@ -0,0 +1,40 @@
+name: Self-hosted runner (past-ci)
+
+
+on:
+  workflow_call:
+    inputs:
+      framework:
+        required: true
+        type: string
+      version:
+        required: true
+        type: string
+      # Use this to control the commit to test against
+      sha:
+        default: 'main'
+        required: false
+        type: string
+
+jobs:
+  model-ci:
+    name: Model CI
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_models_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: past-ci
+      docker: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      ci_event: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+    secrets: inherit
+
+  deepspeed-ci:
+    name: DeepSpeed CI
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_torch_cuda_extensions_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: past-ci
+      docker: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      ci_event: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+    secrets: inherit

.github/workflows/self-past.yml

@@ -1,357 +0,0 @@
-name: Self-hosted runner (past-ci)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  workflow_call:
-    inputs:
-      framework:
-        required: true
-        type: string
-      version:
-        required: true
-        type: string
-      # Use this to control the commit to test against
-      sha:
-        default: 'main'
-        required: false
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-
-jobs:
-  setup:
-    name: Setup
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - id: set-matrix
-        working-directory: /transformers
-        name: Identify models to test
-        run: |
-          cd tests
-          echo "matrix=$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')" >> $GITHUB_OUTPUT
-
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - 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: Update some packages
-        working-directory: /transformers
-        run: python3 -m pip install -U datasets
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Install
-        if: inputs.framework == 'pytorch'
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Save job name
-        if: ${{ always() }}
-        shell: bash
-        run: |
-          matrix_folders=${matrix_folders/'models_'/'models/'}
-          job_name="Model tests ($matrix_folders, ${{ matrix.machine_type }})"
-          echo "$job_name"
-          echo "$job_name" > /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/job_name.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_tests_multi_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - 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: Update some packages
-        working-directory: /transformers
-        run: python3 -m pip install -U datasets
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Install
-        if: inputs.framework == 'pytorch'
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Save job name
-        if: ${{ always() }}
-        shell: bash
-        run: |
-          matrix_folders=${matrix_folders/'models_'/'models/'}
-          job_name="Model tests ($matrix_folders, ${{ matrix.machine_type }})"
-          echo "$job_name"
-          echo "$job_name" > /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/job_name.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
-          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
-
-  run_torch_cuda_extensions_gpu:
-    name: Torch CUDA extension tests
-    if: inputs.framework == 'pytorch'
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    needs: setup
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Update some packages
-        working-directory: /transformers
-        run: python3 -m pip install -U datasets
-
-      - name: Install
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
-          path: /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-      setup,
-      run_tests_single_gpu,
-      run_tests_multi_gpu,
-      run_torch_cuda_extensions_gpu
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-
-      # Create a directory to store test failure tables in the next step
-      - name: Create directory
-        run: mkdir test_failure_tables
-
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
-
-      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
-      - name: Failure table artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: test_failure_tables_${{ inputs.framework }}-${{ inputs.version }}
-          path: test_failure_tables
-
-      # delete-artifact
-      - uses: geekyeggo/delete-artifact@v2
-        with:
-          name: |
-              single-*
-              multi-*

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

@@ -16,6 +16,9 @@ jobs:
     with:
       job: run_models_gpu
       slack_report_channel: "#transformers-ci-daily-models"
+      runner: daily-ci
+      docker: huggingface/transformers-all-latest-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   torch-pipeline:
@@ -24,6 +27,9 @@ jobs:
     with:
       job: run_pipelines_torch_gpu
       slack_report_channel: "#transformers-ci-daily-pipeline-torch"
+      runner: daily-ci
+      docker: huggingface/transformers-pytorch-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   tf-pipeline:
@@ -32,6 +38,9 @@ jobs:
     with:
       job: run_pipelines_tf_gpu
       slack_report_channel: "#transformers-ci-daily-pipeline-tf"
+      runner: daily-ci
+      docker: huggingface/transformers-tensorflow-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   example-ci:
@@ -40,6 +49,9 @@ jobs:
     with:
       job: run_examples_gpu
       slack_report_channel: "#transformers-ci-daily-examples"
+      runner: daily-ci
+      docker: huggingface/transformers-all-latest-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   deepspeed-ci:
@@ -48,6 +60,10 @@ jobs:
     with:
       job: run_torch_cuda_extensions_gpu
       slack_report_channel: "#transformers-ci-daily-deepspeed"
+      runner: daily-ci
+      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      ci_event: Daily CI
+      working-directory-prefix: /workspace
     secrets: inherit
 
   quantization-ci:
@@ -56,4 +72,7 @@ jobs:
     with:
       job: run_quantization_torch_gpu
       slack_report_channel: "#transformers-ci-daily-quantization"
+      runner: daily-ci
+      docker: huggingface/transformers-quantization-latest-gpu
+      ci_event: Daily CI
     secrets: inherit

.github/workflows/self-scheduled.yml

@@ -15,6 +15,19 @@ on:
       slack_report_channel:
         required: true
         type: string
+      runner:
+        required: true
+        type: string
+      docker:
+        required: true
+        type: string
+      ci_event:
+        required: true
+        type: string
+      working-directory-prefix:
+        default: ''
+        required: false
+        type: string
 
 env:
   HF_HOME: /mnt/cache
@@ -38,7 +51,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -96,6 +109,8 @@ jobs:
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
       machine_type: ${{ matrix.machine_type }}
       slice_id: ${{ matrix.slice_id }}
+      runner: ${{ inputs.runner }}
+      docker: ${{ inputs.docker }}
     secrets: inherit
 
   run_pipelines_torch_gpu:
@@ -105,7 +120,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-pytorch-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -155,7 +170,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-tensorflow-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -206,7 +221,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -257,69 +272,88 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      image: ${{ inputs.docker }}
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Update clone
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: git fetch && git checkout ${{ github.sha }}
 
       - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
 
+      - name: Update / Install some packages (for Past CI)
+        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
+        run: |
+          python3 -m pip install -U datasets
+          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
+
       - name: Remove cached torch extensions
         run: rm -rf /github/home/.cache/torch_extensions/
 
       # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
+      - name: Pre build DeepSpeed *again* (for daily CI)
+        if: ${{ contains(inputs.ci_event, 'Daily CI') }}
+        working-directory: ${{ inputs.working-directory-prefix }}/
         run: |
           python3 -m pip uninstall -y deepspeed
           DS_DISABLE_NINJA=1 DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
 
+      # To avoid unknown test failures
+      - name: Pre build DeepSpeed *again* (for nightly & Past CI)
+        if: ${{ contains(inputs.ci_event, 'Nightly CI') || contains(inputs.ci_event, 'Past CI') }}
+        working-directory: ${{ inputs.working-directory-prefix }}/
+        run: |
+          python3 -m pip uninstall -y deepspeed
+          rm -rf DeepSpeed
+          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
+          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
       - name: Environment
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: |
-          python utils/print_env.py
+          python3 utils/print_env.py
 
       - name: Show installed libraries and their versions
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: pip freeze
 
       - name: Run all tests on GPU
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: |
-          python -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
+          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
+        run: cat ${{ inputs.working-directory-prefix }}/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
 
       - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
         if: ${{ always() }}
         uses: actions/upload-artifact@v4
         with:
           name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
+          path: ${{ inputs.working-directory-prefix }}/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
 
   run_quantization_torch_gpu:
     if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
     name: " "
     needs: setup
     strategy:
+      max-parallel: 4
       fail-fast: false
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.quantization_matrix) }}
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-quantization-latest-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -434,5 +468,6 @@ jobs:
       # This would be an empty string if `setup` is skipped.
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
       quantization_matrix: ${{ needs.setup.outputs.quantization_matrix }}
-      
+      ci_event: ${{ inputs.ci_event }}
+
     secrets: inherit

.github/workflows/slack-report.yml

@@ -18,6 +18,9 @@ on:
       quantization_matrix:
         required: true
         type: string
+      ci_event:
+        required: true
+        type: string
 
 env:
   TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN: ${{ secrets.TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN }}
@@ -45,7 +48,7 @@ jobs:
           CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
           SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: scheduled
+          CI_EVENT: ${{ inputs.ci_event }}
           CI_SHA: ${{ github.sha }}
           CI_WORKFLOW_REF: ${{ github.workflow_ref }}
           CI_TEST_JOB: ${{ inputs.job }}
@@ -76,7 +79,7 @@ jobs:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
           SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
-          CI_EVENT: scheduled
+          CI_EVENT: ${{ inputs.ci_event }}
           CI_SHA: ${{ github.sha }}
           CI_TEST_JOB: ${{ inputs.job }}
           SETUP_STATUS: ${{ inputs.setup_status }}

README.md

@@ -36,18 +36,18 @@ limitations under the License.
 <h4 align="center">
     <p>
         <b>English</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_pt-br.md">Рortuguês</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_te.md">తెలుగు</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_fr.md">Français</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_de.md">Deutsch</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_vi.md">Tiếng Việt</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_ko.md">한국어</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_es.md">Español</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_ja.md">日本語</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_hd.md">हिन्दी</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_ru.md">Русский</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_pt-br.md">Рortuguês</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_te.md">తెలుగు</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_fr.md">Français</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_de.md">Deutsch</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_vi.md">Tiếng Việt</a> |
     </p>
 </h4>
 

awesome-transformers.md

@@ -596,7 +596,7 @@ Keywords: Data-Centric AI, Data Quality, Noisy Labels, Outlier Detection, Active
 
 ## [BentoML](https://github.com/bentoml/BentoML)
 
-[BentoML](https://github.com/bentoml) is the unified framework for for building, shipping, and scaling production-ready AI applications incorporating traditional ML, pre-trained AI models, Generative and Large Language Models. 
+[BentoML](https://github.com/bentoml) is the unified framework for building, shipping, and scaling production-ready AI applications incorporating traditional ML, pre-trained AI models, Generative and Large Language Models. 
 All Hugging Face models and pipelines can be seamlessly integrated into BentoML applications, enabling the running of models on the most suitable hardware and independent scaling based on usage.
 
 Keywords: BentoML, Framework, Deployment, AI Applications

docker/consistency.dockerfile

@@ -6,10 +6,10 @@ RUN apt-get update && apt-get install -y time git pkg-config make git-lfs
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools GitPython
 RUN uv pip install --no-cache-dir --upgrade 'torch' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-cache-dir tensorflow-cpu tf-keras
-RUN uv pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,quality,vision,testing]" 
+# tensorflow pin matching setup.py
+RUN uv pip install --no-cache-dir "tensorflow-cpu<2.16" "tf-keras<2.16"
+RUN uv pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,quality,vision,testing]"
 RUN git lfs install
 
 RUN pip uninstall -y transformers
 RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean
-

docs/source/de/testing.md

@@ -185,16 +185,16 @@ pytest -k "test and ada" tests/test_optimization.py
 
 Manchmal müssen Sie `accelerate` Tests für Ihre Modelle ausführen. Dazu fügen Sie einfach `-m accelerate_tests` zu Ihrem Befehl hinzu, wenn Sie diese Tests bei einem `OPT`-Lauf ausführen möchten:
 ```bash
-RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py 
+RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py
 ```
 
 
-### Dokumentationstests ausführen 
+### Dokumentationstests ausführen
 
-Um zu testen, ob die Dokumentationsbeispiele korrekt sind, sollten Sie überprüfen, ob die `doctests` erfolgreich sind. 
-Lassen Sie uns als Beispiel den docstring von [WhisperModel.forward](https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py#L1017-L1035) verwenden: 
+Um zu testen, ob die Dokumentationsbeispiele korrekt sind, sollten Sie überprüfen, ob die `doctests` erfolgreich sind.
+Lassen Sie uns als Beispiel den docstring von [WhisperModel.forward](https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py#L1017-L1035) verwenden:
 
-```python 
+```python
 r"""
 Returns:
 
@@ -217,8 +217,8 @@ Example:
 
 ```
 
-Führen Sie einfach die folgende Zeile aus, um automatisch jedes docstring-Beispiel in der gewünschten Datei zu testen: 
-```bash 
+Führen Sie einfach die folgende Zeile aus, um automatisch jedes docstring-Beispiel in der gewünschten Datei zu testen:
+```bash
 pytest --doctest-modules <path_to_file_or_dir>
 ```
 Wenn die Datei eine Markdown-Erweiterung hat, sollten Sie das Argument `--doctest-glob="*.md"` hinzufügen.
@@ -862,7 +862,7 @@ Code, der fehlerhaft ist, einen schlechten Zustand verursacht, der sich auf ande
 - Hier sehen Sie, wie Sie einen ganzen Test bedingungslos überspringen können:
 
 ```python no-style
-@unittest.skip("this bug needs to be fixed")
+@unittest.skip(reason="this bug needs to be fixed")
 def test_feature_x():
 ```
 

docs/source/en/_toctree.yml

@@ -382,6 +382,8 @@
         title: Fuyu
       - local: model_doc/gemma
         title: Gemma
+      - local: model_doc/gemma2
+        title: Gemma2
       - local: model_doc/openai-gpt
         title: GPT
       - local: model_doc/gpt_neo
@@ -579,6 +581,8 @@
         title: DeiT
       - local: model_doc/depth_anything
         title: Depth Anything
+      - local: model_doc/depth_anything_v2
+        title: Depth Anything V2
       - local: model_doc/deta
         title: DETA
       - local: model_doc/detr
@@ -599,6 +603,8 @@
         title: FocalNet
       - local: model_doc/glpn
         title: GLPN
+      - local: model_doc/hiera
+        title: Hiera
       - local: model_doc/imagegpt
         title: ImageGPT
       - local: model_doc/levit
@@ -627,6 +633,8 @@
         title: RegNet
       - local: model_doc/resnet
         title: ResNet
+      - local: model_doc/rt_detr
+        title: RT-DETR
       - local: model_doc/segformer
         title: SegFormer
       - local: model_doc/seggpt
@@ -661,6 +669,8 @@
         title: ViTMSN
       - local: model_doc/yolos
         title: YOLOS
+      - local: model_doc/zoedepth
+        title: ZoeDepth
       title: Vision models
     - isExpanded: false
       sections:
@@ -672,6 +682,8 @@
         title: CLAP
       - local: model_doc/encodec
         title: EnCodec
+      - local: model_doc/hiera
+        title: Hiera
       - local: model_doc/hubert
         title: Hubert
       - local: model_doc/mctct
@@ -774,6 +786,8 @@
         title: Idefics2
       - local: model_doc/instructblip
         title: InstructBLIP
+      - local: model_doc/instructblipvideo
+        title: InstructBlipVideo
       - local: model_doc/kosmos-2
         title: KOSMOS-2
       - local: model_doc/layoutlm
@@ -790,6 +804,8 @@
         title: Llava
       - local: model_doc/llava_next
         title: LLaVA-NeXT
+      - local: model_doc/llava-next-video
+        title: LLaVa-NeXT-Video
       - local: model_doc/lxmert
         title: LXMERT
       - local: model_doc/matcha

docs/source/en/agents.md

@@ -50,7 +50,7 @@ We implement two versions of ReactJsonAgent:
 
 ![Framework of a React Agent](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/open-source-llms-as-agents/ReAct.png)
 
-For example, here is how a ReAct agent would work its way through the following question.
+For example, here is how a ReAct Code agent would work its way through the following question.
 
 ```py3
 >>> agent.run(
@@ -188,7 +188,7 @@ You can still authorize additional imports by passing the authorized modules as
 >>> from transformers import ReactCodeAgent
 
 >>> agent = ReactCodeAgent(tools=[], additional_authorized_imports=['requests', 'bs4'])
->>>agent.run("Could you get me the title of the page at url 'https://huggingface.co/blog'?")
+>>> agent.run("Could you get me the title of the page at url 'https://huggingface.co/blog'?")
 
 (...)
 'Hugging Face – Blog'
@@ -256,6 +256,13 @@ agent = ReactJsonAgent(tools=[PythonInterpreterTool()], system_prompt="{your_cus
 > Please make sure to define the `<<tool_descriptions>>` string somewhere in the `template` so the agent is aware 
 of the available tools.
 
+
+### Inspecting an agent run
+
+Here are a few useful attributes to inspect what happened after a run:
+- `agent.logs` stores the fine-grained logs of the agent. At every step of the agent's run, everything gets stored in a dictionary that then is appended to `agent.logs`.
+- Running `agent.write_inner_memory_from_logs()` creates an inner memory of the agent's logs for the LLM to view, as a list of chat messages. This method goes over each step of the log and only stores what it's interested in as a message: for instance, it will save the system prompt and task in separate messages, then for each step it will store the LLM output as a message, and the tool call output as another message. Use this if you want a higher-level view of what has happened - but not every log will be transcripted by this method.
+
 ## Tools
 
 A tool is an atomic function to be used by an agent.
@@ -379,7 +386,7 @@ And the output:
 `"The most downloaded model for the 'text-to-video' task is ByteDance/AnimateDiff-Lightning."`
 
 
-### Manage agent toolbox
+### Manage your agent's toolbox
 
 If you have already initialized an agent, it is inconvenient to reinitialize it from scratch with a tool you want to use. With Transformers, you can manage an agent's toolbox by adding or replacing a tool.
 

docs/source/en/chat_templating.md

@@ -199,7 +199,8 @@ effect that `add_generation_prompt` has will depend on the template being used.
 
 ## Can I use chat templates in training?
 
-Yes! We recommend that you apply the chat template as a preprocessing step for your dataset. After this, you
+Yes! This is a good way to ensure that the chat template matches the tokens the model sees during training.
+We recommend that you apply the chat template as a preprocessing step for your dataset. After this, you
 can simply continue like any other language model training task. When training, you should usually set 
 `add_generation_prompt=False`, because the added tokens to prompt an assistant response will not be helpful during 
 training. Let's see an example:
@@ -233,6 +234,16 @@ The sun.</s>
 
 From here, just continue training like you would with a standard language modelling task, using the `formatted_chat` column.
 
+<Tip>
+If you format text with `apply_chat_template(tokenize=False)` and then tokenize it in a separate step, you should set the argument
+`add_special_tokens=False`. If you use `apply_chat_template(tokenize=True)`, you don't need to worry about this!
+
+By default, some tokenizers add special tokens like `<bos>` and `<eos>` to text they tokenize. Chat templates should 
+always include all of the special tokens they need, and so adding extra special tokens with
+the default `add_special_tokens=True` can result in incorrect or duplicated special tokens, which will hurt model
+performance.
+</Tip>
+
 ## Advanced: Extra inputs to chat templates
 
 The only argument that `apply_chat_template` requires is `messages`. However, you can pass any keyword

docs/source/en/deepspeed.md

@@ -16,11 +16,11 @@ rendered properly in your Markdown viewer.
 
 # DeepSpeed
 
-[DeepSpeed](https://www.deepspeed.ai/) is a PyTorch optimization library that makes distributed training memory-efficient and fast. At it's core is the [Zero Redundancy Optimizer (ZeRO)](https://hf.co/papers/1910.02054) which enables training large models at scale. ZeRO works in several stages:
+[DeepSpeed](https://www.deepspeed.ai/) is a PyTorch optimization library that makes distributed training memory-efficient and fast. At its core is the [Zero Redundancy Optimizer (ZeRO)](https://hf.co/papers/1910.02054) which enables training large models at scale. ZeRO works in several stages:
 
-* ZeRO-1, optimizer state partioning across GPUs
+* ZeRO-1, optimizer state partitioning across GPUs
 * ZeRO-2, gradient partitioning across GPUs
-* ZeRO-3, parameteter partitioning across GPUs
+* ZeRO-3, parameter partitioning across GPUs
 
 In GPU-limited environments, ZeRO also enables offloading optimizer memory and computation from the GPU to the CPU to fit and train really large models on a single GPU. DeepSpeed is integrated with the Transformers [`Trainer`] class for all ZeRO stages and offloading. All you need to do is provide a config file or you can use a provided template. For inference, Transformers support ZeRO-3 and offloading since it allows loading huge models.
 
@@ -159,7 +159,7 @@ There are three types of configuration parameters:
 
 You could also modify the DeepSpeed configuration and edit [`TrainingArguments`] from it:
 
-1. Create or load a DeepSpeed configuration to used as the main configuration
+1. Create or load a DeepSpeed configuration to use as the main configuration
 2. Create a [`TrainingArguments`] object based on these DeepSpeed configuration values
 
 Some values, such as `scheduler.params.total_num_steps` are calculated by the [`Trainer`] during training.
@@ -191,7 +191,7 @@ ZeRO-1 shards the optimizer states across GPUs, and you can expect a tiny speed
 </hfoption>
 <hfoption id="ZeRO-2">
 
-ZeRO-2 shards the optimizer and gradients across GPUs. This stage is primarily used for training since it's features are not relevant to inference. Some important parameters to configure for better performance include:
+ZeRO-2 shards the optimizer and gradients across GPUs. This stage is primarily used for training since its features are not relevant to inference. Some important parameters to configure for better performance include:
 
 * `offload_optimizer` should be enabled to reduce GPU memory usage.
 * `overlap_comm` when set to `true` trades off increased GPU memory usage to lower allreduce latency. This feature uses 4.5x the `allgather_bucket_size` and `reduce_bucket_size` values. In this example, they're set to `5e8` which means it requires 9GB of GPU memory. If your GPU memory is 8GB or less, you should reduce `overlap_comm` to lower the memory requirements and prevent an out-of-memory (OOM) error.
@@ -226,7 +226,7 @@ ZeRO-3 shards the optimizer, gradient, and parameters across GPUs. Unlike ZeRO-2
 * `pin_memory: true` can improve throughput, but less memory becomes available for other processes because the pinned memory is reserved for the specific process that requested it and it's typically accessed much faster than normal CPU memory.
 * `stage3_max_live_parameters` is the upper limit on how many full parameters you want to keep on the GPU at any given time. Reduce this value if you encounter an OOM error.
 * `stage3_max_reuse_distance` is a value for determining when a parameter is used again in the future, and it helps decide whether to throw the parameter away or to keep it. If the parameter is going to be reused (if the value is less than `stage3_max_reuse_distance`), then it is kept to reduce communication overhead. This is super helpful when activation checkpointing is enabled and you want to keep the parameter in the forward recompute until the backward pass. But reduce this value if you encounter an OOM error.
-* `stage3_gather_16bit_weights_on_model_save` consolidates fp16 weights when a model is saved. For large models and multiple GPUs, this is an expensive in terms of memory and speed. You should enable it if you're planning on resuming training.
+* `stage3_gather_16bit_weights_on_model_save` consolidates fp16 weights when a model is saved. For large models and multiple GPUs, this is expensive in terms of memory and speed. You should enable it if you're planning on resuming training.
 * `sub_group_size` controls which parameters are updated during the optimizer step. Parameters are grouped into buckets of `sub_group_size` and each bucket is updated one at a time. When used with NVMe offload, `sub_group_size` determines when model states are moved in and out of CPU memory from during the optimization step. This prevents running out of CPU memory for extremely large models. `sub_group_size` can be left to its default value if you aren't using NVMe offload, but you may want to change it if you:
 
     1. Run into an OOM error during the optimizer step. In this case, reduce `sub_group_size` to reduce memory usage of the temporary buffers.

docs/source/en/generation_strategies.md

@@ -178,7 +178,7 @@ An increasing sequence: one, two, three, four, five, six, seven, eight, nine, te
 
 The `generate()` method supports caching keys and values to enhance efficiency and avoid re-computations. However the key and value
 cache can occupy a large portion of memory, becoming a bottleneck for long-context generation, especially for Large Language Models.
-Quantizing the cache when using `generate()` can significantly reduce memory requirements at the cost of speed. 
+Quantizing the cache when using `generate()` can significantly reduce memory requirements at the cost of speed.
 
 KV Cache quantization in `transformers` is largely inspired by the paper [KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache]
 (https://arxiv.org/abs/2402.02750) and currently supports `quanto` and `HQQ` as backends. For more information on the inner workings see the paper.
@@ -213,11 +213,11 @@ I like rock music because it's loud and energetic. I like to listen to it when I
 
 ## Watermarking
 
-The `generate()` supports watermarking the generated text by randomly marking a portion of tokens as "green". 
+The `generate()` supports watermarking the generated text by randomly marking a portion of tokens as "green".
 When generating the "green" will have a small 'bias' value added to their logits, thus having a higher chance to be generated.
 The watermarked text can be detected by calculating the proportion of "green" tokens in the text and estimating how likely it is
-statistically to obtain that amount of "green" tokens for human-generated text. This watermarking strategy was proposed in the paper 
-["On the Reliability of Watermarks for Large Language Models"](https://arxiv.org/abs/2306.04634). For more information on 
+statistically to obtain that amount of "green" tokens for human-generated text. This watermarking strategy was proposed in the paper
+["On the Reliability of Watermarks for Large Language Models"](https://arxiv.org/abs/2306.04634). For more information on
 the inner functioning of watermarking, it is recommended to refer to the paper.
 
 The watermarking can be used with any generative model in `tranformers` and does not require an extra classification model
@@ -484,3 +484,59 @@ just like in multinomial sampling. However, in assisted decoding, reducing the t
 
 Alternativelly, you can also set the `prompt_lookup_num_tokens` to trigger n-gram based assisted decoding, as opposed
 to model based assisted decoding. You can read more about it [here](https://twitter.com/joao_gante/status/1747322413006643259).
+### DoLa Decoding
+
+**D**ecoding by C**o**ntrasting **La**yers (DoLa) is a contrastive decoding strategy to improve the factuality and reduce the
+hallucinations of LLMs, as described in this paper of ICLR 2024 [DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models](https://arxiv.org/abs/2309.03883).
+
+DoLa is achieved by contrasting the differences in logits obtained from final
+layers versus earlier layers, thus amplify the factual knowledge localized to particular part of transformer layers.
+
+Do the following two steps to activate DoLa decoding when calling the `model.generate` function:
+1. Set the `dola_layers` argument, which can be either a string or a list of integers.
+    - If set to a string, it can be one of `low`, `high`.
+    - If set to a list of integers, it should be a list of layer indices between 0 and the total number of layers in the model. The 0-th layer is word embedding, and the 1st layer is the first transformer layer, and so on.
+2. Set `repetition_penalty = 1.2` is suggested to reduce repetition in DoLa decoding.
+
+See the following examples for DoLa decoding with the 32-layer LLaMA-7B model.
+
+```python
+>>> from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
+>>> import torch
+
+>>> tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
+>>> model = AutoModelForCausalLM.from_pretrained("huggyllama/llama-7b", torch_dtype=torch.float16)
+>>> device = 'cuda' if torch.cuda.is_available() else 'cpu'
+>>> model.to(device)
+>>> set_seed(42)
+
+>>> text = "On what date was the Declaration of Independence officially signed?"
+>>> inputs = tokenizer(text, return_tensors="pt").to(device)
+
+# Vanilla greddy decoding
+>>> vanilla_output = model.generate(**inputs, do_sample=False, max_new_tokens=50)
+>>> tokenizer.batch_decode(vanilla_output[:, inputs.input_ids.shape[-1]:], skip_special_tokens=True)
+['\nThe Declaration of Independence was signed on July 4, 1776.\nWhat was the date of the signing of the Declaration of Independence?\nThe Declaration of Independence was signed on July 4,']
+
+# DoLa decoding with contrasting higher part of layers (layers 16,18,...,30)
+>>> dola_high_output = model.generate(**inputs, do_sample=False, max_new_tokens=50, dola_layers='high')
+>>> tokenizer.batch_decode(dola_high_output[:, inputs.input_ids.shape[-1]:], skip_special_tokens=True)
+['\nJuly 4, 1776, when the Continental Congress voted to separate from Great Britain. The 56 delegates to the Continental Congress signed the Declaration on August 2, 1776.']
+
+# DoLa decoding with contrasting specific layers (layers 28 and 30)
+>>> dola_custom_output = model.generate(**inputs, do_sample=False, max_new_tokens=50, dola_layers=[28,30], repetition_penalty=1.2)
+>>> tokenizer.batch_decode(dola_custom_output[:, inputs.input_ids.shape[-1]:], skip_special_tokens=True)
+['\nIt was officially signed on 2 August 1776, when 56 members of the Second Continental Congress, representing the original 13 American colonies, voted unanimously for the resolution for independence. The 2']
+```
+
+#### Understanding the `dola_layers` argument
+
+`dola_layers` stands for the candidate layers in premature layer selection, as described in the DoLa paper. The selected premature layer will be contrasted with the final layer.
+
+Setting `dola_layers` to `'low'` or `'high'` will select the lower or higher part of the layers to contrast, respectively.
+- For `N`-layer models with `N <= 40` layers, the layers of `range(0, N // 2, 2)` and `range(N // 2, N, 2)` are used for `'low'` and `'high'` layers, respectively.
+- For models with `N > 40` layers, the layers of `range(0, 20, 2)` and `range(N - 20, N, 2)` are used for `'low'` and `'high'` layers, respectively.
+- If the model has tied word embeddings, we skip the word embeddings (0-th) layer and start from the 2nd layer, as the early exit from word embeddings will become identity function.
+- Set the `dola_layers` to a list of integers for layer indices to contrast manually specified layers. For example, setting `dola_layers=[28,30]` will contrast the final layer (32-th layer) with the 28-th and 30-th layers.
+
+The paper suggested that contrasting `'high'` layers to improve short-answer tasks like TruthfulQA, and contrasting `'low'` layers to improve all the other long-answer reasoning tasks, such as GSM8K, StrategyQA, FACTOR, and VicunaQA. Applying DoLa to smaller models like GPT-2 is not recommended, as the results shown in the Appendix N of the paper.

docs/source/en/glossary.md

@@ -139,7 +139,7 @@ reading the whole sentence with a mask to hide future tokens at a certain timest
 
 ### deep learning (DL)
 
-Machine learning algorithms which uses neural networks with several layers.
+Machine learning algorithms which use neural networks with several layers.
 
 ## E
 
@@ -519,4 +519,4 @@ A form of model training in which data provided to the model is not labeled. Uns
 Parallelism technique which performs sharding of the tensors somewhat similar to [TensorParallel](#tensor-parallelism-tp), 
 except the whole tensor gets reconstructed in time for a forward or backward computation, therefore the model doesn't need 
 to be modified. This method also supports various offloading techniques to compensate for limited GPU memory. 
-Learn more about ZeRO [here](perf_train_gpu_many#zero-data-parallelism).
+Learn more about ZeRO [here](perf_train_gpu_many#zero-data-parallelism).

docs/source/en/index.md

@@ -145,6 +145,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                  [Funnel Transformer](model_doc/funnel)                  |       ✅        |         ✅         |      ❌      |
 |                          [Fuyu](model_doc/fuyu)                          |       ✅        |         ❌         |      ❌      |
 |                         [Gemma](model_doc/gemma)                         |       ✅        |         ❌         |      ✅      |
+|                        [Gemma2](model_doc/gemma2)                        |       ✅        |         ❌         |      ❌      |
 |                           [GIT](model_doc/git)                           |       ✅        |         ❌         |      ❌      |
 |                          [GLPN](model_doc/glpn)                          |       ✅        |         ❌         |      ❌      |
 |                       [GPT Neo](model_doc/gpt_neo)                       |       ✅        |         ❌         |      ✅      |
@@ -158,6 +159,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                [Grounding DINO](model_doc/grounding-dino)                |       ✅        |         ❌         |      ❌      |
 |                      [GroupViT](model_doc/groupvit)                      |       ✅        |         ✅         |      ❌      |
 |                       [HerBERT](model_doc/herbert)                       |       ✅        |         ✅         |      ✅      |
+|                         [Hiera](model_doc/hiera)                         |       ✅        |         ❌         |      ❌      |
 |                        [Hubert](model_doc/hubert)                        |       ✅        |         ✅         |      ❌      |
 |                        [I-BERT](model_doc/ibert)                         |       ✅        |         ❌         |      ❌      |
 |                       [IDEFICS](model_doc/idefics)                       |       ✅        |         ✅         |      ❌      |
@@ -165,6 +167,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                      [ImageGPT](model_doc/imagegpt)                      |       ✅        |         ❌         |      ❌      |
 |                      [Informer](model_doc/informer)                      |       ✅        |         ❌         |      ❌      |
 |                  [InstructBLIP](model_doc/instructblip)                  |       ✅        |         ❌         |      ❌      |
+|             [InstructBlipVideo](model_doc/instructblipvideo)             |       ✅        |         ❌         |      ❌      |
 |                         [Jamba](model_doc/jamba)                         |       ✅        |         ❌         |      ❌      |
 |                        [JetMoe](model_doc/jetmoe)                        |       ✅        |         ❌         |      ❌      |
 |                       [Jukebox](model_doc/jukebox)                       |       ✅        |         ❌         |      ❌      |
@@ -181,6 +184,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                        [Llama3](model_doc/llama3)                        |       ✅        |         ❌         |      ✅      |
 |                         [LLaVa](model_doc/llava)                         |       ✅        |         ❌         |      ❌      |
 |                    [LLaVA-NeXT](model_doc/llava_next)                    |       ✅        |         ❌         |      ❌      |
+|              [LLaVa-NeXT-Video](model_doc/llava-next-video)              |       ✅        |         ❌         |      ❌      |
 |                    [Longformer](model_doc/longformer)                    |       ✅        |         ✅         |      ❌      |
 |                        [LongT5](model_doc/longt5)                        |       ✅        |         ❌         |      ✅      |
 |                          [LUKE](model_doc/luke)                          |       ✅        |         ❌         |      ❌      |
@@ -262,6 +266,8 @@ Flax), PyTorch, and/or TensorFlow.
 |          [RoBERTa-PreLayerNorm](model_doc/roberta-prelayernorm)          |       ✅        |         ✅         |      ✅      |
 |                      [RoCBert](model_doc/roc_bert)                       |       ✅        |         ❌         |      ❌      |
 |                      [RoFormer](model_doc/roformer)                      |       ✅        |         ✅         |      ✅      |
+|                       [RT-DETR](model_doc/rt_detr)                       |       ✅        |         ❌         |      ❌      |
+|                [RT-DETR-ResNet](model_doc/rt_detr_resnet)                |       ✅        |         ❌         |      ❌      |
 |                          [RWKV](model_doc/rwkv)                          |       ✅        |         ❌         |      ❌      |
 |                           [SAM](model_doc/sam)                           |       ✅        |         ✅         |      ❌      |
 |                  [SeamlessM4T](model_doc/seamless_m4t)                   |       ✅        |         ❌         |      ❌      |
@@ -338,5 +344,6 @@ Flax), PyTorch, and/or TensorFlow.
 |                 [XLSR-Wav2Vec2](model_doc/xlsr_wav2vec2)                 |       ✅        |         ✅         |      ✅      |
 |                         [YOLOS](model_doc/yolos)                         |       ✅        |         ❌         |      ❌      |
 |                          [YOSO](model_doc/yoso)                          |       ✅        |         ❌         |      ❌      |
+|                      [ZoeDepth](model_doc/zoedepth)                      |       ✅        |         ❌         |      ❌      |
 
 <!-- End table-->

docs/source/en/internal/generation_utils.md

@@ -391,6 +391,12 @@ A [`Constraint`] can be used to force the generation to include specific tokens
     - get_seq_length
     - reset
 
+[[autodoc]] EncoderDecoderCache
+    - get_seq_length
+    - to_legacy_cache
+    - from_legacy_cache
+    - reset
+    - reorder_cache
 
 ## Watermark Utils
 

docs/source/en/llm_tutorial_optimization.md

@@ -147,7 +147,7 @@ Let's call it now for the next experiment.
 ```python
 flush()
 ```
-In the recent version of the accelerate library, you can also use an utility method called `release_memory()`
+In the recent version of the accelerate library, you can also use a utility method called `release_memory()`
 
 ```python
 from accelerate.utils import release_memory
@@ -683,7 +683,7 @@ Assistant: Germany has ca. 81 million inhabitants
 
 In this chat, the LLM runs auto-regressive decoding twice:
   1. The first time, the key-value cache is empty and the input prompt is `"User: How many people live in France?"` and the model auto-regressively generates the text `"Roughly 75 million people live in France"` while increasing the key-value cache at every decoding step.
-  2. The second time the input prompt is `"User: How many people live in France? \n Assistant: Roughly 75 million people live in France \n User: And how many in Germany?"`. Thanks to the cache, all key-value vectors for the first two sentences are already computed. Therefore the input prompt only consists of `"User: And how many in Germany?"`. While processing the shortened input prompt, it's computed key-value vectors are concatenated to the key-value cache of the first decoding. The second Assistant's answer `"Germany has ca. 81 million inhabitants"` is then auto-regressively generated with the key-value cache consisting of encoded key-value vectors of `"User: How many people live in France? \n Assistant: Roughly 75 million people live in France \n User: And how many are in Germany?"`.
+  2. The second time the input prompt is `"User: How many people live in France? \n Assistant: Roughly 75 million people live in France \n User: And how many in Germany?"`. Thanks to the cache, all key-value vectors for the first two sentences are already computed. Therefore the input prompt only consists of `"User: And how many in Germany?"`. While processing the shortened input prompt, its computed key-value vectors are concatenated to the key-value cache of the first decoding. The second Assistant's answer `"Germany has ca. 81 million inhabitants"` is then auto-regressively generated with the key-value cache consisting of encoded key-value vectors of `"User: How many people live in France? \n Assistant: Roughly 75 million people live in France \n User: And how many are in Germany?"`.
 
 Two things should be noted here:
   1. Keeping all the context is crucial for LLMs deployed in chat so that the LLM understands all the previous context of the conversation. E.g. for the example above the LLM needs to understand that the user refers to the population when asking `"And how many are in Germany"`.

docs/source/en/main_classes/callback.md

@@ -34,7 +34,7 @@ By default, `TrainingArguments.report_to` is set to `"all"`, so a [`Trainer`] wi
 - [`~integrations.TensorBoardCallback`] if tensorboard is accessible (either through PyTorch >= 1.4
   or tensorboardX).
 - [`~integrations.WandbCallback`] if [wandb](https://www.wandb.com/) is installed.
-- [`~integrations.CometCallback`] if [comet_ml](https://www.comet.ml/site/) is installed.
+- [`~integrations.CometCallback`] if [comet_ml](https://www.comet.com/site/) is installed.
 - [`~integrations.MLflowCallback`] if [mlflow](https://www.mlflow.org/) is installed.
 - [`~integrations.NeptuneCallback`] if [neptune](https://neptune.ai/) is installed.
 - [`~integrations.AzureMLCallback`] if [azureml-sdk](https://pypi.org/project/azureml-sdk/) is

docs/source/en/main_classes/model.md

@@ -40,6 +40,10 @@ for text generation, [`~generation.GenerationMixin`] (for the PyTorch models),
     - push_to_hub
     - all
 
+Custom models should also include a `_supports_assign_param_buffer`, which determines if superfast init can apply
+on the particular model. Signs that your model needs this are if `test_save_and_load_from_pretrained` fails. If so,
+set this to `False`.
+
 ## ModuleUtilsMixin
 
 [[autodoc]] modeling_utils.ModuleUtilsMixin

docs/source/en/main_classes/pipelines.md

@@ -270,6 +270,11 @@ This is a simplified view, since the pipeline can handle automatically the batch
 about how many forward passes you inputs are actually going to trigger, you can optimize the `batch_size`
 independently of the inputs. The caveats from the previous section still apply.
 
+## Pipeline FP16 inference
+Models can be run in FP16 which can be significantly faster on GPU while saving memory. Most models will not suffer noticeable performance loss from this. The larger the model, the less likely that it will.
+
+To enable FP16 inference, you can simply pass `torch_dtype=torch.float16` or `torch_dtype='float16'` to the pipeline constructor. Note that this only works for models with a PyTorch backend. Your inputs will be converted to FP16 internally.
+
 ## Pipeline custom code
 
 If you want to override a specific pipeline.

docs/source/en/model_doc/depth_anything.md

@@ -20,6 +20,12 @@ rendered properly in your Markdown viewer.
 
 The Depth Anything model was proposed in [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) by Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao. Depth Anything is based on the [DPT](dpt) architecture, trained on ~62 million images, obtaining state-of-the-art results for both relative and absolute depth estimation.
 
+<Tip>
+
+[Depth Anything V2](depth_anything_v2) was released in June 2024. It uses the same architecture as Depth Anything and therefore it is compatible with all code examples and existing workflows. However, it leverages synthetic data and a larger capacity teacher model to achieve much finer and robust depth predictions.
+
+</Tip>
+
 The abstract from the paper is the following:
 
 *This work presents Depth Anything, a highly practical solution for robust monocular depth estimation. Without pursuing novel technical modules, we aim to build a simple yet powerful foundation model dealing with any images under any circumstances. To this end, we scale up the dataset by designing a data engine to collect and automatically annotate large-scale unlabeled data (~62M), which significantly enlarges the data coverage and thus is able to reduce the generalization error. We investigate two simple yet effective strategies that make data scaling-up promising. First, a more challenging optimization target is created by leveraging data augmentation tools. It compels the model to actively seek extra visual knowledge and acquire robust representations. Second, an auxiliary supervision is developed to enforce the model to inherit rich semantic priors from pre-trained encoders. We evaluate its zero-shot capabilities extensively, including six public datasets and randomly captured photos. It demonstrates impressive generalization ability. Further, through fine-tuning it with metric depth information from NYUv2 and KITTI, new SOTAs are set. Our better depth model also results in a better depth-conditioned ControlNet.*

docs/source/en/model_doc/depth_anything_v2.md

@@ -0,0 +1,115 @@
+<!--Copyright 2024 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.
+
+-->
+
+# Depth Anything V2
+
+## Overview
+
+Depth Anything V2 was introduced in [the paper of the same name](https://arxiv.org/abs/2406.09414) by Lihe Yang et al. It uses the same architecture as the original [Depth Anything model](depth_anything), but uses synthetic data and a larger capacity teacher model to achieve much finer and robust depth predictions.
+
+The abstract from the paper is the following:
+
+*This work presents Depth Anything V2. Without pursuing fancy techniques, we aim to reveal crucial findings to pave the way towards building a powerful monocular depth estimation model. Notably, compared with V1, this version produces much finer and more robust depth predictions through three key practices: 1) replacing all labeled real images with synthetic images, 2) scaling up the capacity of our teacher model, and 3) teaching student models via the bridge of large-scale pseudo-labeled real images. Compared with the latest models built on Stable Diffusion, our models are significantly more efficient (more than 10x faster) and more accurate. We offer models of different scales (ranging from 25M to 1.3B params) to support extensive scenarios. Benefiting from their strong generalization capability, we fine-tune them with metric depth labels to obtain our metric depth models. In addition to our models, considering the limited diversity and frequent noise in current test sets, we construct a versatile evaluation benchmark with precise annotations and diverse scenes to facilitate future research.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/depth_anything_overview.jpg"
+alt="drawing" width="600"/>
+
+<small> Depth Anything overview. Taken from the <a href="https://arxiv.org/abs/2401.10891">original paper</a>.</small>
+
+The Depth Anything models were contributed by [nielsr](https://huggingface.co/nielsr).
+The original code can be found [here](https://github.com/DepthAnything/Depth-Anything-V2).
+
+## Usage example
+
+There are 2 main ways to use Depth Anything V2: either using the pipeline API, which abstracts away all the complexity for you, or by using the `DepthAnythingForDepthEstimation` class yourself.
+
+### Pipeline API
+
+The pipeline allows to use the model in a few lines of code:
+
+```python
+>>> from transformers import pipeline
+>>> from PIL import Image
+>>> import requests
+
+>>> # load pipe
+>>> pipe = pipeline(task="depth-estimation", model="depth-anything/Depth-Anything-V2-Small-hf")
+
+>>> # load image
+>>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
+>>> image = Image.open(requests.get(url, stream=True).raw)
+
+>>> # inference
+>>> depth = pipe(image)["depth"]
+```
+
+### Using the model yourself
+
+If you want to do the pre- and post-processing yourself, here's how to do that:
+
+```python
+>>> from transformers import AutoImageProcessor, AutoModelForDepthEstimation
+>>> import torch
+>>> import numpy as np
+>>> from PIL import Image
+>>> import requests
+
+>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+>>> image = Image.open(requests.get(url, stream=True).raw)
+
+>>> image_processor = AutoImageProcessor.from_pretrained("depth-anything/Depth-Anything-V2-Small-hf")
+>>> model = AutoModelForDepthEstimation.from_pretrained("depth-anything/Depth-Anything-V2-Small-hf")
+
+>>> # prepare image for the model
+>>> inputs = image_processor(images=image, return_tensors="pt")
+
+>>> with torch.no_grad():
+...     outputs = model(**inputs)
+...     predicted_depth = outputs.predicted_depth
+
+>>> # interpolate to original size
+>>> prediction = torch.nn.functional.interpolate(
+...     predicted_depth.unsqueeze(1),
+...     size=image.size[::-1],
+...     mode="bicubic",
+...     align_corners=False,
+... )
+
+>>> # visualize the prediction
+>>> output = prediction.squeeze().cpu().numpy()
+>>> formatted = (output * 255 / np.max(output)).astype("uint8")
+>>> depth = Image.fromarray(formatted)
+```
+
+## Resources
+
+A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Depth Anything.
+
+- [Monocular depth estimation task guide](../tasks/depth_estimation)
+- [Depth Anything V2 demo](https://huggingface.co/spaces/depth-anything/Depth-Anything-V2).
+- A notebook showcasing inference with [`DepthAnythingForDepthEstimation`] can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Depth%20Anything/Predicting_depth_in_an_image_with_Depth_Anything.ipynb). 🌎
+- [Core ML conversion of the `small` variant for use on Apple Silicon](https://huggingface.co/apple/coreml-depth-anything-v2-small).
+
+If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
+
+## DepthAnythingConfig
+
+[[autodoc]] DepthAnythingConfig
+
+## DepthAnythingForDepthEstimation
+
+[[autodoc]] DepthAnythingForDepthEstimation
+    - forward

docs/source/en/model_doc/gemma2.md

@@ -0,0 +1,58 @@
+
+<!--Copyright 2024 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.
+
+-->
+
+# Gemma2
+
+## Overview
+
+The Gemma2 model was proposed in [Gemma2: Open Models Based on Gemini Technology and Research](https://blog.google/technology/developers/google-gemma-2/) by Gemma2 Team, Google.
+Two Gemma2 models are released, with parameters sizes of 9 billion (9B) and 27 billion (27B).
+
+The abstract from the blog post is the following:
+
+*Now we’re officially releasing Gemma 2 to researchers and developers globally. Available in both 9 billion (9B) and 27 billion (27B) parameter sizes, Gemma 2 is higher-performing and more efficient at inference than the first generation, with significant safety advancements built in. In fact, at 27B, it offers competitive alternatives to models more than twice its size, delivering the kind of performance that was only possible with proprietary models as recently as December.*
+
+Tips:
+
+- The original checkpoints can be converted using the conversion script `src/transformers/models/Gemma2/convert_Gemma2_weights_to_hf.py` 
+
+This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ), [Pedro Cuenca](https://huggingface.co/pcuenq) and [Tom Arsen]().
+
+
+## Gemma2Config
+
+[[autodoc]] Gemma2Config
+
+## Gemma2Model
+
+[[autodoc]] Gemma2Model
+    - forward
+
+## Gemma2ForCausalLM
+
+[[autodoc]] Gemma2ForCausalLM
+    - forward
+
+## Gemma2ForSequenceClassification
+
+[[autodoc]] Gemma2ForSequenceClassification
+    - forward
+
+## Gemma2ForTokenClassification
+
+[[autodoc]] Gemma2ForTokenClassification
+    - forward

docs/source/en/model_doc/gpt_neox.md

@@ -95,6 +95,68 @@ Below is an expected speedup diagram that compares pure inference time between t
 <img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/gpt-neox-1.8b-speedup.jpg">
 </div>
 
+
+## Using Scaled Dot Product Attention (SDPA)
+PyTorch includes a native scaled dot-product attention (SDPA) operator as part of `torch.nn.functional`. This function
+encompasses several implementations that can be applied depending on the inputs and the hardware in use. See the
+[official documentation](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)
+or the [GPU Inference](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention)
+page for more information.
+
+SDPA is used by default for `torch>=2.1.1` when an implementation is available, but you may also set
+`attn_implementation="sdpa"` in `from_pretrained()` to explicitly request SDPA to be used.
+
+```python
+from transformers import GPTNeoXForCausalLM
+model = GPTNeoXForCausalLM.from_pretrained("EleutherAI/gpt-neox-20b", torch_dtype=torch.float16, attn_implementation="sdpa")
+...
+```
+
+For the best speedups, we recommend loading the model in half-precision (e.g. `torch.float16` or `torch.bfloat16`).
+
+On a local benchmark (rtx3080ti-16GB, PyTorch 2.2.1, OS Ubuntu 22.04) using `float16` with
+[pythia-410m-deduped](https://huggingface.co/EleutherAI/pythia-410m-deduped), we saw the
+following speedups during training and inference.
+
+### Training
+| Batch size |    Seq len | Time per batch (Eager - s) |    Time per batch (SDPA - s) | Speedup (%) | Eager peak mem (MB) | SDPA peak mem (MB) |    Mem saving (%) |
+|-----------:|-----------:|---------------------------:|-----------------------------:|------------:|--------------------:|-------------------:|------------------:|
+|          1 |        128 |                      0.024 |                        0.019 |      28.945 |             1789.95 |            1789.95 |                 0 |
+|          1 |        256 |                      0.039 |                        0.031 |       23.18 |             1845.83 |            1844.84 |             0.053 |
+|          1 |        512 |                       0.08 |                        0.055 |      45.524 |             2278.38 |            1953.76 |            16.615 |
+|          1 |       1024 |                       0.19 |                        0.102 |      86.777 |             4772.36 |            2408.35 |            98.159 |
+|          1 |       2048 |                      0.565 |                        0.204 |     177.098 |             13484.1 |            3882.01 |           247.348 |
+|          2 |        128 |                      0.037 |                        0.032 |      15.121 |             1843.86 |            1844.78 |             -0.05 |
+|          2 |        256 |                      0.067 |                        0.055 |      21.706 |             1999.72 |            1951.67 |             2.462 |
+|          2 |        512 |                      0.144 |                        0.096 |      50.046 |             3613.16 |            2406.77 |            50.125 |
+|          2 |       1024 |                      0.366 |                        0.193 |      89.666 |             8707.55 |            3878.86 |           124.487 |
+|          2 |       2048 |                        OOM |                        0.379 |           / |                 OOM |            6825.13 | SDPA does not OOM |
+|          4 |        128 |                       0.06 |                        0.054 |      11.539 |              1947.6 |            1952.06 |            -0.228 |
+|          4 |        256 |                      0.119 |                        0.093 |      28.072 |             3008.39 |            2405.99 |            25.038 |
+|          4 |        512 |                      0.275 |                        0.187 |      47.145 |             6290.58 |            3877.29 |            62.242 |
+|          4 |       1024 |                        OOM |                         0.36 |           / |                 OOM |            6821.98 | SDPA does not OOM |
+|          4 |       2048 |                        OOM |                        0.731 |           / |                 OOM |            12705.1 | SDPA does not OOM |
+
+### Inference
+|    Batch size |      Seq len |    Per token latency Eager (ms) |    Per token latency SDPA (ms) |    Speedup (%) |    Mem Eager (MB) |   Mem SDPA (MB) |    Mem saved (%) |
+|--------------:|-------------:|--------------------------------:|-------------------------------:|---------------:|------------------:|----------------:|-----------------:|
+|             1 |          128 |                           6.569 |                          5.858 |          12.14 |           974.831 |         974.826 |                0 |
+|             1 |          256 |                           7.009 |                          5.863 |         19.542 |           1029.01 |         1028.08 |             0.09 |
+|             1 |          512 |                           7.157 |                          5.965 |         19.983 |           1137.54 |         1137.52 |            0.001 |
+|             1 |         1024 |                           7.523 |                          6.506 |         15.637 |            1329.3 |         1329.26 |            0.003 |
+|             1 |         2048 |                           9.271 |                          9.205 |          0.713 |           1752.47 |         1734.51 |            1.036 |
+|             2 |          128 |                           7.239 |                          5.959 |         21.493 |            1044.8 |         1028.37 |            1.597 |
+|             2 |          256 |                           7.228 |                          6.036 |         19.757 |           1167.32 |         1137.73 |            2.601 |
+|             2 |          512 |                           7.538 |                          6.693 |         12.628 |           1352.93 |         1329.55 |            1.758 |
+|             2 |         1024 |                           8.916 |                          8.632 |          3.291 |           1752.56 |         1734.62 |            1.034 |
+|             2 |         2048 |                          12.628 |                         12.606 |          0.181 |           2558.72 |          2545.8 |            0.508 |
+|             4 |          128 |                           7.278 |                          6.046 |         20.373 |           1168.41 |         1137.79 |            2.691 |
+|             4 |          256 |                           7.614 |                          6.588 |         15.574 |            1353.1 |         1329.79 |            1.753 |
+|             4 |          512 |                           8.798 |                          8.144 |          8.028 |           1752.76 |         1734.85 |            1.032 |
+|             4 |         1024 |                          11.765 |                         11.303 |           4.09 |           2558.96 |         2546.04 |            0.508 |
+|             4 |         2048 |                          19.568 |                         17.735 |          10.33 |            4175.5 |         4165.26 |            0.246 |
+
+
 ## Resources
 
 - [Causal language modeling task guide](../tasks/language_modeling)

docs/source/en/model_doc/hiera.md

@@ -0,0 +1,48 @@
+<!--Copyright 2024 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.
+
+-->
+
+# Hiera
+
+## Overview
+
+Hiera was proposed in [Hiera: A Hierarchical Vision Transformer without the Bells-and-Whistles](https://arxiv.org/abs/2306.00989) by Chaitanya Ryali, Yuan-Ting Hu, Daniel Bolya, Chen Wei, Haoqi Fan, Po-Yao Huang, Vaibhav Aggarwal, Arkabandhu Chowdhury, Omid Poursaeed, Judy Hoffman, Jitendra Malik, Yanghao Li, Christoph Feichtenhofer
+
+The paper introduces "Hiera," a hierarchical Vision Transformer that simplifies the architecture of modern hierarchical vision transformers by removing unnecessary components without compromising on accuracy or efficiency. Unlike traditional transformers that add complex vision-specific components to improve supervised classification performance, Hiera demonstrates that such additions, often termed "bells-and-whistles," are not essential for high accuracy. By leveraging a strong visual pretext task (MAE) for pretraining, Hiera retains simplicity and achieves superior accuracy and speed both in inference and training across various image and video recognition tasks. The approach suggests that spatial biases required for vision tasks can be effectively learned through proper pretraining, eliminating the need for added architectural complexity. 
+
+The abstract from the paper is the following:
+
+*Modern hierarchical vision transformers have added several vision-specific components in the pursuit of supervised classification performance. While these components lead to effective accuracies and attractive FLOP counts, the added complexity actually makes these transformers slower than their vanilla ViT counterparts. In this paper, we argue that this additional bulk is unnecessary. By pretraining with a strong visual pretext task (MAE), we can strip out all the bells-and-whistles from a state-of-the-art multi-stage vision transformer without losing accuracy. In the process, we create Hiera, an extremely simple hierarchical vision transformer that is more accurate than previous models while being significantly faster both at inference and during training. We evaluate Hiera on a variety of tasks for image and video recognition. Our code and models are available at https://github.com/facebookresearch/hiera.*
+
+This model was a joint contibution by [EduardoPacheco](https://huggingface.co/EduardoPacheco) and [namangarg110](https://huggingface.co/namangarg110). The original code can be found [here] (https://github.com/facebookresearch/hiera).
+
+## HieraConfig
+
+[[autodoc]] HieraConfig
+
+## HieraModel
+
+[[autodoc]] HieraModel
+    - forward
+
+## HieraForPreTraining
+
+[[autodoc]] HieraForPreTraining
+    - forward
+  
+## HieraForImageClassification
+
+[[autodoc]] HieraForImageClassification
+    - forward

docs/source/en/model_doc/instructblip.md

@@ -50,6 +50,7 @@ InstructBLIP uses the same architecture as [BLIP-2](blip2) with a tiny but impor
 
 [[autodoc]] InstructBlipProcessor
 
+
 ## InstructBlipVisionModel
 
 [[autodoc]] InstructBlipVisionModel

docs/source/en/model_doc/instructblipvideo.md

@@ -0,0 +1,74 @@
+<!--Copyright 2024 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.
+-->
+
+# InstructBlipVideo
+
+## Overview
+
+## Overview
+
+The InstructBLIPVideo is an extension of the models proposed in [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
+InstructBLIPVideo uses the same architecture as [InstructBLIP](instructblip) and works with the same checkpoints as [InstructBLIP](instructblip). The only difference is the ability to process videos.
+
+The abstract from the paper is the following:
+
+*General-purpose language models that can solve various language-domain tasks have emerged driven by the pre-training and instruction-tuning pipeline. However, building general-purpose vision-language models is challenging due to the increased task discrepancy introduced by the additional visual input. Although vision-language pre-training has been widely studied, vision-language instruction tuning remains relatively less explored. In this paper, we conduct a systematic and comprehensive study on vision-language instruction tuning based on the pre-trained BLIP-2 models. We gather a wide variety of 26 publicly available datasets, transform them into instruction tuning format and categorize them into two clusters for held-in instruction tuning and held-out zero-shot evaluation. Additionally, we introduce instruction-aware visual feature extraction, a crucial method that enables the model to extract informative features tailored to the given instruction. The resulting InstructBLIP models achieve state-of-the-art zero-shot performance across all 13 held-out datasets, substantially outperforming BLIP-2 and the larger Flamingo. Our models also lead to state-of-the-art performance when finetuned on individual downstream tasks (e.g., 90.7% accuracy on ScienceQA IMG). Furthermore, we qualitatively demonstrate the advantages of InstructBLIP over concurrent multimodal models.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/instructblip_architecture.jpg"
+alt="drawing" width="600"/>
+
+<small> InstructBLIPVideo architecture. Taken from the <a href="https://arxiv.org/abs/2305.06500">original paper.</a> </small>
+
+This model was contributed by [RaushanTurganbay](https://huggingface.co/RaushanTurganbay).
+The original code can be found [here](https://github.com/salesforce/LAVIS/tree/main/projects/instructblip).
+
+## Usage tips
+
+- The model was trained by sampling 4 frames per video, so it's recommended to sample 4 frames
+
+## InstructBlipVideoConfig
+
+[[autodoc]] InstructBlipVideoConfig
+    - from_vision_qformer_text_configs
+
+## InstructBlipVideoVisionConfig
+
+[[autodoc]] InstructBlipVideoVisionConfig
+
+## InstructBlipVideoQFormerConfig
+
+[[autodoc]] InstructBlipVideoQFormerConfig
+
+## InstructBlipVideoProcessor
+
+[[autodoc]] InstructBlipVideoProcessor
+
+## InstructBlipVideoImageProcessor
+
+[[autodoc]] InstructBlipVideoImageProcessor
+    - preprocess
+
+## InstructBlipVideoVisionModel
+
+[[autodoc]] InstructBlipVideoVisionModel
+    - forward
+
+## InstructBlipVideoQFormerModel
+
+[[autodoc]] InstructBlipVideoQFormerModel
+    - forward
+
+## InstructBlipVideoForConditionalGeneration
+
+[[autodoc]] InstructBlipVideoForConditionalGeneration
+    - forward
+    - generate

docs/source/en/model_doc/llama3.md

@@ -16,6 +16,15 @@ rendered properly in your Markdown viewer.
 
 # Llama3
 
+```py3
+import transformers
+import torch
+
+model_id = "meta-llama/Meta-Llama-3-8B"
+
+pipeline = transformers.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.bfloat16}, device_map="auto")
+pipeline("Hey how are you doing today?")
+```
 
 ## Overview
 
@@ -66,20 +75,7 @@ model = AutoModelForCausalLM.from_pretrained("/output/path")
 Note that executing the script requires enough CPU RAM to host the whole model in float16 precision (even if the biggest versions
 come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM). For the 75B model, it's thus 145GB of RAM needed.
 
-
 - When using Flash Attention 2 via `attn_implementation="flash_attention_2"`, don't pass `torch_dtype` to the `from_pretrained` class method and use Automatic Mixed-Precision training. When using `Trainer`, it is simply specifying either `fp16` or `bf16` to `True`. Otherwise, make sure you are using `torch.autocast`. This is required because the Flash Attention only support `fp16` and `bf16` data type.
 
-## Quick usage
-
-```py3
-import transformers
-import torch
-
-model_id = "meta-llama/Meta-Llama-3-8B"
-
-pipeline = transformers.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.bfloat16}, device_map="auto")
-pipeline("Hey how are you doing today?")
-```
-
 ## Resources
-A ton of cool resources are already available on the documentation page of [~llama2], inviting contributors to add new resources curated for Llama3 here! 🤗
+A ton of cool resources are already available on the documentation page of [Llama2](./llama2), inviting contributors to add new resources curated for Llama3 here! 🤗

docs/source/en/model_doc/llava-next-video.md

@@ -0,0 +1,259 @@
+<!--Copyright 2024 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.
+
+-->
+
+# LLaVa-NeXT-Video
+
+## Overview
+
+The LLaVa-NeXT-Video model was proposed in [LLaVA-NeXT: A Strong Zero-shot Video Understanding Model
+](https://llava-vl.github.io/blog/2024-04-30-llava-next-video/) by Yuanhan Zhang, Bo Li, Haotian Liu, Yong Jae Lee, Liangke Gui, Di Fu, Jiashi Feng, Ziwei Liu, Chunyuan Li. LLaVa-NeXT-Video improves upon [LLaVa-NeXT](llava_next) by fine-tuning on a mix if video and image dataset thus increasing the model's performance on videos.
+
+[LLaVA-NeXT](llava_next) surprisingly has strong performance in understanding video content in zero-shot fashion with the AnyRes technique that it uses. The AnyRes technique naturally represents a high-resolution image into multiple images. This technique is naturally generalizable to represent videos because videos can be considered as a set of frames (similar to a set of images in LLaVa-NeXT). The current version of LLaVA-NeXT makes use of AnyRes and trains with supervised fine-tuning (SFT) on top of LLaVA-Next on video data to achieves better video understanding capabilities.The model is a current SOTA among open-source models on [VideoMME bench](https://arxiv.org/abs/2405.21075).
+
+
+The introduction from the blog is the following:
+
+On January 30, 2024, we released LLaVA-NeXT, an open-source Large Multimodal Model (LMM) that has been trained exclusively on text-image data. With the proposed AnyRes technique, it boosts capabilities in reasoning, OCR, and world knowledge, demonstrating remarkable performance across a spectrum of image-based multimodal understanding tasks, and even exceeding Gemini-Pro on several image benchmarks, e.g. MMMU and MathVista.
+
+**In today’s exploration, we delve into the performance of LLaVA-NeXT within the realm of video understanding tasks. We reveal that LLaVA-NeXT surprisingly has strong performance in understanding video content. The current version of LLaVA-NeXT for videos has several improvements:
+
+- Zero-shot video representation capabilities with AnyRes: The AnyRes technique naturally represents a high-resolution image into multiple images that a pre-trained VIT is able to digest, and forms them into a concantenated sequence. This technique is naturally generalizable to represent videos (consisting of multiple frames), allowing the image-only-trained LLaVA-Next model to perform surprisingly well on video tasks. Notably, this is the first time that LMMs show strong zero-shot modality transfer ability.
+- Inference with length generalization improves on longer videos. The linear scaling technique enables length generalization, allowing LLaVA-NeXT to effectively handle long-video beyond the limitation of the "max_token_length" of the LLM.
+- Strong video understanding ability. (1) LLaVA-Next-Image, which combines the above two techniques, yields superior zero-shot performance than open-source LMMs tuned on videos. (2) LLaVA-Next-Video, further supervised fine-tuning (SFT) LLaVA-Next-Image on video data, achieves better video understanding capabilities compared to LLaVA-Next-Image. (3) LLaVA-Next-Video-DPO, which aligns the model response with AI feedback using direct preference optimization (DPO), showing significant performance boost.
+- Efficient deployment and inference with SGLang. It allows 5x faster inference on video tasks, allowing more scalable serving such as million-level video re-captioning. See instructions in our repo.**
+
+
+This model was contributed by [RaushanTurganbay](https://huggingface.co/RaushanTurganbay).
+The original code can be found [here](https://github.com/LLaVA-VL/LLaVA-NeXT/tree/inference).
+
+## Usage tips
+
+- We advise users to use `padding_side="left"` when computing batched generation as it leads to more accurate results. Simply make sure to call `processor.tokenizer.padding_side = "left"` before generating.
+
+- Note that each checkpoint has been trained with a specific prompt format, depending on which large language model (LLM) was used. You can use tokenizer's `apply_chat_template` to format your prompts correctly. Below is an example of how to do that.
+
+We will use [LLaVA-NeXT-Video-7B-hf](https://huggingface.co/llava-hf/LLaVA-NeXT-Video-7B-hf) and a conversation history of videos and images. Each content field has to be a list of dicts, as follows:
+
+```python
+from transformers import LlavaNextVideoProcessor
+
+processor = LlavaNextVideoProcessor.from_pretrained("llava-hf/LLaVA-NeXT-Video-7B-hf")
+
+conversation = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions."},
+            ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "What’s shown in this image?"},
+            {"type": "image"},
+            ],
+    },
+    {
+        "role": "assistant",
+        "content": [{"type": "text", "text": "This image shows a red stop sign."},]
+    },
+    {
+
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "Why is this video funny?"},
+            {"type": "video"},
+            ],
+    },
+]
+
+text_prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)
+
+# Note that the template simply formats your prompt, you still have to tokenize it and obtain pixel values for your visuals
+print(text_prompt)
+```
+
+## Usage example
+
+### Single Media Mode
+
+The model can accept both images and videos as input. Here's an example code for inference in half-precision (`torch.float16`):
+
+```python
+import av
+import torch
+import numpy as np
+from transformers import LlavaNextVideoForConditionalGeneration, LlavaNextVideoProcessor
+
+def read_video_pyav(container, indices):
+    '''
+    Decode the video with PyAV decoder.
+    Args:
+        container (`av.container.input.InputContainer`): PyAV container.
+        indices (`List[int]`): List of frame indices to decode.
+    Returns:
+        result (np.ndarray): np array of decoded frames of shape (num_frames, height, width, 3).
+    '''
+    frames = []
+    container.seek(0)
+    start_index = indices[0]
+    end_index = indices[-1]
+    for i, frame in enumerate(container.decode(video=0)):
+        if i > end_index:
+            break
+        if i >= start_index and i in indices:
+            frames.append(frame)
+    return np.stack([x.to_ndarray(format="rgb24") for x in frames])
+
+# Load the model in half-precision
+model = LlavaNextVideoForConditionalGeneration.from_pretrained("llava-hf/LLaVA-NeXT-Video-7B-hf", torch_dtype=torch.float16, device_map="auto")
+processor = LlavaNextVideoProcessor.from_pretrained("llava-hf/LLaVA-NeXT-Video-7B-hf")
+
+# Load the video as an np.array, sampling uniformly 8 frames (can sample more for longer videos)
+video_path = hf_hub_download(repo_id="raushan-testing-hf/videos-test", filename="sample_demo_1.mp4", repo_type="dataset")
+container = av.open(video_path)
+total_frames = container.streams.video[0].frames
+indices = np.arange(0, total_frames, total_frames / 8).astype(int)
+video = read_video_pyav(container, indices)
+
+conversation = [
+    {
+
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "Why is this video funny?"},
+            {"type": "video"},
+            ],
+    },
+]
+
+prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)
+inputs = processor(text=prompt, videos=video, return_tensors="pt")
+
+out = model.generate(**inputs, max_new_tokens=60)
+processor.batch_decode(out, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+```
+
+
+### Mixed Media Mode
+
+The model can also generate from an interleaved image-video inputs. However note, that it was not trained in interleaved image-video setting which might affect the performance. Below is an example usage for mixed media input, add the following lines to the above code snippet: 
+
+```python
+from PIL import Image
+import requests
+
+# Generate from image and video mixed inputs
+# Load and image and write a new prompt
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+conversation = [
+    {
+
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "How many cats are there in the image?"},
+            {"type": "image"},
+            ],
+    },
+    {
+
+        "role": "assistant",
+        "content": [{"type": "text", "text": "There are two cats"}],
+    },
+    {
+
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "Why is this video funny?"},
+            {"type": "video"},
+            ],
+    },
+]
+prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)
+inputs = processor(text=prompt, images=image, videos=clip, padding=True, return_tensors="pt")
+
+# Generate
+generate_ids = model.generate(**inputs, max_length=50)
+processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+
+```
+
+## Model optimization
+
+### Quantization using Bitsandbytes for memory efficiency
+
+The model can be loaded in lower bits, significantly reducing memory burden while maintaining the performance of the original model. This allows for efficient deployment on resource-constrained cases. 
+
+First make sure to install bitsandbytes by running `pip install bitsandbytes` and to have access to a CUDA compatible GPU device. Load the quantized model by simply adding [`BitsAndBytesConfig`](../main_classes/quantization#transformers.BitsAndBytesConfig) as shown below:
+
+
+```python
+from transformers import LlavaNextVideoForConditionalGeneration, LlavaNextVideoProcessor
+
+# specify how to quantize the model
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.float16,
+)
+
+model = LlavaNextVideoForConditionalGeneration.from_pretrained("llava-hf/LLaVA-NeXT-Video-7B-hf", quantization_config=quantization_config, device_map="auto")
+```
+
+
+### Flash-Attention 2 to speed-up generation
+
+Additionally, we can greatly speed-up model inference by using [Flash Attention](../perf_train_gpu_one.md#flash-attention-2), which is a faster implementation of the attention mechanism used inside the model.
+
+First, make sure to install the latest version of Flash Attention 2:
+
+```bash
+pip install -U flash-attn --no-build-isolation
+```
+
+Also, you should have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of the [flash attention repository](https://github.com/Dao-AILab/flash-attention). FlashAttention-2 can only be used when a model is loaded in `torch.float16` or `torch.bfloat16`.
+
+To load and run a model using Flash Attention-2, simply add `attn_implementation="flash_attention_2"` when loading the model as follows:
+
+```python
+from transformers import LlavaNextVideoForConditionalGeneration
+
+model = LlavaNextVideoForConditionalGeneration.from_pretrained(
+    "llava-hf/LLaVA-NeXT-Video-7B-hf", 
+    torch_dtype=torch.float16, 
+    attn_implementation="flash_attention_2",
+).to(0)
+```
+
+
+
+## LlavaNextVideoConfig
+
+[[autodoc]] LlavaNextVideoConfig
+
+## LlavaNextVideoProcessor
+
+[[autodoc]] LlavaNextVideoProcessor
+
+## LlavaNextVideoImageProcessor
+
+[[autodoc]] LlavaNextVideoImageProcessor
+
+## LlavaNextVideoForConditionalGeneration
+
+[[autodoc]] LlavaNextVideoForConditionalGeneration
+    - forward

docs/source/en/model_doc/llava.md

@@ -40,8 +40,20 @@ The original code can be found [here](https://github.com/haotian-liu/LLaVA/tree/
 
 - Note the model has not been explicitly trained to process multiple images in the same prompt, although this is technically possible, you may experience inaccurate results.
 
-- For better results, we recommend users to prompt the model with the correct prompt format: 
+- For better results, we recommend users to prompt the model with the correct prompt format. Below is a list of prompt formats accepted by each llava checkpoint:
 
+[llava-interleave models](https://huggingface.co/collections/llava-hf/llava-interleave-668e19a97da0036aad4a2f19) requires the following format:
+```bash
+"<|im_start|>user <image>\nWhat is shown in this image?<|im_end|><|im_start|>assistant"
+```
+
+For multiple turns conversation:
+
+```bash
+"<|im_start|>user <image>\n<prompt1><|im_end|><|im_start|>assistant <answer1><|im_end|><|im_start|>user <image>\n<prompt1><|im_end|><|im_start|>assistant "
+```
+
+[llava-1.5 models](https://huggingface.co/collections/llava-hf/llava-15-65f762d5b6941db5c2ba07e0) requires the following format:
 ```bash
 "USER: <image>\n<prompt> ASSISTANT:"
 ```

docs/source/en/model_doc/prophetnet.md

@@ -51,7 +51,7 @@ The Authors' code can be found [here](https://github.com/microsoft/ProphetNet).
 
 - ProphetNet is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather than
   the left.
-- The model architecture is based on the original Transformer, but replaces the “standard” self-attention mechanism in the decoder by a a main self-attention mechanism and a self and n-stream (predict) self-attention mechanism.
+- The model architecture is based on the original Transformer, but replaces the “standard” self-attention mechanism in the decoder by a main self-attention mechanism and a self and n-stream (predict) self-attention mechanism.
 
 ## Resources
 

docs/source/en/model_doc/rt_detr.md

@@ -0,0 +1,111 @@
+<!--Copyright 2024 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.
+
+-->
+
+# RT-DETR
+
+## Overview
+
+
+The RT-DETR model was proposed in [DETRs Beat YOLOs on Real-time Object Detection](https://arxiv.org/abs/2304.08069) by Wenyu Lv, Yian Zhao, Shangliang Xu, Jinman Wei, Guanzhong Wang, Cheng Cui, Yuning Du, Qingqing Dang, Yi Liu.
+
+RT-DETR is an object detection model that stands for "Real-Time DEtection Transformer." This model is designed to perform object detection tasks with a focus on achieving real-time performance while maintaining high accuracy. Leveraging the transformer architecture, which has gained significant popularity in various fields of deep learning, RT-DETR processes images to identify and locate multiple objects within them.
+
+The abstract from the paper is the following:
+
+*Recently, end-to-end transformer-based detectors (DETRs) have achieved remarkable performance. However, the issue of the high computational cost of DETRs has not been effectively addressed, limiting their practical application and preventing them from fully exploiting the benefits of no post-processing, such as non-maximum suppression (NMS). In this paper, we first analyze the influence of NMS in modern real-time object detectors on inference speed, and establish an end-to-end speed benchmark. To avoid the inference delay caused by NMS, we propose a Real-Time DEtection TRansformer (RT-DETR), the first real-time end-to-end object detector to our best knowledge. Specifically, we design an efficient hybrid encoder to efficiently process multi-scale features by decoupling the intra-scale interaction and cross-scale fusion, and propose IoU-aware query selection to improve the initialization of object queries. In addition, our proposed detector supports flexibly adjustment of the inference speed by using different decoder layers without the need for retraining, which facilitates the practical application of real-time object detectors. Our RT-DETR-L achieves 53.0% AP on COCO val2017 and 114 FPS on T4 GPU, while RT-DETR-X achieves 54.8% AP and 74 FPS, outperforming all YOLO detectors of the same scale in both speed and accuracy. Furthermore, our RT-DETR-R50 achieves 53.1% AP and 108 FPS, outperforming DINO-Deformable-DETR-R50 by 2.2% AP in accuracy and by about 21 times in FPS.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/rt_detr_overview.png"
+alt="drawing" width="600"/>
+
+<small> RT-DETR performance relative to YOLO models. Taken from the <a href="https://arxiv.org/abs/2304.08069">original paper.</a> </small>
+
+The model version was contributed by [rafaelpadilla](https://huggingface.co/rafaelpadilla) and [sangbumchoi](https://github.com/SangbumChoi). The original code can be found [here](https://github.com/lyuwenyu/RT-DETR/).
+
+
+## Usage tips
+
+Initially, an image is processed using a pre-trained convolutional neural network, specifically a Resnet-D variant as referenced in the original code. This network extracts features from the final three layers of the architecture. Following this, a hybrid encoder is employed to convert the multi-scale features into a sequential array of image features. Then, a decoder, equipped with auxiliary prediction heads is used to refine the object queries. This process facilitates the direct generation of bounding boxes, eliminating the need for any additional post-processing to acquire the logits and coordinates for the bounding boxes.
+
+```py
+>>> import torch
+>>> import requests
+
+>>> from PIL import Image
+>>> from transformers import RTDetrForObjectDetection, RTDetrImageProcessor
+
+>>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg' 
+>>> image = Image.open(requests.get(url, stream=True).raw)
+
+>>> image_processor = RTDetrImageProcessor.from_pretrained("PekingU/rtdetr_r50vd")
+>>> model = RTDetrForObjectDetection.from_pretrained("PekingU/rtdetr_r50vd")
+
+>>> inputs = image_processor(images=image, return_tensors="pt")
+
+>>> with torch.no_grad():
+...     outputs = model(**inputs)
+
+>>> results = image_processor.post_process_object_detection(outputs, target_sizes=torch.tensor([image.size[::-1]]), threshold=0.3)
+
+>>> for result in results:
+...     for score, label_id, box in zip(result["scores"], result["labels"], result["boxes"]):
+...         score, label = score.item(), label_id.item()
+...         box = [round(i, 2) for i in box.tolist()]
+...         print(f"{model.config.id2label[label]}: {score:.2f} {box}")
+sofa: 0.97 [0.14, 0.38, 640.13, 476.21]
+cat: 0.96 [343.38, 24.28, 640.14, 371.5]
+cat: 0.96 [13.23, 54.18, 318.98, 472.22]
+remote: 0.95 [40.11, 73.44, 175.96, 118.48]
+remote: 0.92 [333.73, 76.58, 369.97, 186.99]
+```
+
+## Resources
+
+A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with RT-DETR.
+
+<PipelineTag pipeline="object-detection"/>
+
+- Scripts for finetuning [`RTDetrForObjectDetection`] with [`Trainer`] or [Accelerate](https://huggingface.co/docs/accelerate/index) can be found [here](https://github.com/huggingface/transformers/tree/main/examples/pytorch/object-detection).
+- See also: [Object detection task guide](../tasks/object_detection).
+- Notebooks regarding inference and fine-tuning RT-DETR on a custom dataset can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/RT-DETR). 🌎
+
+## RTDetrConfig
+
+[[autodoc]] RTDetrConfig
+
+## RTDetrResNetConfig
+
+[[autodoc]] RTDetrResNetConfig
+
+## RTDetrImageProcessor
+
+[[autodoc]] RTDetrImageProcessor
+    - preprocess
+    - post_process_object_detection
+
+## RTDetrModel
+
+[[autodoc]] RTDetrModel
+    - forward
+
+## RTDetrForObjectDetection
+
+[[autodoc]] RTDetrForObjectDetection
+    - forward
+
+## RTDetrResNetBackbone
+
+[[autodoc]] RTDetrResNetBackbone
+    - forward

docs/source/en/model_doc/siglip.md

@@ -27,7 +27,7 @@ The abstract from the paper is the following:
 ## Usage tips
 
 - Usage of SigLIP is similar to [CLIP](clip). The main difference is the training loss, which does not require a global view of all the pairwise similarities of images and texts within a batch. One needs to apply the sigmoid activation function to the logits, rather than the softmax.
-- Training is not yet supported. If you want to fine-tune SigLIP or train from scratch, refer to the loss function from [OpenCLIP](https://github.com/mlfoundations/open_clip/blob/73ad04ae7fb93ede1c02dc9040a828634cb1edf1/src/open_clip/loss.py#L307), which leverages various `torch.distributed` utilities.
+- Training is supported but does not use `torch.distributed` utilities which may limit the scalability of batch size. However, DDP and FDSP works on single-node multi-gpu setup.
 - When using the standalone [`SiglipTokenizer`] or [`SiglipProcessor`], make sure to pass `padding="max_length"` as that's how the model was trained.
 - To get the same results as the pipeline, a prompt template of "This is a photo of {label}." should be used.
 
@@ -107,6 +107,88 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 
 If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
 
+
+## Combining SigLIP and Flash Attention 2
+
+First, make sure to install the latest version of Flash Attention 2.
+
+```bash
+pip install -U flash-attn --no-build-isolation
+```
+
+Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of flash-attn repository. Make also sure to load your model in half-precision (e.g. `torch.float16``)
+
+To load and run a model using Flash Attention 2, refer to the snippet below:
+
+```python
+>>> import torch
+>>> import requests
+>>> from PIL import Image
+>>> from transformers import SiglipProcessor, SiglipModel
+>>> device = "cuda" # the device to load the model onto
+
+>>> model = SiglipModel.from_pretrained(
+...     "google/siglip-so400m-patch14-384",
+...     attn_implementation="flash_attention_2",
+...     torch_dtype=torch.float16,
+...     device_map=device,
+... )
+>>> processor = SiglipProcessor.from_pretrained("google/siglip-so400m-patch14-384")
+
+>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+>>> image = Image.open(requests.get(url, stream=True).raw)
+
+>>> candidate_labels = ["2 cats", "2 dogs"]
+# follows the pipeline prompt template to get same results
+>>> candidate_labels = [f'This is a photo of {label}.' for label in candidate_labels]
+# important: we pass `padding=max_length` since the model was trained with this
+>>> inputs = processor(text=candidate_labels, images=image, padding="max_length", return_tensors="pt")
+>>> inputs.to(device)
+
+>>> with torch.no_grad():
+...     with torch.autocast(device):
+...         outputs = model(**inputs)
+
+>>> logits_per_image = outputs.logits_per_image
+>>> probs = torch.sigmoid(logits_per_image) # these are the probabilities
+>>> print(f"{probs[0][0]:.1%} that image 0 is '{candidate_labels[0]}'")
+51.3% that image 0 is 'This is a photo of 2 cats.'
+```
+
+
+## Using Scaled Dot Product Attention (SDPA)
+
+PyTorch includes a native scaled dot-product attention (SDPA) operator as part of `torch.nn.functional`. This function 
+encompasses several implementations that can be applied depending on the inputs and the hardware in use. See the 
+[official documentation](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) 
+or the [GPU Inference](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention)
+page for more information.
+
+You may set `attn_implementation="sdpa"` in `from_pretrained()` to explicitly request SDPA to be used. Make sure you have `torch>=2.1.1`.
+
+```python
+>>> from transformers import SiglipModel
+
+>>> model = SiglipModel.from_pretrained(
+...     "google/siglip-so400m-patch14-384",
+...     attn_implementation="sdpa",
+...     torch_dtype=torch.float16,
+...     device_map=device,
+... )
+```
+
+For the best speedups, we recommend loading the model in half-precision (e.g. `torch.float16` or `torch.bfloat16`).
+
+
+## Expected speedups
+
+Below is an expected speedup diagram that compares inference time between the native implementation in transformers using `google/siglip-so400m-patch14-384` checkpoint in `float16` precision and the Flash Attention 2 / SDPA version of the model using different batch sizes.
+
+<div style="text-align: center">
+<img src="https://i.imgur.com/cWm4rsn.png">
+</div>
+
+
 ## SiglipConfig
 
 [[autodoc]] SiglipConfig

docs/source/en/model_doc/whisper.md

@@ -52,8 +52,6 @@ Here is a step-by-step guide to transcribing an audio sample using a pre-trained
 >>> # Select an audio file and read it:
 >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
 >>> audio_sample = ds[0]["audio"]
->>> waveform = audio_sample["array"]
->>> sampling_rate = audio_sample["sampling_rate"]
 
 >>> # Load the Whisper model in Hugging Face format:
 >>> processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
@@ -61,7 +59,7 @@ Here is a step-by-step guide to transcribing an audio sample using a pre-trained
 
 >>> # Use the model and processor to transcribe the audio:
 >>> input_features = processor(
-...     waveform, sampling_rate=sampling_rate, return_tensors="pt"
+...     audio_sample["array"], sampling_rate=audio_sample["sampling_rate"], return_tensors="pt"
 ... ).input_features
 
 >>> # Generate token ids
@@ -74,6 +72,49 @@ Here is a step-by-step guide to transcribing an audio sample using a pre-trained
 ' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
 ```
 
+Whisper is compatible with the following optimisations:
+- [PyTorch Scaled Dot Product Attention (SDPA)](../perf_infer_gpu_one#pytorch-scaled-dot-product-attention): flash attention and memory-efficient attention kernels. Enabled by default for `torch>=2.1.1`.
+- [Flash Attention 2](../perf_infer_gpu_one#flashattention-2): improved implementation of flash attention through better parallelism and work partitioning. 
+- [torch.compile](../llm_optims#static-kv-cache-and-torchcompile): JIT-compile the forward pass to dispatch to efficient fused kernels.
+
+As an example, the following codesnippet enables SDPA and `torch.compile` for up to 5x faster inference:
+
+```python
+>>> from datasets import load_dataset
+>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
+
+>>> # Select an audio file and read it:
+>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
+>>> audio_sample = ds[0]["audio"]
+
+>>> # Load the Whisper model with SDPA attention
+>>> processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
+>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en", attn_implementation="sdpa")
+
+>>> # Enable static cache and compile the forward pass
+>>> model.generation_config.cache_implementation = "static"
+>>> model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
+
+>>> # Use the model and processor to transcribe the audio:
+>>> input_features = processor(
+...     audio_sample["array"], sampling_rate=audio_sample["sampling_rate"], return_tensors="pt"
+... ).input_features
+
+>>> # Compile the forward pass
+>>> _ = model.generate(input_features)
+
+>>> # Generate token ids using compiled graph (fast!)
+>>> predicted_ids = model.generate(input_features)
+
+>>> # Decode token ids to text
+>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
+
+>>> transcription[0]
+' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
+```
+
+For more details on each optimisation, refer to the documentation linked above.
+
 ## Resources
 
 A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Whisper. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.

docs/source/en/model_doc/zoedepth.md

@@ -0,0 +1,108 @@
+<!--Copyright 2024 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.
+
+-->
+
+# ZoeDepth
+
+## Overview
+
+The ZoeDepth model was proposed in [ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth](https://arxiv.org/abs/2302.12288) by Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, Matthias Müller. ZoeDepth extends the [DPT](dpt) framework for metric (also called absolute) depth estimation. ZoeDepth is pre-trained on 12 datasets using relative depth and fine-tuned on two domains (NYU and KITTI) using metric depth. A lightweight head is used with a novel bin adjustment design called metric bins module for each domain. During inference, each input image is automatically routed to the appropriate head using a latent classifier.
+
+The abstract from the paper is the following:
+
+*This paper tackles the problem of depth estimation from a single image. Existing work either focuses on generalization performance disregarding metric scale, i.e. relative depth estimation, or state-of-the-art results on specific datasets, i.e. metric depth estimation. We propose the first approach that combines both worlds, leading to a model with excellent generalization performance while maintaining metric scale. Our flagship model, ZoeD-M12-NK, is pre-trained on 12 datasets using relative depth and fine-tuned on two datasets using metric depth. We use a lightweight head with a novel bin adjustment design called metric bins module for each domain. During inference, each input image is automatically routed to the appropriate head using a latent classifier. Our framework admits multiple configurations depending on the datasets used for relative depth pre-training and metric fine-tuning. Without pre-training, we can already significantly improve the state of the art (SOTA) on the NYU Depth v2 indoor dataset. Pre-training on twelve datasets and fine-tuning on the NYU Depth v2 indoor dataset, we can further improve SOTA for a total of 21% in terms of relative absolute error (REL). Finally, ZoeD-M12-NK is the first model that can jointly train on multiple datasets (NYU Depth v2 and KITTI) without a significant drop in performance and achieve unprecedented zero-shot generalization performance to eight unseen datasets from both indoor and outdoor domains.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/zoedepth_architecture_bis.png"
+alt="drawing" width="600"/>
+
+<small> ZoeDepth architecture. Taken from the <a href="https://arxiv.org/abs/2302.12288">original paper.</a> </small>
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr).
+The original code can be found [here](https://github.com/isl-org/ZoeDepth).
+
+## Usage tips
+
+- ZoeDepth is an absolute (also called metric) depth estimation model, unlike DPT which is a relative depth estimation model. This means that ZoeDepth is able to estimate depth in metric units like meters.
+
+The easiest to perform inference with ZoeDepth is by leveraging the [pipeline API](../main_classes/pipelines.md):
+
+```python
+from transformers import pipeline
+from PIL import Image
+import requests
+
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+
+pipe = pipeline(task="depth-estimation", model="Intel/zoedepth-nyu-kitti")
+result = pipe(image)
+depth = result["depth"]
+```
+
+Alternatively, one can also perform inference using the classes:
+
+```python
+from transformers import AutoImageProcessor, ZoeDepthForDepthEstimation
+import torch
+import numpy as np
+from PIL import Image
+import requests
+
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+
+image_processor = AutoImageProcessor.from_pretrained("Intel/zoedepth-nyu-kitti")
+model = ZoeDepthForDepthEstimation.from_pretrained("Intel/zoedepth-nyu-kitti")
+
+# prepare image for the model
+inputs = image_processor(images=image, return_tensors="pt")
+
+with torch.no_grad():
+    outputs = model(**inputs)
+    predicted_depth = outputs.predicted_depth
+
+# interpolate to original size
+prediction = torch.nn.functional.interpolate(
+    predicted_depth.unsqueeze(1),
+    size=image.size[::-1],
+    mode="bicubic",
+    align_corners=False,
+)
+
+# visualize the prediction
+output = prediction.squeeze().cpu().numpy()
+formatted = (output * 255 / np.max(output)).astype("uint8")
+depth = Image.fromarray(formatted)
+```
+
+## Resources
+
+A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with ZoeDepth.
+
+- A demo notebook regarding inference with ZoeDepth models can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/ZoeDepth). 🌎
+
+## ZoeDepthConfig
+
+[[autodoc]] ZoeDepthConfig
+
+## ZoeDepthImageProcessor
+
+[[autodoc]] ZoeDepthImageProcessor
+    - preprocess
+
+## ZoeDepthForDepthEstimation
+
+[[autodoc]] ZoeDepthForDepthEstimation
+    - forward

docs/source/en/perf_hardware.md

@@ -116,7 +116,7 @@ Each new generation provides a faster bandwidth, e.g. here is a quote from [Nvid
 
 So the higher `X` you get in the report of `NVX` in the output of `nvidia-smi topo -m` the better. The generation will depend on your GPU architecture.
 
-Let's compare the execution of a openai-community/gpt2 language model training over a small sample of wikitext.
+Let's compare the execution of an openai-community/gpt2 language model training over a small sample of wikitext.
 
 The results are:
 

docs/source/en/perf_infer_gpu_one.md

@@ -43,6 +43,7 @@ FlashAttention-2 is currently supported for the following architectures:
 * [Dbrx](https://huggingface.co/docs/transformers/model_doc/dbrx#transformers.DbrxModel)
 * [DistilBert](https://huggingface.co/docs/transformers/model_doc/distilbert#transformers.DistilBertModel)
 * [Gemma](https://huggingface.co/docs/transformers/model_doc/gemma#transformers.GemmaModel)
+* [Gemma2](https://huggingface.co/docs/transformers/model_doc/gemma2#transformers.Gemma2Model)
 * [GPT2](https://huggingface.co/docs/transformers/model_doc/gpt2)
 * [GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode#transformers.GPTBigCodeModel)
 * [GPTNeo](https://huggingface.co/docs/transformers/model_doc/gpt_neo#transformers.GPTNeoModel)
@@ -55,6 +56,7 @@ FlashAttention-2 is currently supported for the following architectures:
 * [Llama](https://huggingface.co/docs/transformers/model_doc/llama#transformers.LlamaModel)
 * [Llava](https://huggingface.co/docs/transformers/model_doc/llava)
 * [Llava-NeXT](https://huggingface.co/docs/transformers/model_doc/llava_next)
+* [Llava-NeXT-Video](https://huggingface.co/docs/transformers/model_doc/llava_next_video)
 * [VipLlava](https://huggingface.co/docs/transformers/model_doc/vipllava)
 * [VideoLlava](https://huggingface.co/docs/transformers/model_doc/video_llava)
 * [M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)
@@ -68,6 +70,7 @@ FlashAttention-2 is currently supported for the following architectures:
 * [OPT](https://huggingface.co/docs/transformers/model_doc/opt#transformers.OPTModel)
 * [Phi](https://huggingface.co/docs/transformers/model_doc/phi#transformers.PhiModel)
 * [Phi3](https://huggingface.co/docs/transformers/model_doc/phi3#transformers.Phi3Model)
+* [SigLIP](https://huggingface.co/docs/transformers/model_doc/siglip)
 * [StableLm](https://huggingface.co/docs/transformers/model_doc/stablelm#transformers.StableLmModel)
 * [Starcoder2](https://huggingface.co/docs/transformers/model_doc/starcoder2#transformers.Starcoder2Model)
 * [Qwen2](https://huggingface.co/docs/transformers/model_doc/qwen2#transformers.Qwen2Model)
@@ -201,8 +204,10 @@ For now, Transformers supports SDPA inference and training for the following arc
 * [Dpr](https://huggingface.co/docs/transformers/model_doc/dpr#transformers.DprReader)
 * [Falcon](https://huggingface.co/docs/transformers/model_doc/falcon#transformers.FalconModel)
 * [Gemma](https://huggingface.co/docs/transformers/model_doc/gemma#transformers.GemmaModel)
+* [Gemma2](https://huggingface.co/docs/transformers/model_doc/gemma2#transformers.Gemma2Model)
 * [GPT2](https://huggingface.co/docs/transformers/model_doc/gpt2)
 * [GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode#transformers.GPTBigCodeModel)
+* [GPTNeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox#transformers.GPTNeoXModel)
 * [JetMoe](https://huggingface.co/docs/transformers/model_doc/jetmoe#transformers.JetMoeModel)
 * [Jamba](https://huggingface.co/docs/transformers/model_doc/jamba#transformers.JambaModel)
 * [Llama](https://huggingface.co/docs/transformers/model_doc/llama#transformers.LlamaModel)
@@ -227,6 +232,7 @@ For now, Transformers supports SDPA inference and training for the following arc
 * [wav2vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2#transformers.Wav2Vec2Model)
 * [Hubert](https://huggingface.co/docs/transformers/model_doc/hubert#transformers.HubertModel)
 * [data2vec_audio](https://huggingface.co/docs/transformers/main/en/model_doc/data2vec#transformers.Data2VecAudioModel)
+* [SigLIP](https://huggingface.co/docs/transformers/model_doc/siglip)
 * [Sew](https://huggingface.co/docs/transformers/main/en/model_doc/sew#transformers.SEWModel)
 * [UniSpeech](https://huggingface.co/docs/transformers/v4.39.3/en/model_doc/unispeech#transformers.UniSpeechModel)
 * [unispeech_sat](https://huggingface.co/docs/transformers/v4.39.3/en/model_doc/unispeech-sat#transformers.UniSpeechSatModel)

docs/source/en/perf_train_gpu_one.md

@@ -41,21 +41,22 @@ hyperparameter tuning, you should determine which batch size yields the best res
 
 The methods and tools covered in this guide can be classified based on the effect they have on the training process:
 
-| Method/tool                                                | Improves training speed | Optimizes memory utilization |
-|:-----------------------------------------------------------|:------------------------|:-----------------------------|
-| [Batch size choice](#batch-size-choice)                    | Yes                     | Yes                          |
-| [Gradient accumulation](#gradient-accumulation)            | No                      | Yes                          |
-| [Gradient checkpointing](#gradient-checkpointing)          | No                      | Yes                          |
-| [Mixed precision training](#mixed-precision-training)      | Yes                     | (No)                         |
-| [Optimizer choice](#optimizer-choice)                      | Yes                     | Yes                          |
-| [Data preloading](#data-preloading)                        | Yes                     | No                           |
-| [DeepSpeed Zero](#deepspeed-zero)                          | No                      | Yes                          |
-| [torch.compile](#using-torchcompile)                       | Yes                     | No                           |
-| [Parameter-Efficient Fine Tuning (PEFT)](#using--peft)            | No                      | Yes                          |
+| Method/tool                                                                                                                                             | Improves training speed | Optimizes memory utilization |
+|:--------------------------------------------------------------------------------------------------------------------------------------------------------|:------------------------|:-----------------------------|
+| [Batch size choice](#batch-size-choice)                                                                                                                 | Yes                     | Yes                          |
+| [Gradient accumulation](#gradient-accumulation)                                                                                                         | No                      | Yes                          |
+| [Gradient checkpointing](#gradient-checkpointing)                                                                                                       | No                      | Yes                          |
+| [Mixed precision training](#mixed-precision-training)                                                                                                   | Yes                     | Maybe*                       |
+| [torch_empty_cache_steps](https://huggingface.co/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments.torch_empty_cache_steps) | No                      | Yes                          |
+| [Optimizer choice](#optimizer-choice)                                                                                                                   | Yes                     | Yes                          |
+| [Data preloading](#data-preloading)                                                                                                                     | Yes                     | No                           |
+| [DeepSpeed Zero](#deepspeed-zero)                                                                                                                       | No                      | Yes                          |
+| [torch.compile](#using-torchcompile)                                                                                                                    | Yes                     | No                           |
+| [Parameter-Efficient Fine Tuning (PEFT)](#using--peft)                                                                                                  | No                      | Yes                          |
  
 <Tip>
 
-Note: when using mixed precision with a small model and a large batch size, there will be some memory savings but with a 
+*Note: when using mixed precision with a small model and a large batch size, there will be some memory savings but with a 
 large model and a small batch size, the memory use will be larger.
 
 </Tip>

docs/source/en/pipeline_tutorial.md

@@ -113,7 +113,9 @@ This will work regardless of whether you are using PyTorch or Tensorflow.
 transcriber = pipeline(model="openai/whisper-large-v2", device=0)
 ```
 
-If the model is too large for a single GPU and you are using PyTorch, you can set `device_map="auto"` to automatically 
+If the model is too large for a single GPU and you are using PyTorch, you can set `torch_dtype='float16'` to enable FP16 precision inference. Usually this would not cause significant performance drops but make sure you evaluate it on your models!
+
+Alternatively, you can set `device_map="auto"` to automatically 
 determine how to load and store the model weights. Using the `device_map` argument requires the 🤗 [Accelerate](https://huggingface.co/docs/accelerate)
 package:
 
@@ -342,4 +344,3 @@ gr.Interface.from_pipeline(pipe).launch()
 
 By default, the web demo runs on a local server. If you'd like to share it with others, you can generate a temporary public
 link by setting `share=True` in `launch()`. You can also host your demo on [Hugging Face Spaces](https://huggingface.co/spaces) for a permanent link. 
-

docs/source/en/preprocessing.md

@@ -471,7 +471,7 @@ from [`DetrImageProcessor`] and define a custom `collate_fn` to batch images tog
 
 ## Multimodal
 
-For tasks involving multimodal inputs, you'll need a [processor](main_classes/processors) to prepare your dataset for the model. A processor couples together two processing objects such as as tokenizer and feature extractor.
+For tasks involving multimodal inputs, you'll need a [processor](main_classes/processors) to prepare your dataset for the model. A processor couples together two processing objects such as tokenizer and feature extractor.
 
 Load the [LJ Speech](https://huggingface.co/datasets/lj_speech) dataset (see the 🤗 [Datasets tutorial](https://huggingface.co/docs/datasets/load_hub) for more details on how to load a dataset) to see how you can use a processor for automatic speech recognition (ASR):
 

docs/source/en/tasks/mask_generation.md

@@ -124,6 +124,7 @@ the processor.
 
 ```python
 from transformers import SamModel, SamProcessor
+import torch
 
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
@@ -147,7 +148,6 @@ masks = processor.image_processor.post_process_masks(outputs.pred_masks.cpu(), i
 We can visualize the three masks in the `masks` output.
 
 ```python
-import torch
 import matplotlib.pyplot as plt
 import numpy as np
 
@@ -211,7 +211,7 @@ import matplotlib.patches as patches
 fig, ax = plt.subplots()
 ax.imshow(image)
 
-rectangle = patches.Rectangle((2350, 1600, 500, 500, linewidth=2, edgecolor='r', facecolor='none')
+rectangle = patches.Rectangle((2350, 1600), 500, 500, linewidth=2, edgecolor='r', facecolor='none')
 ax.add_patch(rectangle)
 ax.axis("off")
 plt.show()

docs/source/en/tasks/monocular_depth_estimation.md

@@ -23,23 +23,26 @@ a single camera viewpoint.
 Monocular depth estimation has various applications, including 3D reconstruction, augmented reality, autonomous driving,
 and robotics. It is a challenging task as it requires the model to understand the complex relationships between objects
 in the scene and the corresponding depth information, which can be affected by factors such as lighting conditions,
-occlusion, and texture.
+occlusion, and texture. 
+
+There are two main depth estimation categories:
+
+- **Absolute depth estimation**: This task variant aims to provide exact depth measurements from the camera. The term is used interchangeably with metric depth estimation, where depth is provided in precise measurements in meters or feet. Absolute depth estimation models output depth maps with numerical values that represent real-world distances.
+
+- **Relative depth estimation**: Relative depth estimation aims to predict the depth order of objects or points in a scene without providing the precise measurements. These models output a depth map that indicates which parts of the scene are closer or farther relative to each other without the actual distances to A and B.
+
+In this guide, we will see how to infer with [Depth Anything V2](https://huggingface.co/depth-anything/Depth-Anything-V2-Large), a state-of-the-art zero-shot relative depth estimation model, and [ZoeDepth](https://huggingface.co/docs/transformers/main/en/model_doc/zoedepth), an absolute depth estimation model.
 
 <Tip>
 
-To see all architectures and checkpoints compatible with this task, we recommend checking the [task-page](https://huggingface.co/tasks/depth-anything)
+Check the [Depth Estimation](https://huggingface.co/tasks/depth-estimation) task page to view all compatible architectures and checkpoints.
 
 </Tip>
 
-In this guide you'll learn how to:
-
-* create a depth estimation pipeline
-* run depth estimation inference by hand
-
-Before you begin, make sure you have all the necessary libraries installed:
+Before we begin, we need to install the latest version of Transformers:
 
 ```bash
-pip install -q transformers
+pip install -q -U transformers
 ```
 
 ## Depth estimation pipeline
@@ -49,9 +52,11 @@ Instantiate a pipeline from a [checkpoint on the Hugging Face Hub](https://huggi
 
 ```py
 >>> from transformers import pipeline
+>>> import torch
 
->>> checkpoint = "vinvino02/glpn-nyu"
->>> depth_estimator = pipeline("depth-estimation", model=checkpoint)
+>>> device = "cuda" if torch.cuda.is_available() else "cpu"
+>>> checkpoint = "depth-anything/Depth-Anything-V2-base-hf"
+>>> pipe = pipeline("depth-estimation", model=checkpoint, device=device)
 ```
 
 Next, choose an image to analyze:
@@ -60,19 +65,19 @@ Next, choose an image to analyze:
 >>> from PIL import Image
 >>> import requests
 
->>> url = "https://unsplash.com/photos/HwBAsSbPBDU/download?ixid=MnwxMjA3fDB8MXxzZWFyY2h8MzR8fGNhciUyMGluJTIwdGhlJTIwc3RyZWV0fGVufDB8MHx8fDE2Nzg5MDEwODg&force=true&w=640"
+>>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"
 >>> image = Image.open(requests.get(url, stream=True).raw)
 >>> image
 ```
 
 <div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/depth-estimation-example.jpg" alt="Photo of a busy street"/>
+     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg" alt="Photo of a bee"/>
 </div>
 
 Pass the image to the pipeline.
 
 ```py
->>> predictions = depth_estimator(image)
+>>> predictions = pipe(image)
 ```
 
 The pipeline returns a dictionary with two entries. The first one, called `predicted_depth`, is a tensor with the values
@@ -99,17 +104,17 @@ Here we'll use the same checkpoint as before:
 ```py
 >>> from transformers import AutoImageProcessor, AutoModelForDepthEstimation
 
->>> checkpoint = "vinvino02/glpn-nyu"
+>>> checkpoint = "Intel/zoedepth-nyu-kitti"
 
 >>> image_processor = AutoImageProcessor.from_pretrained(checkpoint)
->>> model = AutoModelForDepthEstimation.from_pretrained(checkpoint)
+>>> model = AutoModelForDepthEstimation.from_pretrained(checkpoint).to(device)
 ```
 
 Prepare the image input for the model using the `image_processor` that will take care of the necessary image transformations
 such as resizing and normalization:
 
 ```py
->>> pixel_values = image_processor(image, return_tensors="pt").pixel_values
+>>> pixel_values = image_processor(image, return_tensors="pt").pixel_values.to(device)
 ```
 
 Pass the prepared inputs through the model:
@@ -119,28 +124,100 @@ Pass the prepared inputs through the model:
 
 >>> with torch.no_grad():
 ...     outputs = model(pixel_values)
-...     predicted_depth = outputs.predicted_depth
 ```
 
-Visualize the results:
+Let's post-process and visualize the results. 
+
+We need to pad and then resize the outputs so that predicted depth map has the same dimension as the original image. After resizing we will remove the padded regions from the depth. 
 
 ```py
 >>> import numpy as np
+>>> import torch.nn.functional as F
 
->>> # interpolate to original size
->>> prediction = torch.nn.functional.interpolate(
-...     predicted_depth.unsqueeze(1),
-...     size=image.size[::-1],
-...     mode="bicubic",
-...     align_corners=False,
-... ).squeeze()
->>> output = prediction.numpy()
+>>> predicted_depth = outputs.predicted_depth.unsqueeze(dim=1)
+>>> height, width = pixel_values.shape[2:]
 
->>> formatted = (output * 255 / np.max(output)).astype("uint8")
->>> depth = Image.fromarray(formatted)
->>> depth
+>>> height_padding_factor = width_padding_factor = 3
+>>> pad_h = int(np.sqrt(height/2) * height_padding_factor)
+>>> pad_w = int(np.sqrt(width/2) * width_padding_factor)
+
+>>> if predicted_depth.shape[-2:] != pixel_values.shape[-2:]:
+>>>    predicted_depth = F.interpolate(predicted_depth, size= (height, width), mode='bicubic', align_corners=False)
+
+>>> if pad_h > 0:
+     predicted_depth = predicted_depth[:, :, pad_h:-pad_h,:]
+>>> if pad_w > 0:
+     predicted_depth = predicted_depth[:, :, :, pad_w:-pad_w]
 ```
 
+We can now visualize the results (the function below is taken from the [GaussianObject](https://github.com/GaussianObject/GaussianObject/blob/ad6629efadb57902d5f8bc0fa562258029a4bdf1/pred_monodepth.py#L11) framework).
+
+```py
+import matplotlib
+
+def colorize(value, vmin=None, vmax=None, cmap='gray_r', invalid_val=-99, invalid_mask=None, background_color=(128, 128, 128, 255), gamma_corrected=False, value_transform=None):
+    """Converts a depth map to a color image.
+
+    Args:
+        value (torch.Tensor, numpy.ndarry): Input depth map. Shape: (H, W) or (1, H, W) or (1, 1, H, W). All singular dimensions are squeezed
+        vmin (float, optional): vmin-valued entries are mapped to start color of cmap. If None, value.min() is used. Defaults to None.
+        vmax (float, optional):  vmax-valued entries are mapped to end color of cmap. If None, value.max() is used. Defaults to None.
+        cmap (str, optional): matplotlib colormap to use. Defaults to 'magma_r'.
+        invalid_val (int, optional): Specifies value of invalid pixels that should be colored as 'background_color'. Defaults to -99.
+        invalid_mask (numpy.ndarray, optional): Boolean mask for invalid regions. Defaults to None.
+        background_color (tuple[int], optional): 4-tuple RGB color to give to invalid pixels. Defaults to (128, 128, 128, 255).
+        gamma_corrected (bool, optional): Apply gamma correction to colored image. Defaults to False.
+        value_transform (Callable, optional): Apply transform function to valid pixels before coloring. Defaults to None.
+
+    Returns:
+        numpy.ndarray, dtype - uint8: Colored depth map. Shape: (H, W, 4)
+    """
+    if isinstance(value, torch.Tensor):
+        value = value.detach().cpu().numpy()
+
+    value = value.squeeze()
+    if invalid_mask is None:
+        invalid_mask = value == invalid_val
+    mask = np.logical_not(invalid_mask)
+
+    # normalize
+    vmin = np.percentile(value[mask],2) if vmin is None else vmin
+    vmax = np.percentile(value[mask],85) if vmax is None else vmax
+    if vmin != vmax:
+        value = (value - vmin) / (vmax - vmin)  # vmin..vmax
+    else:
+        # Avoid 0-division
+        value = value * 0.
+
+    # squeeze last dim if it exists
+    # grey out the invalid values
+
+    value[invalid_mask] = np.nan
+    cmapper = matplotlib.colormaps.get_cmap(cmap)
+    if value_transform:
+        value = value_transform(value)
+        # value = value / value.max()
+    value = cmapper(value, bytes=True)  # (nxmx4)
+
+    # img = value[:, :, :]
+    img = value[...]
+    img[invalid_mask] = background_color
+
+    #     return img.transpose((2, 0, 1))
+    if gamma_corrected:
+        # gamma correction
+        img = img / 255
+        img = np.power(img, 2.2)
+        img = img * 255
+        img = img.astype(np.uint8)
+    return img
+
+>>> result = colorize(predicted_depth.cpu().squeeze().numpy())
+>>> Image.fromarray(result)
+```
+
+
+
 <div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/depth-visualization.png" alt="Depth estimation visualization"/>
+     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/depth-visualization-zoe.png" alt="Depth estimation visualization"/>
 </div>

docs/source/en/testing.md

@@ -184,16 +184,16 @@ pytest -k "test and ada" tests/test_optimization.py
 Sometimes you need to run `accelerate` tests on your models. For that you can just add `-m accelerate_tests` to your command, if let's say you want to run these tests on `OPT` run:
 
 ```bash
-RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py 
+RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py
 ```
 
 
-### Run documentation tests 
+### Run documentation tests
 
-In order to test whether the documentation examples are correct, you should check that the `doctests` are passing. 
-As an example, let's use [`WhisperModel.forward`'s docstring](https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py#L1017-L1035): 
+In order to test whether the documentation examples are correct, you should check that the `doctests` are passing.
+As an example, let's use [`WhisperModel.forward`'s docstring](https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py#L1017-L1035):
 
-```python 
+```python
 r"""
 Returns:
 
@@ -216,8 +216,8 @@ Example:
 
 ```
 
-Just run the following line to automatically test every docstring example in the desired file: 
-```bash 
+Just run the following line to automatically test every docstring example in the desired file:
+```bash
 pytest --doctest-modules <path_to_file_or_dir>
 ```
 If the file has a markdown extention, you should add the `--doctest-glob="*.md"` argument.
@@ -881,7 +881,7 @@ code that's buggy causes some bad state that will affect other tests, do not use
 - Here is how to skip whole test unconditionally:
 
 ```python no-style
-@unittest.skip("this bug needs to be fixed")
+@unittest.skip(reason="this bug needs to be fixed")
 def test_feature_x():
 ```
 
@@ -1011,7 +1011,7 @@ slow models to do qualitative testing. To see the use of these simply look for *
 grep tiny tests examples
 ```
 
-Here is a an example of a [script](https://github.com/huggingface/transformers/tree/main/scripts/fsmt/fsmt-make-tiny-model.py) that created the tiny model
+Here is an example of a [script](https://github.com/huggingface/transformers/tree/main/scripts/fsmt/fsmt-make-tiny-model.py) that created the tiny model
 [stas/tiny-wmt19-en-de](https://huggingface.co/stas/tiny-wmt19-en-de). You can easily adjust it to your specific
 model's architecture.
 

docs/source/fr/_toctree.yml

@@ -15,7 +15,7 @@
       title: Préparation des données
     - local: in_translation
       title: Fine-tune un modèle pré-entraîné
-    - local: in_translation
+    - local: run_scripts_fr
       title: Entraînement avec un script
     - local: in_translation
       title: Entraînement distribué avec 🤗 Accelerate

docs/source/fr/run_scripts_fr.md

@@ -0,0 +1,355 @@
+<!--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.
+
+-->
+
+# Entraîner avec un script
+
+En plus des [notebooks](./notebooks) de 🤗 Transformers, il existe également des exemples de scripts démontrant comment entraîner un modèle pour une tâche avec [PyTorch](https://github.com/huggingface/transformers/tree/main/examples/pytorch), [TensorFlow](https://github.com/huggingface/transformers/tree/main/examples/tensorflow) ou [JAX/Flax](https://github.com/huggingface/transformers/tree/main/examples/flax).
+
+
+Vous trouverez également des scripts que nous avons utilisé dans nos [projets de recherche](https://github.com/huggingface/transformers/tree/main/examples/research_projects) et des [exemples "legacy"](https://github.com/huggingface/transformers/tree/main/examples/legacy) qui sont des contributions de la communauté. Ces scripts ne sont pas activement maintenus et nécessitent une version spécifique de 🤗 Transformers qui sera probablement incompatible avec la dernière version de la librairie.
+
+Les exemples de scripts ne sont pas censés fonctionner immédiatement pour chaque problème, et il se peut que vous ayez besoin d'adapter le script au problème que vous essayez de résoudre. Pour vous aider dans cette tâche, la plupart des scripts exposent entièrement la manière dont les données sont prétraitées, vous permettant de les modifier selon vos besoins.
+
+Pour toute fonctionnalité que vous souhaitez implémenter dans un script d'exemple, veuillez en discuter sur le [forum](https://discuss.huggingface.co/) ou dans une [issue](https://github.com/huggingface/transformers/issues) avant de soumettre une Pull Request. Bien que nous acceptions les corrections de bugs, il est peu probable que nous fusionnions une Pull Request (opération "merge" dans Git) ajoutant plus de fonctionnalités au détriment de la lisibilité.
+
+Ce guide vous montrera comment exécuter un script d'entraînement de résumé en exemple avec [PyTorch](https://github.com/huggingface/transformers/tree/main/examples/pytorch/summarization) et [TensorFlow](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/summarization). Tous les exemples sont censés fonctionner avec les deux frameworks, sauf indication contraire.
+
+## Configuration
+
+Pour exécuter avec succès la dernière version des scripts d'exemple, vous devez **installer 🤗 Transformers à partir du code source** dans un nouvel environnement virtuel :
+
+```bash
+git clone https://github.com/huggingface/transformers
+cd transformers
+pip install .
+```
+
+Pour les versions plus anciennes des exemples de scripts, cliquez sur le bouton ci-dessous :
+
+<details>
+  <summary>Exemples pour les anciennes versions de Transformers 🤗</summary>
+	<ul>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.5.1/examples">v4.5.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.4.2/examples">v4.4.2</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.3.3/examples">v4.3.3</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.2.2/examples">v4.2.2</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.1.1/examples">v4.1.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v4.0.1/examples">v4.0.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.5.1/examples">v3.5.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.4.0/examples">v3.4.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.3.1/examples">v3.3.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.2.0/examples">v3.2.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.1.0/examples">v3.1.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v3.0.2/examples">v3.0.2</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.11.0/examples">v2.11.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.10.0/examples">v2.10.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.9.1/examples">v2.9.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.8.0/examples">v2.8.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.7.0/examples">v2.7.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.6.0/examples">v2.6.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.5.1/examples">v2.5.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.4.0/examples">v2.4.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.3.0/examples">v2.3.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.2.0/examples">v2.2.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.1.0/examples">v2.1.1</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v2.0.0/examples">v2.0.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v1.2.0/examples">v1.2.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v1.1.0/examples">v1.1.0</a></li>
+		<li><a href="https://github.com/huggingface/transformers/tree/v1.0.0/examples">v1.0.0</a></li>
+	</ul>
+</details>
+
+Ensuite, changez votre clone actuel de  🤗 Transformers pour une version spécifique, comme par exemple v3.5.1 :
+
+```bash
+git checkout tags/v3.5.1
+```
+
+Après avoir configuré la bonne version de la librairie, accédez au dossier d'exemple de votre choix et installez les prérequis spécifiques à l'exemple.
+
+```bash
+pip install -r requirements.txt
+```
+
+## Exécuter un script
+
+<frameworkcontent>
+<pt>
+
+Le script d'exemple télécharge et prétraite un jeu de données à partir de la bibliothèque 🤗 [Datasets](https://huggingface.co/docs/datasets/). Ensuite, le script affine un ensemble de données à l'aide de [Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) sur une architecture qui prend en charge la tâche de résumé. L'exemple suivant montre comment ajuster le modèle [T5-small](https://huggingface.co/google-t5/t5-small) sur les données [CNN/DailyMail](https://huggingface.co/datasets/cnn_dailymail). Le modèle T5 nécessite un argument supplémentaire `source_prefix` en raison de la façon dont il a été entraîné. Cette invite permet à T5 de savoir qu'il s'agit d'une tâche de résumé.
+
+```bash
+python examples/pytorch/summarization/run_summarization.py \
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+</pt>
+<tf>
+
+Le script d'exemple télécharge et prétraite un jeu de données à partir de la bibliothèque  🤗 [Datasets](https://huggingface.co/docs/datasets/). Ensuite, le script ajuste un modèle à l'aide de Keras sur une architecture qui prend en charge la tâche de résumé. L'exemple suivant montre comment ajuster le modèle [T5-small](https://huggingface.co/google-t5/t5-small) sur le jeu de données [CNN/DailyMail](https://huggingface.co/datasets/cnn_dailymail). Le modèle T5 nécessite un argument supplémentaire source_prefix en raison de la façon dont il a été entraîné. Cette invite permet à T5 de savoir qu'il s'agit d'une tâche de résumé.
+
+```bash
+python examples/tensorflow/summarization/run_summarization.py  \
+    --model_name_or_path google-t5/t5-small \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --output_dir /tmp/tst-summarization  \
+    --per_device_train_batch_size 8 \
+    --per_device_eval_batch_size 16 \
+    --num_train_epochs 3 \
+    --do_train \
+    --do_eval
+```
+</tf>
+</frameworkcontent>
+
+## Entraînement distribué et précision mixte
+
+[Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) prend en charge l'entraînement distribué et la précision mixte, ce qui signifie que vous pouvez également les utiliser dans un script. Pour activer ces deux fonctionnalités :
+
+- Ajoutez l'argument fp16 pour activer la précision mixte.
+- Définissez le nombre de GPU à utiliser avec l'argument `nproc_per_node`.
+
+```bash
+torchrun \
+    --nproc_per_node 8 pytorch/summarization/run_summarization.py \
+    --fp16 \
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+
+Les scripts TensorFlow utilisent une Strategie en Miroir [`MirroredStrategy`](https://www.tensorflow.org/guide/distributed_training#mirroredstrategy) pour l'entraînement distribué, et vous n'avez pas besoin d'ajouter d'arguments supplémentaires au script d'entraînement. Le script TensorFlow utilisera plusieurs GPU par défaut s'ils sont disponibles.
+
+## Exécuter un script sur un TPU 
+
+<frameworkcontent>
+<pt>
+
+Les unités de traitement de tenseurs (UTT) (TPU) sont spécialement conçues pour accélérer les performances. PyTorch prend en charge les TPU avec le compilateur de deep learning [XLA](https://www.tensorflow.org/xla). Pour utiliser un TPU, lancez le script xla_spawn.py et utilisez l'argument num_cores pour définir le nombre de cœurs TPU que vous souhaitez utilise
+
+```bash
+python xla_spawn.py --num_cores 8 \
+    summarization/run_summarization.py \
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+</pt>
+<tf>
+Les scripts TensorFlow utilisent une [`TPUStrategy`](https://www.tensorflow.org/guide/distributed_training#tpustrategy) pour l'entraînement sur TPU. Pour utiliser un TPU, passez le nom de la ressource TPU à l'argument tpu.
+
+```bash
+python run_summarization.py  \
+    --tpu name_of_tpu_resource \
+    --model_name_or_path google-t5/t5-small \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --output_dir /tmp/tst-summarization  \
+    --per_device_train_batch_size 8 \
+    --per_device_eval_batch_size 16 \
+    --num_train_epochs 3 \
+    --do_train \
+    --do_eval
+```
+</tf>
+</frameworkcontent>
+
+## Exécuter un script avec 🤗 Accelerate 
+
+🤗 [Accelerate](https://huggingface.co/docs/accelerate) est une bibliothèque uniquement pour PyTorch qui offre une méthode unifiée pour entraîner un modèle sur plusieurs types de configurations (CPU uniquement, plusieurs GPU, TPU) tout en maintenant une visibilité complète sur la boucle d'entraînement PyTorch. Assurez-vous que vous avez installé 🤗 Accelerate si ce n'est pas déjà le cas.
+
+> Note : Comme Accelerate est en développement rapide, la version git d'accelerate doit être installée pour exécuter les scripts.
+```bash
+pip install git+https://github.com/huggingface/accelerate
+```
+
+Au lieu du script `run_summarization.py`, vous devez utiliser le script `run_summarization_no_trainer.py`. Les scripts compatibles avec 🤗 Accelerate auront un fichier `task_no_trainer.py` dans le dossier. Commencez par exécuter la commande suivante pour créer et enregistrer un fichier de configuration.
+
+```bash
+accelerate config
+```
+
+Testez votre configuration pour vous assurer qu'elle est correctement configurée :
+
+```bash
+accelerate test
+```
+
+Maintenant, vous êtes prêt à lancer l'entraînement :
+
+```bash
+accelerate launch run_summarization_no_trainer.py \
+    --model_name_or_path google-t5/t5-small \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir ~/tmp/tst-summarization
+```
+
+## Utiliser un jeu de données personnalisé 
+
+Le script de résumé prend en charge les jeux de données personnalisés tant qu'ils sont au format CSV ou JSON Line. Lorsque vous utilisez votre propre jeu de données, vous devez spécifier plusieurs arguments supplémentaires :
+
+- `train_file` et `validation_file` spécifient le chemin vers vos fichiers d'entraînement et de validation.
+- `text_column` est le texte d'entrée à résumer.
+- `summary_column` est le texte cible à produire.
+
+Un exemple de script de résumé utilisant un ensemble de données personnalisé ressemblerait à ceci :
+
+```bash
+python examples/pytorch/summarization/run_summarization.py \
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --train_file path_to_csv_or_jsonlines_file \
+    --validation_file path_to_csv_or_jsonlines_file \
+    --text_column text_column_name \
+    --summary_column summary_column_name \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --overwrite_output_dir \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --predict_with_generate
+```
+
+## Tester un script
+Il est souvent judicieux d'exécuter votre script sur un plus petit nombre d'exemples de jeu de données pour s'assurer que tout fonctionne comme prévu avant de s'engager sur un jeu de données complet qui pourrait prendre des heures à traiter. Utilisez les arguments suivants pour tronquer le jeu de données à un nombre maximal d'échantillons :
+
+- `max_train_samples`
+- `max_eval_samples`
+- `max_predict_samples`
+
+```bash
+python examples/pytorch/summarization/run_summarization.py \
+    --model_name_or_path google-t5/t5-small \
+    --max_train_samples 50 \
+    --max_eval_samples 50 \
+    --max_predict_samples 50 \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```
+
+Tous les scripts d'exemple ne prennent pas en charge l'argument `max_predict_samples`. Si vous n'êtes pas sûr que votre script prenne en charge cet argument, ajoutez l'argument `-h` pour vérifier.
+
+```bash
+examples/pytorch/summarization/run_summarization.py -h
+```
+
+## Reprendre l'entraînement à partir d'un point de contrôle
+
+Une autre option utile est de reprendre l'entraînement à partir d'un point de contrôle précédent. Cela vous permettra de reprendre là où vous vous étiez arrêté sans recommencer si votre entraînement est interrompu. Il existe deux méthodes pour reprendre l'entraînement à partir d'un point de contrôle.
+
+La première méthode utilise l'argument `output_dir previous_output_dir` pour reprendre l'entraînement à partir du dernier point de contrôle stocké dans `output_dir`. Dans ce cas, vous devez supprimer l'argument `overwrite_output_dir`.
+
+```bash
+python examples/pytorch/summarization/run_summarization.py
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --output_dir previous_output_dir \
+    --predict_with_generate
+```
+
+La seconde méthode utilise l'argument `resume_from_checkpoint path_to_specific_checkpoint` pour reprendre l'entraînement à partir d'un dossier de point de contrôle spécifique.
+
+```bash
+python examples/pytorch/summarization/run_summarization.py
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --resume_from_checkpoint path_to_specific_checkpoint \
+    --predict_with_generate
+```
+
+## Partage ton modèle
+
+Tous les scripts peuvent télécharger votre modèle final sur le Model Hub. Assurez-vous que vous êtes connecté à Hugging Face avant de commencer :
+
+```bash
+huggingface-cli login
+```
+
+Ensuite, ajoutez l'argument `push_to_hub` au script. Cet argument créera un dépôt avec votre nom d'utilisateur Hugging Face et le nom du dossier spécifié dans `output_dir`.
+
+
+Pour donner un nom spécifique à votre dépôt, utilisez l'argument `push_to_hub_model_id` pour l'ajouter. Le dépôt sera automatiquement listé sous votre namespace. 
+
+L'exemple suivant montre comment télécharger un modèle avec un nom de dépôt spécifique :
+
+```bash
+python examples/pytorch/summarization/run_summarization.py
+    --model_name_or_path google-t5/t5-small \
+    --do_train \
+    --do_eval \
+    --dataset_name cnn_dailymail \
+    --dataset_config "3.0.0" \
+    --source_prefix "summarize: " \
+    --push_to_hub \
+    --push_to_hub_model_id finetuned-t5-cnn_dailymail \
+    --output_dir /tmp/tst-summarization \
+    --per_device_train_batch_size=4 \
+    --per_device_eval_batch_size=4 \
+    --overwrite_output_dir \
+    --predict_with_generate
+```

docs/source/ja/main_classes/callback.md

@@ -35,7 +35,7 @@ rendered properly in your Markdown viewer.
 - [`~integrations.TensorBoardCallback`] (PyTorch >= 1.4 を介して) tensorboard にアクセスできる場合
   またはテンソルボードX）。
 - [`~integrations.WandbCallback`] [wandb](https://www.wandb.com/) がインストールされている場合。
-- [`~integrations.CometCallback`] [comet_ml](https://www.comet.ml/site/) がインストールされている場合。
+- [`~integrations.CometCallback`] [comet_ml](https://www.comet.com/site/) がインストールされている場合。
 - [mlflow](https://www.mlflow.org/) がインストールされている場合は [`~integrations.MLflowCallback`]。
 - [`~integrations.NeptuneCallback`] [neptune](https://neptune.ai/) がインストールされている場合。
 - [`~integrations.AzureMLCallback`] [azureml-sdk](https://pypi.org/project/azureml-sdk/) の場合

docs/source/ja/testing.md

@@ -171,16 +171,16 @@ pytest -k "test and ada" tests/test_optimization.py
 時々、モデルに対して `accelerate` テストを実行する必要があります。たとえば、`OPT` 実行に対してこれらのテストを実行したい場合、コマンドに `-m accelerate_tests` を追加するだけで済みます：
 
 ```bash
-RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py 
+RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py
 ```
 
-### Run documentation tests 
+### Run documentation tests
 
 ドキュメンテーションの例が正しいかどうかをテストするには、`doctests` が合格しているかを確認する必要があります。
 例として、[`WhisperModel.forward` のドックストリング](https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py#L1017-L1035)を使用しましょう。
 
 
-```python 
+```python
 r"""
 Returns:
 
@@ -205,7 +205,7 @@ Example:
 
 指定したファイル内のすべてのドックストリング例を自動的にテストするために、以下の行を実行してください：
 
-```bash 
+```bash
 pytest --doctest-modules <path_to_file_or_dir>
 ```
 
@@ -809,7 +809,7 @@ with ExtendSysPath(f"{bindir}/.."):
 
 
 ```python no-style
-@unittest.skip("this bug needs to be fixed")
+@unittest.skip(reason="this bug needs to be fixed")
 def test_feature_x():
 ```
 
@@ -1211,4 +1211,3 @@ cmd_that_may_fail || true
 
 - [Github Actions:](https://github.com/actions/toolkit/issues/399)
 - [CircleCI:](https://ideas.circleci.com/ideas/CCI-I-344)
-

docs/source/ko/testing.md

@@ -26,19 +26,19 @@ rendered properly in your Markdown viewer.
 
 ## Transformers 테스트 방법[[how-transformers-are-tested]]
 
-1. PR이 제출되면 9개의 CircleCi 작업으로 테스트가 진행됩니다. 해당 PR에 대해 새로운 커밋이 생성될 때마다 테스트는 다시 진행됩니다. 이 작업들은 
-   이 [config 파일](https://github.com/huggingface/transformers/tree/main/.circleci/config.yml)에 정의되어 있으므로 필요하다면 
+1. PR이 제출되면 9개의 CircleCi 작업으로 테스트가 진행됩니다. 해당 PR에 대해 새로운 커밋이 생성될 때마다 테스트는 다시 진행됩니다. 이 작업들은
+   이 [config 파일](https://github.com/huggingface/transformers/tree/main/.circleci/config.yml)에 정의되어 있으므로 필요하다면
    사용자의 로컬 환경에서 동일하게 재현해 볼 수 있습니다.
 
    이 CI 작업은 `@slow` 테스트를 실행하지 않습니다.
 
 2. [github actions](https://github.com/huggingface/transformers/actions)에 의해 실행되는 작업은 3개입니다:
 
-   - [torch hub integration](https://github.com/huggingface/transformers/tree/main/.github/workflows/github-torch-hub.yml): 
+   - [torch hub integration](https://github.com/huggingface/transformers/tree/main/.github/workflows/github-torch-hub.yml):
     torch hub integration이 작동하는지 확인합니다.
 
-   - [self-hosted (push)](https://github.com/huggingface/transformers/tree/main/.github/workflows/self-push.yml): `main` 브랜치에서 커밋이 업데이트된 경우에만 GPU를 이용한 빠른 테스트를 실행합니다. 
-    이는 `src`, `tests`, `.github` 폴더 중 하나에 코드가 업데이트된 경우에만 실행됩니다. 
+   - [self-hosted (push)](https://github.com/huggingface/transformers/tree/main/.github/workflows/self-push.yml): `main` 브랜치에서 커밋이 업데이트된 경우에만 GPU를 이용한 빠른 테스트를 실행합니다.
+    이는 `src`, `tests`, `.github` 폴더 중 하나에 코드가 업데이트된 경우에만 실행됩니다.
     (model card, notebook, 기타 등등을 추가한 경우 실행되지 않도록 하기 위해서입니다)
 
    - [self-hosted runner](https://github.com/huggingface/transformers/tree/main/.github/workflows/self-scheduled.yml): `tests` 및 `examples`에서
@@ -61,7 +61,7 @@ RUN_SLOW=1 pytest examples/
 
 ### 실행할 테스트 선택[[choosing-which-tests-to-run]]
 
-이 문서는 테스트를 실행하는 다양한 방법에 대해 자세히 설명합니다. 
+이 문서는 테스트를 실행하는 다양한 방법에 대해 자세히 설명합니다.
 모든 내용을 읽은 후에도, 더 자세한 내용이 필요하다면 [여기](https://docs.pytest.org/en/latest/usage.html)에서 확인할 수 있습니다.
 
 다음은 가장 유용한 테스트 실행 방법 몇 가지입니다.
@@ -186,7 +186,7 @@ pytest -k "test and ada" tests/test_optimization.py
 모델에서 `accelerate` 테스트를 실행해야 할 때가 있습니다. 이를 위해서는 명령어에 `-m accelerate_tests`를 추가하면 됩니다.
 예를 들어, `OPT`에서 이러한 테스트를 실행하려면 다음과 같습니다:
 ```bash
-RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py 
+RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py
 ```
 
 ### 문서 테스트 실행[[run-documentation-tests]]
@@ -194,7 +194,7 @@ RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py
 예시 문서가 올바른지 테스트하려면 `doctests`가 통과하는지 확인해야 합니다.
 예를 들어, [`WhisperModel.forward`'s docstring](https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py#L1017-L1035)를 사용해 봅시다:
 
-```python 
+```python
 r"""
 Returns:
 
@@ -218,7 +218,7 @@ Example:
 ```
 
 원하는 파일의 모든 docstring 예제를 자동으로 테스트하려면 다음 명령을 실행하면 됩니다:
-```bash 
+```bash
 pytest --doctest-modules <path_to_file_or_dir>
 ```
 파일의 확장자가 markdown인 경우 `--doctest-glob="*.md"` 인수를 추가해야 합니다.
@@ -240,9 +240,9 @@ pytest --picked
 
 ### 소스 수정 시 실패한 테스트 자동 재실행[[automatically-rerun-failed-tests-on-source-modification]]
 
-[pytest-xdist](https://github.com/pytest-dev/pytest-xdist)는 모든 실패한 테스트를 감지하고, 
+[pytest-xdist](https://github.com/pytest-dev/pytest-xdist)는 모든 실패한 테스트를 감지하고,
 파일을 수정한 후에 파일을 계속 재실행하여 테스트가 성공할 때까지 기다리는 매우 유용한 기능을 제공합니다.
-따라서 수정한 내용을 확인한 후 pytest를 다시 시작할 필요가 없습니다. 
+따라서 수정한 내용을 확인한 후 pytest를 다시 시작할 필요가 없습니다.
 모든 테스트가 통과될 때까지 이 과정을 반복한 후 다시 전체 실행이 이루어집니다.
 
 ```bash
@@ -252,7 +252,7 @@ pip install pytest-xdist
 재귀적 모드의 사용: `pytest -f` 또는 `pytest --looponfail`
 
 파일의 변경 사항은 `looponfailroots` 루트 디렉터리와 해당 내용을 (재귀적으로) 확인하여 감지됩니다.
-이 값의 기본값이 작동하지 않는 경우, 
+이 값의 기본값이 작동하지 않는 경우,
 `setup.cfg`의 설정 옵션을 변경하여 프로젝트에서 변경할 수 있습니다:
 
 ```ini
@@ -275,7 +275,7 @@ looponfailroots = transformers tests
 
 ### 특정 테스트 모듈 건너뛰기[[skip-a-test-module]]
 
-모든 테스트 모듈을 실행하되 특정 모듈을 제외하려면, 실행할 테스트 목록을 명시적으로 지정할 수 있습니다. 
+모든 테스트 모듈을 실행하되 특정 모듈을 제외하려면, 실행할 테스트 목록을 명시적으로 지정할 수 있습니다.
 예를 들어, `test_modeling_*.py` 테스트를 제외한 모든 테스트를 실행하려면 다음을 사용할 수 있습니다:
 
 ```bash
@@ -292,19 +292,19 @@ pytest --cache-clear tests
 
 ### 테스트를 병렬로 실행[[running-tests-in-parallel]]
 
-이전에 언급한 것처럼 `make test`는 테스트를 병렬로 실행하기 위해 
+이전에 언급한 것처럼 `make test`는 테스트를 병렬로 실행하기 위해
 `pytest-xdist` 플러그인(`-n X` 인수, 예를 들어 `-n 2`를 사용하여 2개의 병렬 작업 실행)을 통해 실행됩니다.
 
-`pytest-xdist`의 `--dist=` 옵션을 사용하여 테스트를 어떻게 그룹화할지 제어할 수 있습니다. 
+`pytest-xdist`의 `--dist=` 옵션을 사용하여 테스트를 어떻게 그룹화할지 제어할 수 있습니다.
 `--dist=loadfile`은 하나의 파일에 있는 테스트를 동일한 프로세스로 그룹화합니다.
 
 실행된 테스트의 순서가 다르고 예측할 수 없기 때문에, `pytest-xdist`로 테스트 스위트를 실행하면 실패가 발생할 수 있습니다 (검출되지 않은 결합된 테스트가 있는 경우).
-이 경우 [pytest-replay](https://github.com/ESSS/pytest-replay)를 사용하면 동일한 순서로 테스트를 다시 실행해서 
+이 경우 [pytest-replay](https://github.com/ESSS/pytest-replay)를 사용하면 동일한 순서로 테스트를 다시 실행해서
 실패하는 시퀀스를 최소화하는 데에 도움이 됩니다.
 
 ### 테스트 순서와 반복[[test-order-and-repetition]]
 
-잠재적인 종속성 및 상태 관련 버그(tear down)를 감지하기 위해 
+잠재적인 종속성 및 상태 관련 버그(tear down)를 감지하기 위해
 테스트를 여러 번, 연속으로, 무작위로 또는 세트로 반복하는 것이 좋습니다.
 그리고 직접적인 여러 번의 반복은 DL의 무작위성에 의해 발견되는 일부 문제를 감지하는 데에도 유용합니다.
 
@@ -341,10 +341,10 @@ pytest --flake-finder --flake-runs=5 tests/test_failing_test.py
 pip install pytest-random-order
 ```
 
-중요: `pytest-random-order`가 설치되면 테스트가 자동으로 임의의 순서로 섞입니다. 
+중요: `pytest-random-order`가 설치되면 테스트가 자동으로 임의의 순서로 섞입니다.
 구성 변경이나 커맨드 라인 옵션이 필요하지 않습니다.
 
-앞서 설명한 것처럼 이를 통해 한 테스트의 상태가 다른 테스트의 상태에 영향을 미치는 결합된 테스트를 감지할 수 있습니다. 
+앞서 설명한 것처럼 이를 통해 한 테스트의 상태가 다른 테스트의 상태에 영향을 미치는 결합된 테스트를 감지할 수 있습니다.
 `pytest-random-order`가 설치되면 해당 세션에서 사용된 랜덤 시드가 출력되며 예를 들어 다음과 같습니다:
 
 ```bash
@@ -364,7 +364,7 @@ Using --random-order-seed=573663
 ```
 
 정확히 동일한 테스트 목록(또는 목록이 없음)을 사용하는 경우에만 정확한 순서를 재현합니다.
-목록을 수동으로 좁히기 시작하면 더 이상 시드에 의존할 수 없고 실패했던 정확한 순서로 수동으로 목록을 나열해야합니다. 그리고 `--random-order-bucket=none`을 사용하여 pytest에게 순서를 임의로 설정하지 않도록 알려야 합니다. 
+목록을 수동으로 좁히기 시작하면 더 이상 시드에 의존할 수 없고 실패했던 정확한 순서로 수동으로 목록을 나열해야합니다. 그리고 `--random-order-bucket=none`을 사용하여 pytest에게 순서를 임의로 설정하지 않도록 알려야 합니다.
 예를 들어 다음과 같습니다:
 
 ```bash
@@ -377,19 +377,19 @@ pytest --random-order-bucket=none tests/test_a.py tests/test_c.py tests/test_b.p
 pytest --random-order-bucket=none
 ```
 
-기본적으로 `--random-order-bucket=module`이 내재되어 있으므로, 모듈 수준에서 파일을 섞습니다. 
+기본적으로 `--random-order-bucket=module`이 내재되어 있으므로, 모듈 수준에서 파일을 섞습니다.
 또한 `class`, `package`, `global` 및 `none` 수준에서도 섞을 수 있습니다.
 자세한 내용은 해당 [문서](https://github.com/jbasko/pytest-random-order)를 참조하세요.
 
 또 다른 무작위화의 대안은 [`pytest-randomly`](https://github.com/pytest-dev/pytest-randomly)입니다.
-이 모듈은 매우 유사한 기능/인터페이스를 가지고 있지만, `pytest-random-order`에 있는 버킷 모드를 사용할 수는 없습니다. 
+이 모듈은 매우 유사한 기능/인터페이스를 가지고 있지만, `pytest-random-order`에 있는 버킷 모드를 사용할 수는 없습니다.
 설치 후에는 자동으로 적용되는 문제도 동일하게 가집니다.
 
 ### 외관과 느낌을 변경[[look-and-feel-variations]
 
 #### pytest-sugar 사용[[pytest-sugar]]
 
-[pytest-sugar](https://github.com/Frozenball/pytest-sugar)는 테스트가 보여지는 형태를 개선하고, 
+[pytest-sugar](https://github.com/Frozenball/pytest-sugar)는 테스트가 보여지는 형태를 개선하고,
 진행 상황 바를 추가하며, 실패한 테스트와 검증을 즉시 표시하는 플러그인입니다. 설치하면 자동으로 활성화됩니다.
 
 ```bash
@@ -416,7 +416,7 @@ pytest --pspec tests/test_optimization.py
 
 #### 실패한 테스트 즉시 표시[[instantly-shows-failed-tests]]
 
-[pytest-instafail](https://github.com/pytest-dev/pytest-instafail)은 테스트 세션의 끝까지 기다리지 않고 
+[pytest-instafail](https://github.com/pytest-dev/pytest-instafail)은 테스트 세션의 끝까지 기다리지 않고
 실패 및 오류를 즉시 표시합니다.
 
 ```bash
@@ -435,7 +435,7 @@ GPU가 활성화된 환경에서, CPU 전용 모드로 테스트하려면 `CUDA_
 CUDA_VISIBLE_DEVICES="" pytest tests/utils/test_logging.py
 ```
 
-또는 다중 GPU가 있는 경우 `pytest`에서 사용할 GPU를 지정할 수도 있습니다. 
+또는 다중 GPU가 있는 경우 `pytest`에서 사용할 GPU를 지정할 수도 있습니다.
 예를 들어, GPU `0` 및 `1`이 있는 경우 다음을 실행할 수 있습니다:
 
 ```bash
@@ -444,7 +444,7 @@ CUDA_VISIBLE_DEVICES="1" pytest tests/utils/test_logging.py
 
 이렇게 하면 다른 GPU에서 다른 작업을 실행하려는 경우 유용합니다.
 
-일부 테스트는 반드시 CPU 전용으로 실행해야 하며, 일부는 CPU 또는 GPU 또는 TPU에서 실행해야 하고, 일부는 여러 GPU에서 실행해야 합니다. 
+일부 테스트는 반드시 CPU 전용으로 실행해야 하며, 일부는 CPU 또는 GPU 또는 TPU에서 실행해야 하고, 일부는 여러 GPU에서 실행해야 합니다.
 다음 스킵 데코레이터는 테스트의 요구 사항을 CPU/GPU/TPU별로 설정하는 데 사용됩니다:
 
 - `require_torch` - 이 테스트는 torch에서만 실행됩니다.
@@ -480,7 +480,7 @@ def test_example_with_multi_gpu():
 def test_tf_thing_with_tensorflow():
 ```
 
-이러한 데코레이터는 중첩될 수 있습니다. 
+이러한 데코레이터는 중첩될 수 있습니다.
 예를 들어, 느린 테스트로 진행되고 pytorch에서 적어도 하나의 GPU가 필요한 경우 다음과 같이 설정할 수 있습니다:
 
 ```python no-style
@@ -489,7 +489,7 @@ def test_tf_thing_with_tensorflow():
 def test_example_slow_on_gpu():
 ```
 
-`@parametrized`와 같은 일부 데코레이터는 테스트 이름을 다시 작성하기 때문에 `@require_*` 스킵 데코레이터는 올바르게 작동하려면 항상 맨 마지막에 나열되어야 합니다. 
+`@parametrized`와 같은 일부 데코레이터는 테스트 이름을 다시 작성하기 때문에 `@require_*` 스킵 데코레이터는 올바르게 작동하려면 항상 맨 마지막에 나열되어야 합니다.
 다음은 올바른 사용 예입니다:
 
 ```python no-style
@@ -498,7 +498,7 @@ def test_example_slow_on_gpu():
 def test_integration_foo():
 ```
 
-`@pytest.mark.parametrize`에는 이러한 순서 문제는 없으므로 처음 혹은 마지막에 위치시킬 수 있고 이러한 경우에도 잘 작동할 것입니다. 
+`@pytest.mark.parametrize`에는 이러한 순서 문제는 없으므로 처음 혹은 마지막에 위치시킬 수 있고 이러한 경우에도 잘 작동할 것입니다.
 하지만 unittest가 아닌 경우에만 작동합니다.
 
 테스트 내부에서 다음을 사용할 수 있습니다:
@@ -513,7 +513,7 @@ n_gpu = get_gpu_count()  #torch와 tf와 함께 작동
 
 ### 분산 훈련[[distributed-training]]
 
-`pytest`는 분산 훈련을 직접적으로 다루지 못합니다. 
+`pytest`는 분산 훈련을 직접적으로 다루지 못합니다.
 이를 시도하면 하위 프로세스가 올바른 작업을 수행하지 않고 `pytest`라고 생각하기에 테스트 스위트를 반복해서 실행하게 됩니다.
 그러나 일반 프로세스를 생성한 다음 여러 워커를 생성하고 IO 파이프를 관리하도록 하면 동작합니다.
 
@@ -532,7 +532,7 @@ CUDA_VISIBLE_DEVICES=0,1 RUN_SLOW=1 pytest -sv tests/test_trainer_distributed.py
 
 ### 출력 캡처[[output-capture]]
 
-테스트 실행 중 `stdout` 및 `stderr`로 전송된 모든 출력이 캡처됩니다. 
+테스트 실행 중 `stdout` 및 `stderr`로 전송된 모든 출력이 캡처됩니다.
 테스트나 설정 메소드가 실패하면 캡처된 출력은 일반적으로 실패 추적 정보와 함께 표시됩니다.
 
 출력 캡처를 비활성화하고 `stdout` 및 `stderr`를 정상적으로 받으려면 `-s` 또는 `--capture=no`를 사용하세요:
@@ -563,7 +563,7 @@ pytest --color=no tests/utils/test_logging.py
 pytest --pastebin=failed tests/utils/test_logging.py
 ```
 
-이렇게 하면 각 실패에 대한 URL을 제공하는 remote Paste service에 테스트 실행 정보를 제출합니다. 
+이렇게 하면 각 실패에 대한 URL을 제공하는 remote Paste service에 테스트 실행 정보를 제출합니다.
 일반적인 테스트를 선택할 수도 있고 혹은 특정 실패만 보내려면 `-x`와 같이 추가할 수도 있습니다.
 
 전체 테스트 세션 로그에 대한 URL을 생성합니다:
@@ -574,17 +574,17 @@ pytest --pastebin=all tests/utils/test_logging.py
 
 ## 테스트 작성[[writing-tests]]
 
-🤗 transformers 테스트는 대부분 `unittest`를 기반으로 하지만, 
+🤗 transformers 테스트는 대부분 `unittest`를 기반으로 하지만,
 `pytest`에서 실행되므로 대부분의 경우 두 시스템의 기능을 사용할 수 있습니다.
 
-지원되는 기능에 대해 [여기](https://docs.pytest.org/en/stable/unittest.html)에서 확인할 수 있지만, 
+지원되는 기능에 대해 [여기](https://docs.pytest.org/en/stable/unittest.html)에서 확인할 수 있지만,
 기억해야 할 중요한 점은 대부분의 `pytest` fixture가 작동하지 않는다는 것입니다.
 파라미터화도 작동하지 않지만, 우리는 비슷한 방식으로 작동하는 `parameterized` 모듈을 사용합니다.
 
 
 ### 매개변수화[[parametrization]]
 
-동일한 테스트를 다른 인수로 여러 번 실행해야 하는 경우가 종종 있습니다. 
+동일한 테스트를 다른 인수로 여러 번 실행해야 하는 경우가 종종 있습니다.
 테스트 내에서 이 작업을 수행할 수 있지만, 그렇게 하면 하나의 인수 세트에 대해 테스트를 실행할 수 없습니다.
 
 ```python
@@ -605,7 +605,7 @@ class TestMathUnitTest(unittest.TestCase):
         assert_equal(math.floor(input), expected)
 ```
 
-이제 기본적으로 이 테스트는 `test_floor`의 마지막 3개 인수가 
+이제 기본적으로 이 테스트는 `test_floor`의 마지막 3개 인수가
 매개변수 목록의 해당 인수에 할당되는 것으로 3번 실행될 것입니다.
 
 그리고 `negative` 및 `integer` 매개변수 집합만 실행하려면 다음과 같이 실행할 수 있습니다:
@@ -620,7 +620,7 @@ pytest -k "negative and integer" tests/test_mytest.py
 pytest -k "not negative" tests/test_mytest.py
 ```
 
-앞에서 언급한 `-k` 필터를 사용하는 것 외에도, 
+앞에서 언급한 `-k` 필터를 사용하는 것 외에도,
 각 서브 테스트의 정확한 이름을 확인한 후에 일부 혹은 전체 서브 테스트를 실행할 수 있습니다.
 
 ```bash
@@ -641,10 +641,10 @@ test_this1.py::TestMathUnitTest::test_floor_2_large_fraction
 pytest test_this1.py::TestMathUnitTest::test_floor_0_negative  test_this1.py::TestMathUnitTest::test_floor_1_integer
 ```
 
-`transformers`의 개발자 종속성에 이미 있는 [parameterized](https://pypi.org/project/parameterized/) 모듈은 
+`transformers`의 개발자 종속성에 이미 있는 [parameterized](https://pypi.org/project/parameterized/) 모듈은
 `unittests`와 `pytest` 테스트 모두에서 작동합니다.
 
-그러나 테스트가 `unittest`가 아닌 경우 `pytest.mark.parametrize`를 사용할 수 있습니다(이미 있는 일부 테스트에서 사용되는 경우도 있습니다. 
+그러나 테스트가 `unittest`가 아닌 경우 `pytest.mark.parametrize`를 사용할 수 있습니다(이미 있는 일부 테스트에서 사용되는 경우도 있습니다.
 주로 `examples` 하위에 있습니다).
 
 다음은 `pytest`의 `parametrize` 마커를 사용한 동일한 예입니다:
@@ -666,8 +666,8 @@ def test_floor(name, input, expected):
     assert_equal(math.floor(input), expected)
 ```
 
-`parameterized`와 마찬가지로 `pytest.mark.parametrize`를 사용하면 
-`-k` 필터가 작동하지 않는 경우에도 실행할 서브 테스트를 정확하게 지정할 수 있습니다. 
+`parameterized`와 마찬가지로 `pytest.mark.parametrize`를 사용하면
+`-k` 필터가 작동하지 않는 경우에도 실행할 서브 테스트를 정확하게 지정할 수 있습니다.
 단, 이 매개변수화 함수는 서브 테스트의 이름 집합을 약간 다르게 생성합니다. 다음과 같은 모습입니다:
 
 ```bash
@@ -694,7 +694,7 @@ pytest test_this2.py::test_floor[negative--1.5--2.0] test_this2.py::test_floor[i
 
 ### 파일 및 디렉터리[[files-and-directories]]
 
-테스트에서 종종 현재 테스트 파일과 관련된 상대적인 위치를 알아야 하는 경우가 있습니다. 
+테스트에서 종종 현재 테스트 파일과 관련된 상대적인 위치를 알아야 하는 경우가 있습니다.
 테스트가 여러 디렉터리에서 호출되거나 깊이가 다른 하위 디렉터리에 있을 수 있기 때문에 그 위치를 아는 것은 간단하지 않습니다.
 `transformers.test_utils.TestCasePlus`라는 헬퍼 클래스는 모든 기본 경로를 처리하고 간단한 액세서를 제공하여 이 문제를 해결합니다:
 
@@ -717,7 +717,7 @@ pytest test_this2.py::test_floor[negative--1.5--2.0] test_this2.py::test_floor[i
   - `repo_root_dir_str`
   - `src_dir_str`
 
-위의 내용을 사용하려면 테스트가 'transformers.test_utils.TestCasePlus'의 서브클래스에 있는지 확인해야 합니다. 
+위의 내용을 사용하려면 테스트가 'transformers.test_utils.TestCasePlus'의 서브클래스에 있는지 확인해야 합니다.
 예를 들어 다음과 같습니다:
 
 ```python
@@ -729,7 +729,7 @@ class PathExampleTest(TestCasePlus):
         data_dir = self.tests_dir / "fixtures/tests_samples/wmt_en_ro"
 ```
 
-만약 `pathlib`를 통해 경로를 조작할 필요가 없거나 경로를 문자열로만 필요로 하는 경우에는 `pathlib` 객체에 `str()`을 호출하거나 `_str`로 끝나는 접근자를 사용할 수 있습니다. 
+만약 `pathlib`를 통해 경로를 조작할 필요가 없거나 경로를 문자열로만 필요로 하는 경우에는 `pathlib` 객체에 `str()`을 호출하거나 `_str`로 끝나는 접근자를 사용할 수 있습니다.
 예를 들어 다음과 같습니다:
 
 ```python
@@ -743,14 +743,14 @@ class PathExampleTest(TestCasePlus):
 
 ### 임시 파일 및 디렉터리[[temporary-files-and-directories]]
 
-고유한 임시 파일 및 디렉터리를 사용하는 것은 병렬 테스트 실행에 있어 필수적입니다. 
-이렇게 함으로써 테스트들이 서로의 데이터를 덮어쓰지 않게 할 수 있습니다. 또한 우리는 생성된 테스트의 종료 단계에서 이러한 임시 파일 및 디렉터리를 제거하고 싶습니다.  
+고유한 임시 파일 및 디렉터리를 사용하는 것은 병렬 테스트 실행에 있어 필수적입니다.
+이렇게 함으로써 테스트들이 서로의 데이터를 덮어쓰지 않게 할 수 있습니다. 또한 우리는 생성된 테스트의 종료 단계에서 이러한 임시 파일 및 디렉터리를 제거하고 싶습니다.
 따라서 이러한 요구 사항을 충족시켜주는 `tempfile`과 같은 패키지를 사용하는 것이 중요합니다.
 
-그러나 테스트를 디버깅할 때는 임시 파일이나 디렉터리에 들어가는 내용을 확인할 수 있어야 하며, 
+그러나 테스트를 디버깅할 때는 임시 파일이나 디렉터리에 들어가는 내용을 확인할 수 있어야 하며,
 재실행되는 각 테스트마다 임시 파일이나 디렉터리의 경로에 대해 무작위 값이 아닌 정확한 값을 알고 싶을 것입니다.
 
-`transformers.test_utils.TestCasePlus`라는 도우미 클래스는 이러한 목적에 가장 적합합니다. 
+`transformers.test_utils.TestCasePlus`라는 도우미 클래스는 이러한 목적에 가장 적합합니다.
 이 클래스는 `unittest.TestCase`의 하위 클래스이므로, 우리는 이것을 테스트 모듈에서 쉽게 상속할 수 있습니다.
 
 다음은 해당 클래스를 사용하는 예시입니다:
@@ -773,7 +773,7 @@ def test_whatever(self):
     tmp_dir = self.get_auto_remove_tmp_dir()
 ```
 
-`tmp_dir`에는 생성된 임시 디렉터리의 경로가 포함됩니다. 
+`tmp_dir`에는 생성된 임시 디렉터리의 경로가 포함됩니다.
 이는 테스트의 종료 단계에서 자동으로 제거됩니다.
 
 - 선택한 경로로 임시 디렉터리 생성 후에 테스트 시작 전에 비어 있는 상태인지 확인하고, 테스트 후에는 비우지 마세요.
@@ -783,10 +783,10 @@ def test_whatever(self):
     tmp_dir = self.get_auto_remove_tmp_dir("./xxx")
 ```
 
-이것은 디버깅할 때 특정 디렉터리를 모니터링하고, 
+이것은 디버깅할 때 특정 디렉터리를 모니터링하고,
 그 디렉터리에 이전에 실행된 테스트가 데이터를 남기지 않도록 하는 데에 유용합니다.
 
-- `before` 및 `after` 인수를 직접 오버라이딩하여 기본 동작을 변경할 수 있으며 
+- `before` 및 `after` 인수를 직접 오버라이딩하여 기본 동작을 변경할 수 있으며
 다음 중 하나의 동작으로 이어집니다:
 
   - `before=True`: 테스트 시작 시 임시 디렉터리가 항상 지워집니다.
@@ -804,7 +804,7 @@ def test_whatever(self):
 
 <Tip>
 
-각 테스트는 여러 개의 임시 디렉터리를 등록할 수 있으며, 
+각 테스트는 여러 개의 임시 디렉터리를 등록할 수 있으며,
 별도로 요청하지 않는 한 모두 자동으로 제거됩니다.
 
 </Tip>
@@ -826,17 +826,17 @@ with ExtendSysPath(f"{bindir}/.."):
 
 ### 테스트 건너뛰기[[skipping-tests]]
 
-이것은 버그가 발견되어 새로운 테스트가 작성되었지만 아직 그 버그가 수정되지 않은 경우에 유용합니다. 
+이것은 버그가 발견되어 새로운 테스트가 작성되었지만 아직 그 버그가 수정되지 않은 경우에 유용합니다.
 이 테스트를 주 저장소에 커밋하려면 `make test` 중에 건너뛰도록 해야 합니다.
 
 방법:
 
-- **skip**은 테스트가 일부 조건이 충족될 경우에만 통과될 것으로 예상되고, 그렇지 않으면 pytest가 전체 테스트를 건너뛰어야 함을 의미합니다. 
-일반적인 예로는 Windows가 아닌 플랫폼에서 Windows 전용 테스트를 건너뛰거나 
+- **skip**은 테스트가 일부 조건이 충족될 경우에만 통과될 것으로 예상되고, 그렇지 않으면 pytest가 전체 테스트를 건너뛰어야 함을 의미합니다.
+일반적인 예로는 Windows가 아닌 플랫폼에서 Windows 전용 테스트를 건너뛰거나
 외부 리소스(예를 들어 데이터베이스)에 의존하는 테스트를 건너뛰는 것이 있습니다.
 
-- **xfail**은 테스트가 특정한 이유로 인해 실패할 것으로 예상하는 것을 의미합니다. 
-일반적인 예로는 아직 구현되지 않은 기능이나 아직 수정되지 않은 버그의 테스트가 있습니다. 
+- **xfail**은 테스트가 특정한 이유로 인해 실패할 것으로 예상하는 것을 의미합니다.
+일반적인 예로는 아직 구현되지 않은 기능이나 아직 수정되지 않은 버그의 테스트가 있습니다.
 `xfail`로 표시된 테스트가 예상대로 실패하지 않고 통과된 경우, 이것은 xpass이며 테스트 결과 요약에 기록됩니다.
 
 두 가지 중요한 차이점 중 하나는 `skip`은 테스트를 실행하지 않지만 `xfail`은 실행한다는 것입니다.
@@ -847,7 +847,7 @@ with ExtendSysPath(f"{bindir}/.."):
 - 전체 테스트를 무조건 건너뛰려면 다음과 같이 할 수 있습니다:
 
 ```python no-style
-@unittest.skip("this bug needs to be fixed")
+@unittest.skip(reason="this bug needs to be fixed")
 def test_feature_x():
 ```
 
@@ -920,7 +920,7 @@ class TestClass():
 
 ### 느린 테스트[[slow-tests]]
 
-테스트 라이브러리는 지속적으로 확장되고 있으며, 일부 테스트는 실행하는 데 몇 분이 걸립니다. 
+테스트 라이브러리는 지속적으로 확장되고 있으며, 일부 테스트는 실행하는 데 몇 분이 걸립니다.
 그리고 우리에게는 테스트 스위트가 CI를 통해 완료되기까지 한 시간을 기다릴 여유가 없습니다.
 따라서 필수 테스트를 위한 일부 예외를 제외하고 느린 테스트는 다음과 같이 표시해야 합니다.
 
@@ -936,7 +936,7 @@ def test_integration_foo():
 RUN_SLOW=1 pytest tests
 ```
 
-`@parameterized`와 같은 몇 가지 데코레이터는 테스트 이름을 다시 작성합니다. 
+`@parameterized`와 같은 몇 가지 데코레이터는 테스트 이름을 다시 작성합니다.
 그러므로 `@slow`와 나머지 건너뛰기 데코레이터 `@require_*`가 올바르게 작동되려면 마지막에 나열되어야 합니다. 다음은 올바른 사용 예입니다.
 
 ```python no-style
@@ -945,25 +945,25 @@ RUN_SLOW=1 pytest tests
 def test_integration_foo():
 ```
 
-이 문서의 초반부에 설명된 것처럼 느린 테스트는 PR의 CI 확인이 아닌 예약된 일정 기반으로 실행됩니다. 
+이 문서의 초반부에 설명된 것처럼 느린 테스트는 PR의 CI 확인이 아닌 예약된 일정 기반으로 실행됩니다.
 따라서 PR 제출 중에 일부 문제를 놓친 채로 병합될 수 있습니다.
-이러한 문제들은 다음번의 예정된 CI 작업 중에 감지됩니다. 
+이러한 문제들은 다음번의 예정된 CI 작업 중에 감지됩니다.
 하지만 PR을 제출하기 전에 자신의 컴퓨터에서 느린 테스트를 실행하는 것 또한 중요합니다.
 
 느린 테스트로 표시해야 하는지 여부를 결정하는 대략적인 결정 기준은 다음과 같습니다.
 
-만약 테스트가 라이브러리의 내부 구성 요소 중 하나에 집중되어 있다면(예: 모델링 파일, 토큰화 파일, 파이프라인), 
+만약 테스트가 라이브러리의 내부 구성 요소 중 하나에 집중되어 있다면(예: 모델링 파일, 토큰화 파일, 파이프라인),
 해당 테스트를 느린 테스트 스위트에서 실행해야 합니다.
-만약 라이브러리의 다른 측면(예: 문서 또는 예제)에 집중되어 있다면, 
+만약 라이브러리의 다른 측면(예: 문서 또는 예제)에 집중되어 있다면,
 해당 테스트를 느린 테스트 스위트에서 실행해야 합니다. 그리고 이 접근 방식을 보완하기 위해 예외를 만들어야 합니다.
 
-- 무거운 가중치 세트나 50MB보다 큰 데이터셋을 다운로드해야 하는 모든 테스트(예: 모델 통합 테스트, 토크나이저 통합 테스트, 파이프라인 통합 테스트)를 
+- 무거운 가중치 세트나 50MB보다 큰 데이터셋을 다운로드해야 하는 모든 테스트(예: 모델 통합 테스트, 토크나이저 통합 테스트, 파이프라인 통합 테스트)를
   느린 테스트로 설정해야 합니다.
-  새로운 모델을 추가하는 경우 통합 테스트용으로 무작위 가중치로 작은 버전을 만들어 허브에 업로드해야 합니다. 
+  새로운 모델을 추가하는 경우 통합 테스트용으로 무작위 가중치로 작은 버전을 만들어 허브에 업로드해야 합니다.
   이 내용은 아래 단락에서 설명됩니다.
 - 특별히 빠르게 실행되도록 최적화되지 않은 학습을 수행해야 하는 테스트는 느린 테스트로 설정해야 합니다.
-- 느리지 않아야 할 테스트 중 일부가 극도로 느린 경우 
-  예외를 도입하고 이를 `@slow`로 설정할 수 있습니다. 
+- 느리지 않아야 할 테스트 중 일부가 극도로 느린 경우
+  예외를 도입하고 이를 `@slow`로 설정할 수 있습니다.
   대용량 파일을 디스크에 저장하고 불러오는 자동 모델링 테스트는 `@slow`으로 표시된 테스트의 좋은 예입니다.
 - CI에서 1초 이내에 테스트가 완료되는 경우(다운로드 포함)에는 느린 테스트가 아니어야 합니다.
 
@@ -976,22 +976,22 @@ def test_integration_foo():
 grep tiny tests examples
 ```
 
-다음은 작은 모델[stas/tiny-wmt19-en-de](https://huggingface.co/stas/tiny-wmt19-en-de)을 만든 
-[script](https://github.com/huggingface/transformers/tree/main/scripts/fsmt/fsmt-make-tiny-model.py) 예시입니다. 
+다음은 작은 모델[stas/tiny-wmt19-en-de](https://huggingface.co/stas/tiny-wmt19-en-de)을 만든
+[script](https://github.com/huggingface/transformers/tree/main/scripts/fsmt/fsmt-make-tiny-model.py) 예시입니다.
 특정 모델의 아키텍처에 맞게 쉽게 조정할 수 있습니다.
 
-예를 들어 대용량 모델을 다운로드하는 경우 런타임을 잘못 측정하기 쉽지만, 
-로컬에서 테스트하면 다운로드한 파일이 캐시되어 다운로드 시간이 측정되지 않습니다. 
+예를 들어 대용량 모델을 다운로드하는 경우 런타임을 잘못 측정하기 쉽지만,
+로컬에서 테스트하면 다운로드한 파일이 캐시되어 다운로드 시간이 측정되지 않습니다.
 대신 CI 로그의 실행 속도 보고서를 확인하세요(`pytest --durations=0 tests`의 출력).
 
-이 보고서는 느린 이상값으로 표시되지 않거나 빠르게 다시 작성해야 하는 느린 이상값을 찾는 데도 유용합니다. 
+이 보고서는 느린 이상값으로 표시되지 않거나 빠르게 다시 작성해야 하는 느린 이상값을 찾는 데도 유용합니다.
 CI에서 테스트 스위트가 느려지기 시작하면 이 보고서의 맨 위 목록에 가장 느린 테스트가 표시됩니다.
 
 
 
 ### stdout/stderr 출력 테스트[[testing-the-stdout/stderr-output]]
 
-`stdout` 및/또는 `stderr`로 쓰는 함수를 테스트하려면 `pytest`의 [capsys 시스템](https://docs.pytest.org/en/latest/capture.html)을 사용하여 해당 스트림에 액세스할 수 있습니다. 
+`stdout` 및/또는 `stderr`로 쓰는 함수를 테스트하려면 `pytest`의 [capsys 시스템](https://docs.pytest.org/en/latest/capture.html)을 사용하여 해당 스트림에 액세스할 수 있습니다.
 다음과 같이 수행할 수 있습니다.
 
 ```python
@@ -1019,7 +1019,7 @@ def test_result_and_stdout(capsys):
     assert msg in err
 ```
 
-그리고, 물론 대부분의 경우에는 `stderr`는 예외의 일부로 제공됩니다. 
+그리고, 물론 대부분의 경우에는 `stderr`는 예외의 일부로 제공됩니다.
 그러므로 해당 경우에는 try/except를 사용해야 합니다.
 
 ```python
@@ -1061,11 +1061,11 @@ def test_result_and_stdout():
 ```
 
 `stdout` 캡처에 관련된 중요한 문제 중 하나는 보통 `print`에서 이전에 인쇄된 내용을 재설정하는 `\r` 문자가 포함될 수 있다는 것입니다.
-`pytest`에서는 문제가 없지만 `pytest -s`에서는 이러한 문자가 버퍼에 포함되므로 
+`pytest`에서는 문제가 없지만 `pytest -s`에서는 이러한 문자가 버퍼에 포함되므로
 `-s`가 있거나 없는 상태에서 태스트를 수행할 수 있으려면 캡처된 출력에 대해 추가적인 정리가 필요합니다.
 이 경우에는 `re.sub(r'~.*\r', '', buf, 0, re.M)`을 사용할 수 있습니다.
 
-하지만 도우미 컨텍스트 관리자 래퍼를 사용하면 
+하지만 도우미 컨텍스트 관리자 래퍼를 사용하면
 출력에 `\r`이 포함되어 있는지의 여부에 관계없이 모든 것을 자동으로 처리하므로 편리합니다.
 
 ```python
@@ -1108,7 +1108,7 @@ with CaptureStd() as cs:
 print(cs.err, cs.out)
 ```
 
-또한, 테스트의 디버깅을 지원하기 위해 
+또한, 테스트의 디버깅을 지원하기 위해
 이러한 컨텍스트 관리자는 기본적으로 컨텍스트에서 종료할 때 캡처된 스트림을 자동으로 다시 실행합니다.
 
 
@@ -1130,7 +1130,7 @@ assert cl.out, msg + "\n"
 
 ### 환경 변수를 이용하여 테스트[[testing-with-environment-variables]]
 
-특정 테스트의 환경 변수 영향을 검증하려면 
+특정 테스트의 환경 변수 영향을 검증하려면
 `transformers.testing_utils.mockenv`라는 도우미 데코레이터를 사용할 수 있습니다.
 
 ```python
@@ -1143,7 +1143,7 @@ class HfArgumentParserTest(unittest.TestCase):
         env_level_str = os.getenv("TRANSFORMERS_VERBOSITY", None)
 ```
 
-일부 경우에는 외부 프로그램을 호출해야할 수도 있는데, 이 때에는 여러 개의 로컬 경로를 포함하는 `os.environ`에서 `PYTHONPATH`의 설정이 필요합니다.  
+일부 경우에는 외부 프로그램을 호출해야할 수도 있는데, 이 때에는 여러 개의 로컬 경로를 포함하는 `os.environ`에서 `PYTHONPATH`의 설정이 필요합니다.
 헬퍼 클래스 `transformers.test_utils.TestCasePlus`가 도움이 됩니다:
 
 ```python
@@ -1156,8 +1156,8 @@ class EnvExampleTest(TestCasePlus):
         # 이제 `env`를 사용하여 외부 프로그램 호출
 ```
 
-테스트 파일이 `tests` 테스트 스위트 또는 `examples`에 있는지에 따라 
-`env[PYTHONPATH]`가 두 디렉터리 중 하나를 포함하도록 설정되며, 
+테스트 파일이 `tests` 테스트 스위트 또는 `examples`에 있는지에 따라
+`env[PYTHONPATH]`가 두 디렉터리 중 하나를 포함하도록 설정되며,
 현재 저장소에 대해 테스트가 수행되도록 `src` 디렉터리도 포함됩니다.
 테스트 호출 이전에 설정된 경우에는 `env[PYTHONPATH]`를 그대로 사용합니다.
 
@@ -1166,7 +1166,7 @@ class EnvExampleTest(TestCasePlus):
 
 ### 재현 가능한 결과 얻기[[getting-reproducible-results]]
 
-일부 상황에서 테스트에서 임의성을 제거하여 동일하게 재현 가능한 결과를 얻고 싶을 수 있습니다. 
+일부 상황에서 테스트에서 임의성을 제거하여 동일하게 재현 가능한 결과를 얻고 싶을 수 있습니다.
 이를 위해서는 다음과 같이 시드를 고정해야 합니다.
 
 ```python
@@ -1207,11 +1207,11 @@ pytest tests/utils/test_logging.py -W error::UserWarning --pdb
 셀프 푸시 워크플로우 CI 작업을 트리거하려면, 다음을 수행해야 합니다.
 
 1. `transformers` 원본에서 새 브랜치를 만듭니다(포크가 아닙니다!).
-2. 브랜치 이름은 `ci_` 또는 `ci-`로 시작해야 합니다(`main`도 트리거하지만 `main`에서는 PR을 할 수 없습니다). 
-   또한 특정 경로에 대해서만 트리거되므로 이 문서가 작성된 후에 변경된 내용은 
+2. 브랜치 이름은 `ci_` 또는 `ci-`로 시작해야 합니다(`main`도 트리거하지만 `main`에서는 PR을 할 수 없습니다).
+   또한 특정 경로에 대해서만 트리거되므로 이 문서가 작성된 후에 변경된 내용은
    [여기](https://github.com/huggingface/transformers/blob/main/.github/workflows/self-push.yml)의 *push:*에서 확인할 수 있습니다.
 3. 이 브랜치에서 PR을 생성합니다
-4. 그런 다음 [여기](https://github.com/huggingface/transformers/actions/workflows/self-push.yml)에서 작업이 나타나는지 확인할 수 있습니다. 
+4. 그런 다음 [여기](https://github.com/huggingface/transformers/actions/workflows/self-push.yml)에서 작업이 나타나는지 확인할 수 있습니다.
    백로그가 있는 경우, 바로 실행되지 않을 수도 있습니다.
 
 
@@ -1219,13 +1219,13 @@ pytest tests/utils/test_logging.py -W error::UserWarning --pdb
 
 ## 실험적인 CI 기능 테스트[[testing-Experimental-CI-Features]]
 
-CI 기능을 테스트하는 것은 일반 CI 작동에 방해가 될 수 있기 때문에 잠재적으로 문제가 발생할 수 있습니다. 
+CI 기능을 테스트하는 것은 일반 CI 작동에 방해가 될 수 있기 때문에 잠재적으로 문제가 발생할 수 있습니다.
 따라서 새로운 CI 기능을 추가하는 경우 다음과 같이 수행해야 합니다.
 
 1. 테스트해야 할 내용을 테스트하는 새로운 전용 작업을 생성합니다.
 2. 새로운 작업은 항상 성공해야만 녹색 ✓를 받을 수 있습니다(아래에 자세한 내용이 있습니다).
-3. 다양한 PR 유형에 대한 확인을  위해 
-   (사용자 포크 브랜치, 포크되지 않은 브랜치, github.com UI 직접 파일 편집에서 생성된 브랜치, 강제 푸시 등 PR의 유형은 아주 다양합니다.) 
+3. 다양한 PR 유형에 대한 확인을  위해
+   (사용자 포크 브랜치, 포크되지 않은 브랜치, github.com UI 직접 파일 편집에서 생성된 브랜치, 강제 푸시 등 PR의 유형은 아주 다양합니다.)
    며칠 동안 실험 작업의 로그를 모니터링하면서 실행해봅니다.
    (의도적으로 항상 녹색을 표시하므로 작업 전체가 녹색은 아니라는 점에 유의합니다.)
 4. 모든 것이 안정적인지 확인한 후, 새로운 변경 사항을 기존 작업에 병합합니다.
@@ -1234,7 +1234,7 @@ CI 기능을 테스트하는 것은 일반 CI 작동에 방해가 될 수 있기
 
 그러나 새로운 CI 기능이 개발 중인 동안, 항상 성공하도록 할 수 있는 방법은 무엇일까요?
 
-TravisCI와 같은 일부 CI는 `ignore-step-failure`를 지원하며 전체 작업을 성공한 것으로 보고하지만, 
+TravisCI와 같은 일부 CI는 `ignore-step-failure`를 지원하며 전체 작업을 성공한 것으로 보고하지만,
 현재 우리가 사용하는 CircleCI와 Github Actions는 이를 지원하지 않습니다.
 
 따라서 다음과 같은 해결책을 사용할 수 있습니다.
@@ -1264,12 +1264,12 @@ TravisCI와 같은 일부 CI는 `ignore-step-failure`를 지원하며 전체 작
 cmd_that_may_fail || true
 ```
 
-결과에 만족한 후에는 물론, 실험적인 단계 또는 작업을 일반 작업의 나머지 부분과 통합하면서 
-`set +euo pipefail` 또는 기타 추가한 요소를 제거하여 
+결과에 만족한 후에는 물론, 실험적인 단계 또는 작업을 일반 작업의 나머지 부분과 통합하면서
+`set +euo pipefail` 또는 기타 추가한 요소를 제거하여
 실험 작업이 일반 CI 작동에 방해되지 않도록 해야 합니다.
 
-이 전반적인 과정은 실험 단계가 PR의 전반적인 상태에 영향을 주지 않고 실패하도록 
-`allow-failure`와 같은 기능을 설정할 수 있다면 훨씬 더 쉬웠을 것입니다. 
+이 전반적인 과정은 실험 단계가 PR의 전반적인 상태에 영향을 주지 않고 실패하도록
+`allow-failure`와 같은 기능을 설정할 수 있다면 훨씬 더 쉬웠을 것입니다.
 그러나 앞에서 언급한 바와 같이 CircleCI와 Github Actions는 현재 이러한 기능들 지원하지 않습니다.
 
 이 기능의 지원을 위한 투표에 참여하고 CI 관련 스레드들에서 이러한 상황을 확인할 수도 있습니다.

docs/source/zh/main_classes/callback.md

@@ -28,7 +28,7 @@ Callbacks是“只读”的代码片段，除了它们返回的[TrainerControl]
 - [`PrinterCallback`] 或 [`ProgressCallback`]，用于显示进度和打印日志（如果通过[`TrainingArguments`]停用tqdm，则使用第一个函数；否则使用第二个）。
 - [`~integrations.TensorBoardCallback`]，如果TensorBoard可访问（通过PyTorch版本 >= 1.4 或者 tensorboardX）。
 - [`~integrations.WandbCallback`]，如果安装了[wandb](https://www.wandb.com/)。
-- [`~integrations.CometCallback`]，如果安装了[comet_ml](https://www.comet.ml/site/)。
+- [`~integrations.CometCallback`]，如果安装了[comet_ml](https://www.comet.com/site/)。
 - [`~integrations.MLflowCallback`]，如果安装了[mlflow](https://www.mlflow.org/)。
 - [`~integrations.NeptuneCallback`]，如果安装了[neptune](https://neptune.ai/)。
 - [`~integrations.AzureMLCallback`]，如果安装了[azureml-sdk](https://pypi.org/project/azureml-sdk/)。

examples/flax/language-modeling/run_bart_dlm_flax.py

@@ -290,7 +290,7 @@ class FlaxDataCollatorForBartDenoisingLM:
     def __post_init__(self):
         if self.tokenizer.mask_token is None or self.tokenizer.eos_token is None:
             raise ValueError(
-                "This tokenizer does not have a mask token or eos token token which is necessary for denoising"
+                "This tokenizer does not have a mask token or eos token which is necessary for denoising"
                 " language modeling. "
             )
 

examples/flax/language-modeling/run_clm_flax.py

@@ -225,9 +225,6 @@ class DataTrainingArguments:
             )
         },
     )
-    overwrite_cache: bool = field(
-        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
-    )
     validation_split_percentage: Optional[int] = field(
         default=5,
         metadata={

examples/flax/question-answering/run_qa.py

@@ -61,7 +61,7 @@ from transformers.utils import check_min_version, send_example_telemetry
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 Array = Any
 Dataset = datasets.arrow_dataset.Dataset

examples/flax/speech-recognition/run_flax_speech_recognition_seq2seq.py

@@ -60,7 +60,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risk.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/flax/speech-recognition/requirements.txt")
 

examples/flax/text-classification/run_flax_glue.py

@@ -56,7 +56,7 @@ from transformers.utils import check_min_version, send_example_telemetry
 
 logger = logging.getLogger(__name__)
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 Array = Any
 Dataset = datasets.arrow_dataset.Dataset

examples/flax/token-classification/run_flax_ner.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 logger = logging.getLogger(__name__)
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/token-classification/requirements.txt")
 

examples/pytorch/README.md

@@ -200,7 +200,7 @@ You can easily log and monitor your runs code. The following are currently suppo
 
 * [TensorBoard](https://www.tensorflow.org/tensorboard)
 * [Weights & Biases](https://docs.wandb.ai/integrations/huggingface)
-* [Comet ML](https://www.comet.ml/docs/python-sdk/huggingface/)
+* [Comet ML](https://www.comet.com/docs/v2/integrations/ml-frameworks/transformers/)
 * [Neptune](https://docs.neptune.ai/integrations-and-supported-tools/model-training/hugging-face)
 * [ClearML](https://clear.ml/docs/latest/docs/getting_started/ds/ds_first_steps)
 * [DVCLive](https://dvc.org/doc/dvclive/ml-frameworks/huggingface)
@@ -244,7 +244,7 @@ Additional configuration options are available through generic [wandb environmen
 
 Refer to related [documentation & examples](https://docs.wandb.ai/integrations/huggingface).
 
-### Comet.ml
+### Comet
 
 To use `comet_ml`, install the Python package with:
 

examples/pytorch/audio-classification/run_audio_classification.py

@@ -45,7 +45,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.14.0", "To fix: pip install -r examples/pytorch/audio-classification/requirements.txt")
 

examples/pytorch/contrastive-image-text/run_clip.py

@@ -54,7 +54,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/contrastive-image-text/requirements.txt")
 
@@ -190,9 +190,9 @@ class DataTrainingArguments:
             if self.validation_file is not None:
                 extension = self.validation_file.split(".")[-1]
                 assert extension in ["csv", "json"], "`validation_file` should be a csv or a json file."
-            if self.validation_file is not None:
-                extension = self.validation_file.split(".")[-1]
-                assert extension == "json", "`validation_file` should be a json file."
+            if self.test_file is not None:
+                extension = self.test_file.split(".")[-1]
+                assert extension in ["csv", "json"], "`test_file` should be a csv or a json file."
 
 
 dataset_name_mapping = {

examples/pytorch/image-classification/run_image_classification.py

@@ -56,7 +56,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/pytorch/image-classification/requirements.txt")
 

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -49,7 +49,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 

examples/pytorch/image-pretraining/run_mae.py

@@ -43,7 +43,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-pretraining/requirements.txt")
 

examples/pytorch/image-pretraining/run_mim.py

@@ -48,7 +48,7 @@ Any model supported by the AutoModelForMaskedImageModeling API can be used.
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-pretraining/requirements.txt")
 

examples/pytorch/image-pretraining/run_mim_no_trainer.py

@@ -53,7 +53,7 @@ Any model supported by the AutoModelForMaskedImageModeling API can be used.
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-pretraining/requirements.txt")
 

examples/pytorch/instance-segmentation/run_instance_segmentation.py

@@ -46,7 +46,8 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
+
 require_version("datasets>=2.0.0", "To fix: pip install -r examples/pytorch/instance-segmentation/requirements.txt")
 
 

examples/pytorch/instance-segmentation/run_instance_segmentation_no_trainer.py

@@ -52,7 +52,8 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
+
 require_version("datasets>=2.0.0", "To fix: pip install -r examples/pytorch/instance-segmentation/requirements.txt")
 
 

examples/pytorch/language-modeling/run_clm.py

@@ -55,7 +55,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 

examples/pytorch/language-modeling/run_fim.py

@@ -58,7 +58,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/language-modeling/run_fim_no_trainer.py

@@ -60,7 +60,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 

examples/pytorch/language-modeling/run_mlm.py

@@ -54,7 +54,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")

examples/pytorch/language-modeling/run_plm.py

@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.14.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/multiple-choice/run_swag.py

@@ -47,7 +47,7 @@ from transformers.utils import PaddingStrategy, check_min_version, send_example_
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -56,7 +56,7 @@ from transformers.utils import PaddingStrategy, check_min_version, send_example_
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 # You should update this to your particular problem to have better documentation of `model_type`

examples/pytorch/object-detection/run_object_detection.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.0.0", "To fix: pip install -r examples/pytorch/object-detection/requirements.txt")
 

examples/pytorch/object-detection/run_object_detection_no_trainer.py

@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logging.basicConfig(level=logging.INFO)
 logger = get_logger(__name__)

examples/pytorch/question-answering/run_qa.py

@@ -50,7 +50,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/question-answering/run_qa_beam_search.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -56,7 +56,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/question-answering/run_seq2seq_qa.py

@@ -46,7 +46,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/semantic-segmentation/run_semantic_segmentation.py

@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=2.0.0", "To fix: pip install -r examples/pytorch/semantic-segmentation/requirements.txt")
 

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -50,7 +50,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 

examples/pytorch/speech-recognition/run_speech_recognition_ctc.py

@@ -50,7 +50,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
 

examples/pytorch/speech-recognition/run_speech_recognition_ctc_adapter.py

@@ -53,7 +53,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
 

examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
 

examples/pytorch/summarization/run_summarization.py

@@ -52,7 +52,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")
 

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -56,7 +56,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 logger = get_logger(__name__)
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")

examples/pytorch/text-classification/run_classification.py

@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
 
@@ -133,6 +133,10 @@ class DataTrainingArguments:
             )
         },
     )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
     overwrite_cache: bool = field(
         default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
     )
@@ -573,6 +577,7 @@ def main():
         raw_datasets = raw_datasets.map(
             preprocess_function,
             batched=True,
+            num_proc=data_args.preprocessing_num_workers,
             load_from_cache_file=not data_args.overwrite_cache,
             desc="Running tokenizer on dataset",
         )

examples/pytorch/text-classification/run_glue.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.42.0.dev0")
+check_min_version("4.43.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")