trigger CI

* fix

* trigger CI

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

.github/ISSUE_TEMPLATE/i18n.md

@@ -23,7 +23,7 @@ Some notes:
 * Please translate in a gender-neutral way.
 * Add your translations to the folder called `<languageCode>` inside the [source folder](https://github.com/huggingface/transformers/tree/main/docs/source).
 * Register your translation in `<languageCode>/_toctree.yml`; please follow the order of the [English version](https://github.com/huggingface/transformers/blob/main/docs/source/en/_toctree.yml).
-* Once you're finished, open a pull request and tag this issue by including #issue-number in the description, where issue-number is the number of this issue. Please ping @stevhliu and @MKhalusova for review.
+* Once you're finished, open a pull request and tag this issue by including #issue-number in the description, where issue-number is the number of this issue. Please ping @stevhliu for review.
 * 🙋 If you'd like others to help you with the translation, you can also post in the 🤗 [forums](https://discuss.huggingface.co/).
 
 ## Get Started section

.github/scripts/assign_reviewers.py

@@ -54,6 +54,21 @@ def get_file_owners(file_path, codeowners_lines):
             return owners  # Remember, can still be empty!
     return []  # Should never happen, but just in case
 
+def pr_author_is_in_hf(pr_author, codeowners_lines):
+    # Check if the PR author is in the codeowners file
+    for line in codeowners_lines:
+        line = line.split('#')[0].strip()
+        if not line:
+            continue
+
+        # Split into pattern and owners
+        parts = line.split()
+        owners = [owner.removeprefix("@") for owner in parts[1:]]
+
+        if pr_author in owners:
+            return True
+    return False
+
 def main():
     script_dir = Path(__file__).parent.absolute()
     with open(script_dir / "codeowners_for_review_action") as f:
@@ -68,6 +83,9 @@ def main():
     pr_number = event['pull_request']['number']
     pr = repo.get_pull(pr_number)
     pr_author = pr.user.login
+    if pr_author_is_in_hf(pr_author, codeowners_lines):
+        print(f"PR author {pr_author} is in codeowners, skipping review request.")
+        return
 
     existing_reviews = list(pr.get_reviews())
     if existing_reviews:

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

@@ -63,14 +63,14 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-all-latest-gpu-push-ci docker build 
+          title: 🤗 Results of the transformers-all-latest-gpu-push-ci docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
   latest-torch-deepspeed-docker:
     name: "Latest PyTorch + DeepSpeed"
     runs-on:
-      group: aws-general-8-plus
+      group: aws-g4dn-2xlarge-cache
     steps:
       -
         name: Set up Docker Buildx
@@ -99,7 +99,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER}}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu docker build 
+          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -140,7 +140,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu-push-ci docker build 
+          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu-push-ci docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -176,7 +176,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-doc-builder docker build 
+          title: 🤗 Results of the huggingface/transformers-doc-builder docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -214,7 +214,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-pytorch-gpudocker build 
+          title: 🤗 Results of the huggingface/transformers-pytorch-gpudocker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -223,19 +223,19 @@ jobs:
     runs-on:
       group: aws-general-8-plus
     steps:
-      - 
+      -
         name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
-      - 
+      -
         name: Check out code
         uses: actions/checkout@v4
-      - 
+      -
         name: Login to DockerHub
         uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      - 
+      -
         name: Build and push
         uses: docker/build-push-action@v5
         with:
@@ -263,7 +263,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-pytorch-amd-gpu-push-ci build 
+          title: 🤗 Results of the huggingface/transformers-pytorch-amd-gpu-push-ci build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -301,7 +301,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-tensorflow-gpu build 
+          title: 🤗 Results of the huggingface/transformers-tensorflow-gpu build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -310,19 +310,19 @@ jobs:
     runs-on:
       group: aws-general-8-plus
     steps:
-      - 
+      -
         name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
-      - 
+      -
         name: Check out code
         uses: actions/checkout@v4
-      - 
+      -
         name: Login to DockerHub
         uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      - 
+      -
         name: Build and push
         uses: docker/build-push-action@v5
         with:
@@ -350,7 +350,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-amd-gpu build 
+          title: 🤗 Results of the transformers-pytorch-deepspeed-amd-gpu build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -388,6 +388,6 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-quantization-latest-gpu build 
+          title: 🤗 Results of the transformers-quantization-latest-gpu build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

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

@@ -42,7 +42,7 @@ jobs:
   nightly-torch-deepspeed-docker:
     name: "Nightly PyTorch + DeepSpeed"
     runs-on:
-      group: aws-general-8-plus
+      group: aws-g4dn-2xlarge-cache
     steps:
       -
         name: Set up Docker Buildx

.github/workflows/model_jobs.yml

@@ -18,6 +18,10 @@ on:
       docker:
         required: true
         type: string
+      report_name_prefix:
+        required: false
+        default: run_models_gpu
+        type: string
 
 env:
   HF_HOME: /mnt/cache
@@ -116,23 +120,23 @@ jobs:
 
       - name: Run all tests on GPU
         working-directory: /transformers
-        run: python3 -m pytest -rsfE -v --make-reports=${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
+        run: python3 -m pytest -rsfE -v --make-reports=${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
+        run: cat /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports/failures_short.txt
 
       - name: Run test
         shell: bash
         run: |
-          mkdir -p /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-          echo "hello" > /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/hello.txt
-          echo "${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports"
+          mkdir -p /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports
+          echo "hello" > /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports/hello.txt
+          echo "${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports"
 
-      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
+      - name: "Test suite reports artifacts: ${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ env.matrix_folders }}_test_reports"
         if: ${{ always() }}
         uses: actions/upload-artifact@v4
         with:
-          name: ${{ env.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
+          name: ${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports

.github/workflows/self-comment-ci.yml

@@ -29,7 +29,7 @@ jobs:
     runs-on: ubuntu-22.04
     name: Get PR number
     # For security: only allow team members to run
-    if: ${{ github.event.issue.state == 'open' && contains(fromJSON('["ydshieh", "ArthurZucker", "zucchini-nlp", "qubvel", "molbap", "gante", "LysandreJik", "Cyrilvallez", "Rocketknight1", "SunMarc", "muellerzr", "eustlb"]'), github.actor) && (startsWith(github.event.comment.body, 'run-slow') || startsWith(github.event.comment.body, 'run slow') || startsWith(github.event.comment.body, 'run_slow')) }}
+    if: ${{ github.event.issue.state == 'open' && contains(fromJSON('["ydshieh", "ArthurZucker", "zucchini-nlp", "qubvel", "molbap", "gante", "LysandreJik", "Cyrilvallez", "Rocketknight1", "SunMarc", "muellerzr", "eustlb", "MekkCyber"]'), github.actor) && (startsWith(github.event.comment.body, 'run-slow') || startsWith(github.event.comment.body, 'run slow') || startsWith(github.event.comment.body, 'run_slow')) }}
     outputs:
       PR_NUMBER: ${{ steps.set_pr_number.outputs.PR_NUMBER }}
     steps:

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

@@ -54,12 +54,23 @@ jobs:
       ci_event: Daily CI
     secrets: inherit
 
+  trainer-fsdp-ci:
+    name: Trainer/FSDP CI
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_trainer_and_fsdp_gpu
+      slack_report_channel: "#transformers-ci-daily-training"
+      runner: daily-ci
+      docker: huggingface/transformers-all-latest-gpu
+      ci_event: Daily CI
+    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-daily-deepspeed"
+      slack_report_channel: "#transformers-ci-daily-training"
       runner: daily-ci
       docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
       ci_event: Daily CI

.github/workflows/self-scheduled.yml

@@ -45,7 +45,7 @@ env:
 
 jobs:
   setup:
-    if: contains(fromJSON('["run_models_gpu", "run_quantization_torch_gpu"]'), inputs.job)
+    if: contains(fromJSON('["run_models_gpu", "run_trainer_and_fsdp_gpu", "run_quantization_torch_gpu"]'), inputs.job)
     name: Setup
     strategy:
       matrix:
@@ -77,12 +77,17 @@ jobs:
         run: pip freeze
 
       - id: set-matrix
-        if: ${{ inputs.job == 'run_models_gpu' }}
+        if: contains(fromJSON('["run_models_gpu", "run_trainer_and_fsdp_gpu"]'), inputs.job)
         name: Identify models to test
         working-directory: /transformers/tests
         run: |
-          echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
-          echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
+          if [ "${{ inputs.job }}" = "run_models_gpu" ]; then
+            echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
+            echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
+          elif [ "${{ inputs.job }}" = "run_trainer_and_fsdp_gpu" ]; then
+            echo "folder_slices=[['trainer'], ['fsdp']]" >> $GITHUB_OUTPUT
+            echo "slice_ids=[0, 1]" >> $GITHUB_OUTPUT
+          fi
 
       - id: set-matrix-quantization
         if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
@@ -113,6 +118,25 @@ jobs:
       docker: ${{ inputs.docker }}
     secrets: inherit
 
+  run_trainer_and_fsdp_gpu:
+    if: ${{ inputs.job == 'run_trainer_and_fsdp_gpu' }}
+    name: " "
+    needs: setup
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [aws-g4dn-2xlarge-cache, aws-g4dn-12xlarge-cache]
+        slice_id: [0, 1]
+    uses: ./.github/workflows/model_jobs.yml
+    with:
+      folder_slices: ${{ needs.setup.outputs.folder_slices }}
+      machine_type: ${{ matrix.machine_type }}
+      slice_id: ${{ matrix.slice_id }}
+      runner: ${{ inputs.runner }}
+      docker: ${{ inputs.docker }}
+      report_name_prefix: run_trainer_and_fsdp_gpu
+    secrets: inherit
+
   run_pipelines_torch_gpu:
     if: ${{ inputs.job == 'run_pipelines_torch_gpu' }}
     name: PyTorch pipelines
@@ -382,7 +406,7 @@ jobs:
         run: pip freeze
 
       - name: Set `machine_type` for report and artifact names
-        working-directory: /transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         shell: bash
         run: |
           echo "${{ matrix.machine_type }}"
@@ -541,6 +565,7 @@ jobs:
     needs: [
       setup,
       run_models_gpu,
+      run_trainer_and_fsdp_gpu,
       run_pipelines_torch_gpu,
       run_pipelines_tf_gpu,
       run_examples_gpu,

ISSUES.md

@@ -26,7 +26,7 @@ There are two main venues to receive support: [the forums](https://discuss.huggi
 
 [The user forums](https://discuss.huggingface.co/) are supported by the wide community of the library users and backed up by developers when needed.
 
-If you have a difficulty with deploying this library or some questions, or you'd like to discuss a new feature, please first consider discussing those things at the forums. Only when you feel your subject matter has been crystalized and you still need support from the library developers do proceed to file an [issue](https://github.com/huggingface/transformers/issues).
+If you have a difficulty with deploying this library or some questions, or you'd like to discuss a new feature, please first consider discussing those things at the forums. Only when you feel your subject matter has been crystallized and you still need support from the library developers do proceed to file an [issue](https://github.com/huggingface/transformers/issues).
 
 In particular all "Please explain" questions or objectively very user-specific feature requests belong to the forums. Here are some example of such questions:
 

README.md

@@ -70,7 +70,7 @@ Explore the [Hub](https://huggingface.com/) today to find a model and use Transf
 
 ## Installation
 
-Transformers works with Python 3.9+ [PyTorch](https://pytorch.org/get-started/locally/) 2.0+, [TensorFlow](https://www.tensorflow.org/install/pip) 2.6+, and [Flax](https://flax.readthedocs.io/en/latest/) 0.4.1+.
+Transformers works with Python 3.9+ [PyTorch](https://pytorch.org/get-started/locally/) 2.1+, [TensorFlow](https://www.tensorflow.org/install/pip) 2.6+, and [Flax](https://flax.readthedocs.io/en/latest/) 0.4.1+.
 
 Create and activate a virtual environment with [venv](https://docs.python.org/3/library/venv.html) or [uv](https://docs.astral.sh/uv/), a fast Rust-based Python package and project manager.
 

SECURITY.md

@@ -27,13 +27,6 @@ These models require the `trust_remote_code=True` parameter to be set when using
 the content of the modeling files when using this argument. We recommend setting a revision in order to ensure you
 protect yourself from updates on the repository.
 
-#### Tools
-
-Through the `Agent` framework, remote tools can be downloaded to be used by the Agent. You're to specify these tools
-yourself, but please keep in mind that their code will be run on your machine if the Agent chooses to run them.
-
-Please inspect the code of the tools before passing them to the Agent to protect your runtime and local setup.
-
 ## Reporting a Vulnerability
 
 Feel free to submit vulnerability reports to [security@huggingface.co](mailto:security@huggingface.co), where someone from the HF security team will review and recommend next steps. If reporting a vulnerability specific to open source, please note [Huntr](https://huntr.com) is a vulnerability disclosure program for open source software.

conftest.py

@@ -66,7 +66,6 @@ NOT_DEVICE_TESTS = {
     "ModelTester::test_pipeline_",
     "/repo_utils/",
     "/utils/",
-    "/agents/",
 }
 
 # allow having multiple repository checkouts and not needing to remember to rerun
@@ -83,7 +82,6 @@ def pytest_configure(config):
     config.addinivalue_line("markers", "is_pipeline_test: mark test to run only when pipelines are tested")
     config.addinivalue_line("markers", "is_staging_test: mark test to run only in the staging environment")
     config.addinivalue_line("markers", "accelerate_tests: mark test that require accelerate")
-    config.addinivalue_line("markers", "agent_tests: mark the agent tests that are run on their specific schedule")
     config.addinivalue_line("markers", "not_device_test: mark the tests always running on cpu")
 
 

docker/transformers-all-latest-gpu/Dockerfile

@@ -14,6 +14,8 @@ ARG PYTORCH='2.6.0'
 ARG INTEL_TORCH_EXT='2.3.0'
 # Example: `cu102`, `cu113`, etc.
 ARG CUDA='cu121'
+# Disable kernel mapping for now until all tests pass
+ENV DISABLE_KERNEL_MAPPING=1
 
 RUN apt update
 RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs

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

@@ -1,12 +1,12 @@
-# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
-FROM nvcr.io/nvidia/pytorch:23.11-py3
+# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-08.html
+FROM nvcr.io/nvidia/pytorch:24.08-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
-ARG PYTORCH='2.2.0'
+ARG PYTORCH='2.6.0'
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu126'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -15,7 +15,8 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
 
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
+# `datasets` requires pandas, pandas has some modules compiled with numpy=1.x causing errors
+RUN python3 -m pip install --no-cache-dir './transformers[deepspeed-testing]' 'pandas<2' 'numpy<2'
 
 # Install latest release PyTorch
 # (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)

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

@@ -1,11 +1,11 @@
 # https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
-FROM nvcr.io/nvidia/pytorch:23.11-py3
+FROM nvcr.io/nvidia/pytorch:24.08-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu126'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -21,7 +21,8 @@ RUN python3 -m pip uninstall -y torch torchvision torchaudio
 # (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
 RUN python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/$CUDA
 
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
+# `datasets` requires pandas, pandas has some modules compiled with numpy=1.x causing errors
+RUN python3 -m pip install --no-cache-dir './transformers[deepspeed-testing]' 'pandas<2' 'numpy<2'
 
 RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
 

docker/transformers-quantization-latest-gpu/Dockerfile

@@ -12,6 +12,8 @@ SHELL ["sh", "-lc"]
 ARG PYTORCH='2.6.0'
 # Example: `cu102`, `cu113`, etc.
 ARG CUDA='cu121'
+# Disable kernel mapping for quantization tests
+ENV DISABLE_KERNEL_MAPPING=1
 
 RUN apt update
 RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg

docs/source/ar/_toctree.yml

@@ -23,8 +23,6 @@
     title: تحميل النماذج المخصصة وتدريبها باستخدام 🤗 PEFT
   - local: model_sharing
     title: مشاركة نموذجك
-  - local: agents
-    title: الوكلاء
   - local: llm_tutorial
     title: التوليد باستخدام LLMs
   - local: conversations
@@ -252,8 +250,6 @@
   title: أطر مفاهيمية
 # - sections:
 #   - sections:
-#     - local: main_classes/agent
-#       title: الوكلاء والأدوات
 #     - local: model_doc/auto
 #       title: فئات يتم إنشاؤها ديناميكيًا
 #     - local: main_classes/backbones

docs/source/ar/agents.md

@@ -1,539 +0,0 @@
-# الوكلاء والأدوات
-
-[[open-in-colab]]
-
-### ما هو الوكيل؟
-
-يمكن للنظم اللغوية الكبيرة (LLMs) التي تم تدريبها على أداء [نمذجة اللغة السببية](./tasks/language_modeling.) التعامل مع مجموعة واسعة من المهام، ولكنها غالبًا ما تواجه صعوبات في المهام الأساسية مثل المنطق والحساب والبحث. وعندما يتم استدعاؤها في مجالات لا تؤدي فيها أداءً جيدًا، فإنها غالبًا ما تفشل في توليد الإجابة التي نتوقعها منها.
-
-يتمثل أحد النهج للتغلب على هذا القصور في إنشاء "وكيل".
-
-الوكيل هو نظام يستخدم LLM كمحرك له، ولديه حق الوصول إلى وظائف تسمى "أدوات".
-
-هذه "الأدوات" هي وظائف لأداء مهمة، وتحتوي على جميع الأوصاف اللازمة للوكيل لاستخدامها بشكل صحيح.
-
-يمكن برمجة الوكيل للقيام بما يلي:
-- وضع سلسلة من الإجراءات/الأدوات وتشغيلها جميعًا في نفس الوقت مثل [`CodeAgent`] على سبيل المثال
-- التخطيط للاجراءات/الأدوات وتنفيذها واحدة تلو الأخرى والانتظار حتى انتهاء كل إجراء قبل إطلاق التالي مثل [`ReactJsonAgent`] على سبيل المثال
-
-### أنواع الوكلاء
-
-#### الوكيل البرمجي (Code agent)
-
-يتمتع هذا الوكيل يتبع خطوات محددة: أولًا، يخطط لسلسلة من الإجراءات التي يريد تنفيذها، ثم شفرة Python لتنفيذ جميع الإجراءات في نفس الوقت. وهو يتعامل بشكل أصلي مع أنواع مختلفة من المدخلات والمخرجات للأدوات التي يستخدمها، وبالتالي فهو الخيار الموصى به للمهام متعددة الوسائط.
-
-#### وكلاء التفاعل
-
-هذا هو الوكيل الذي يتم اللجوء إليه لحل مهام الاستدلال، حيث يجعل إطار ReAct ([Yao et al.، 2022](https://huggingface.co/papers/2210.03629)) من الكفاءة حقًا التفكير على أساس ملاحظاته السابقة.
-
-نقوم بتنفيذ إصدارين من ReactJsonAgent: 
-- [`ReactJsonAgent`] يقوم بتوليد استدعاءات الأدوات كـ JSON في إخراجها.
-- [`ReactCodeAgent`] هو نوع جديد من ReactJsonAgent يقوم بتوليد استدعاءات أدواته كمقاطع من التعليمات البرمجية، والتي تعمل بشكل جيد حقًا مع LLMs التي تتمتع بأداء  قوي في البرمجة.
-
-> [!TIP]
-> اقرأ منشور المدونة [Open-source LLMs as LangChain Agents](https://huggingface.co/blog/open-source-llms-as-agents) لمعرفة المزيد عن وكيل ReAct.
-
-![إطار عمل وكيل ReAct](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/open-source-llms-as-agents/ReAct.png)
-
-على سبيل المثال، إليك كيف يعمل وكيل ReAct Code طريقه من خلال السؤال التالي.
-
-```py3
->>> agent.run(
-...     "How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?",
-... )
-=====New task=====
-How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?
-====Agent is executing the code below:
-bert_blocks = search(query="number of blocks in BERT base encoder")
-print("BERT blocks:", bert_blocks)
-====
-Print outputs:
-BERT blocks: twelve encoder blocks
-
-====Agent is executing the code below:
-attention_layer = search(query="number of layers in Attention is All You Need")
-print("Attention layers:", attention_layer)
-====
-Print outputs:
-Attention layers: Encoder: The encoder is composed of a stack of N = 6 identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position- 2 Page 3 Figure 1: The Transformer - model architecture.
-
-====Agent is executing the code below:
-bert_blocks = 12
-attention_layers = 6
-diff = bert_blocks - attention_layers
-print("Difference in blocks:", diff)
-final_answer(diff)
-====
-
-Print outputs:
-Difference in blocks: 6
-
-Final answer: 6
-```
-
-### كيف يمكنني بناء وكيل؟
-
-لتهيئة وكيل، تحتاج إلى هذه الوسائط:
-
-- نموذج لغوي كبير (LLM) يشكل المحرك الأساسي للوكيل. الوكيل نفسه ليس النموذج اللغوي، بل هو برنامج يستخدم النموذج اللغوي كمحرك له.
-- موجه النظام (system prompt): هذه هي التعليمات التي يتم إعطاؤها للنموذج اللغوي لإنشاء مخرجاته.
-- صندوق أدوات (toolbox) يختار الوكيل منه الأدوات لتنفيذها
-- محلل (parser) لاستخراج الأدوات التي يجب استدعاؤها من مخرجات النموذج اللغوي LLM والأدوات التي يجب استخدامها
-
-عند تهيئة نظام الوكيل، يتم استخدام سمات الأداة لإنشاء وصف للأداة، ثم يتم دمجها في موجه النظام الخاص `system_prompt` للوكيل لإعلامه بالأدوات التي يمكنه استخدامها ولماذا.
-
-للبدء، يرجى تثبيت `agents` الإضافية لتثبيت جميع التبعيات الافتراضية.
-
-```bash
-pip install transformers[agents]
-```
-
-قم ببناء محرك LLM الخاص بك من خلال تعريف طريقة `llm_engine` التي تقبل قائمة من [الرسائل](./chat_templating.) وتعيد النص. يجب أن تقبل هذه الدالة القابلة للاستدعاء أيضًا معامل `stop` يشير إلى متى يجب التوقف عن التوليد.
-
-```python
-from huggingface_hub import login, InferenceClient
-
-login("<YOUR_HUGGINGFACEHUB_API_TOKEN>")
-
-client = InferenceClient(model="meta-llama/Meta-Llama-3-70B-Instruct")
-
-def llm_engine(messages, stop_sequences=["Task"]) -> str:
-    response = client.chat_completion(messages, stop=stop_sequences, max_tokens=1000)
-    answer = response.choices[0].message.content
-    return answer
-```
-
-يمكنك استخدام أي طريقة `llm_engine` طالما أنها:
-1. يتبع تنسيق [رسائل](./chat_templating.md) لإدخاله (`List [Dict [str، str]]`) ويعيد `str`
-2. يتوقف عن توليد المخراجات من التسلسلات التي تم تمريرها في معامل `stop`
-
-أنت بحاجة أيضًا إلى معامل "الأدوات" الذي يقبل قائمة من "الأدوات". يمكنك توفير قائمة فارغة لـ "الأدوات"، ولكن استخدم صندوق الأدوات الافتراضي مع معامل اختياري `add_base_tools=True`.
-
-الآن يمكنك إنشاء وكيل، مثل [`CodeAgent`], وتشغيله. ولتسهيل الأمر، نقدم أيضًا فئة [`HfEngine`] التي تستخدم `huggingface_hub.InferenceClient` بشكل مخفى.
-
-```python
-from transformers import CodeAgent, HfEngine
-
-llm_engine = HfEngine(model="meta-llama/Meta-Llama-3-70B-Instruct")
-agent = CodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-
-agent.run(
-    "Could you translate this sentence from French, say it out loud and return the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-هذه الميزة ستكون مفيدة في حالة الحاجة الملحة! يمكنك حتى ترك معامل `llm_engine` غير محدد، وسيتم إنشاء [`HfEngine`] بشكل تلقائي.
-
-```python
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], add_base_tools=True)
-
-agent.run(
-    "Could you translate this sentence from French, say it out loud and give me the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-لاحظ أننا استخدمنا معامل "sentence" إضافي: يمكنك تمرير النص كمعامل إضافي إلى النموذج.
-
-يمكنك أيضًا استخدام هذا للإشارة إلى مسار الملفات المحلية أو البعيدة للنموذج لاستخدامها:
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-
-agent.run("Why does Mike not know many people in New York?", audio="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/recording.mp3")
-```
-
-
-تم تحديد موجه النظام ومحلل المخرجات تلقائيًا، ولكن يمكنك فحصهما بسهولة عن طريق استدعاء `system_prompt_template` على وكيلك.
-
-```python
-print(agent.system_prompt_template)
-```
-
-من المهم أن تشرح بأكبر قدر ممكن من الوضوح المهمة التي تريد تنفيذها.
-كل عملية [`~Agent.run`] مستقلة، وبما أن الوكيل مدعوم من LLM، فقد تؤدي الاختلافات الطفيفة في موجهك إلى نتائج مختلفة تمامًا.
-يمكنك أيضًا تشغيل وكيل بشكل متتالي لمهام مختلفة: في كل مرة يتم فيها إعادة تهيئة سمتي `agent.task` و`agent.logs`.
-
-
-#### تنفيذ التعليمات البرمجية
-
-يقوم مفسر Python بتنفيذ التعليمات البرمجية على مجموعة من المدخلات التي يتم تمريرها جنبًا إلى جنب مع أدواتك.
-يجب أن يكون هذا الأمر آمنًا لأن الوظائف الوحيدة التي يمكن استدعاؤها هي الأدوات التي قدمتها (خاصة إذا كانت أدوات من Hugging Face فقط) ووظيفة الطباعة، لذا فأنت مقيد بالفعل بما يمكن تنفيذه.
-
-مفسر Python لا يسمح أيضًا باستدعاء دوال بشكل افتراضي خارج قائمة آمنة، لذا فإن جميع الهجمات الأكثر وضوحًا لا ينبغي أن تكون مشكلة.
-يمكنك أيضًا الإذن باستيرادات إضافية عن طريق تمرير الوحدات النمطية المصرح بها كقائمة من السلاسل في معامل  `additional_authorized_imports` عند تهيئة [`ReactCodeAgent`] أو [`CodeAgent`]:
-
-```py
->>> 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'?")
-
-(...)
-'Hugging Face – Blog'
-```
-
-سيتم إيقاف التنفيذ عند أي رمز يحاول تنفيذ عملية غير قانونية أو إذا كان هناك خطأ Python عادي في التعليمات البرمجية التي تم إنشاؤها بواسطة الوكيل.
-
-> [!WARNING]
-> يمكن لـ LLM توليد شفرة برمجية عشوائية سيتم تنفيذها بعد ذلك: لا تقمب استدعاء أى دوال غير آمنة!
-
-### موجه النظام
-
-ينشئ الوكيل، أو بالأحرى LLM الذي يقود الوكيل، يولد مخرجات بناءً على موجه النظام. يمكن تخصيص موجه النظام وتصميمه للمهام المقصودة. على سبيل المثال، تحقق من موجه النظام لـ [`ReactCodeAgent`] (الإصدار أدناه مبسط قليلاً).
-
-```text
-You will be given a task to solve as best you can.
-You have access to the following tools:
-<<tool_descriptions>>
-
-To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
-
-At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task, then the tools that you want to use.
-Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '/End code' sequence.
-During each intermediate step, you can use 'print()' to save whatever important information you will then need.
-These print outputs will then be available in the 'Observation:' field, for using this information as input for the next step.
-
-In the end you have to return a final answer using the `final_answer` tool.
-
-Here are a few examples using notional tools:
----
-{examples}
-
-Above example were using notional tools that might not exist for you. You only have access to those tools:
-<<tool_names>>
-You also can perform computations in the python code you generate.
-
-Always provide a 'Thought:' and a 'Code:\n```py' sequence ending with '```<end_code>' sequence. You MUST provide at least the 'Code:' sequence to move forward.
-
-Remember to not perform too many operations in a single code block! You should split the task into intermediate code blocks.
-Print results at the end of each step to save the intermediate results. Then use final_answer() to return the final result.
-
-Remember to make sure that variables you use are all defined.
-
-Now Begin!
-```
-
-يتضمن موجه النظام:
-- *مقدمة* تشرح كيف يجب أن يتصرف الوكيل والأدوات التي يجب عليه استخدامها.
-- وصف لجميع الأدوات التي يتم تحديدها بواسطة رمز `<<tool_descriptions>>` الذي يتم استبداله ديناميكيًا في وقت التشغيل بالأدوات التي يحددها المستخدم أو يختارها.
-    - يأتي وصف الأداة من سمات الأداة، `name`، و`description`، و`inputs` و`output_type`، وقالب `jinja2` بسيط يمكنك تحسينه.
-- شكل المخرج المتوقع.
-
-يمكنك تحسين موجه النظام، على سبيل المثال، عن طريق إضافة شرح لتنسيق المخرجات.
-
-للحصول على أقصى قدر من المرونة، يمكنك الكتابة فوق قالب موجه النظام بالكامل عن طريق تمرير موجه مخصص كمعامل إلى معلمة `system_prompt`.
-
-```python
-from transformers import ReactJsonAgent
-from transformers.agents import PythonInterpreterTool
-
-agent = ReactJsonAgent(tools=[PythonInterpreterTool()], system_prompt="{your_custom_prompt}")
-```
-
-> [!WARNING]
-> يرجى التأكد من تحديد سلسلة `<<tool_descriptions>>` في مكان ما في `template` حتى يكون الوكيل على علم 
-بالأدوات المتاحة.
-
-
-### فحص تشغيل الوكيل
-
-فيما يلي بعض السمات المفيدة لفحص ما حدث بعد التشغيل:
-- تخزن  `agent.logs` سجلات مفصلة للوكيل. في كل خطوة من تشغيل الوكيل، يتم تخزين كل شيء في قاموس إلحاقه بـ `agent.logs`.
-- تشغيل `agent.write_inner_memory_from_logs()` يخلق ذاكرة داخلية لسجلات الوكيل للنظام LLM لعرضها، كقائمة من رسائل الدردشة. تنتقل هذه الطريقة عبر كل خطوة من سجل الوكيل ولا تخزن سوى ما يهمها كرسالة: على سبيل المثال، سيحفظ موجه النظام والمهمة في رسائل منفصلة، ثم لكل خطوة سيخزن مخرج LLM كرسالة، ومخرج استدعاء الأداة كرسالة أخرى. استخدم هذا إذا كنت تريد عرضًا عامًا لما حدث - ولكن لن يتم نسخ كل سجل بواسطة هذه الطريقة.
-
-## الأدوات
-
-الأداة هي عبارة عن وظيفة أساسية يستخدمها الوكيل لتنفيذ مهمة محددة.
-
-يمكنك على سبيل المثال التحقق من [`PythonInterpreterTool`]: لديه اسم ووصف ووصف للمدخلات ونوع للمخرج، وطريقة `__call__` التي تقوم بتنفيذ المهمة المطلوبة.
-
-عند تهيئة الوكيل، يتم استخدام سمات الأداة لتوليد وصف للأداة يتم تضمينه في موجه النظام الخاص بالوكيل. يتيح هذا للوكيل معرفة الأدوات التي يمكنه استخدامها ولماذا.
-
-### صندوق الأدوات الافتراضي
-
-يأتي Transformers مع صندوق أدوات افتراضي لتمكين الوكلاء، والذي يمكنك إضافته إلى وكيلك عند التهيئة باستخدام معامل `add_base_tools = True`:
-
-- **الإجابة على أسئلة المستند**: الإجابة على سؤال حول المستند (مثل ملف PDF) بتنسيق صورة ([Donut](./model_doc/donut))
-- **الإجابة على أسئلة الصور**: الإجابة على سؤال حول صورة ([VILT](./model_doc/vilt))
-- **التحدث إلى النص**: قم بتفريغ الكلام إلى نص ([Whisper](./model_doc/whisper))
-- **النص إلى كلام**: تحويل النص إلى كلام ([SpeechT5](./model_doc/speecht5))
-- **الترجمة**: ترجمة جملة معينة من لغة المصدر إلى لغة الهدف.
-- **مفسر كود Python**: تشغيل كود Python الذي تم إنشاؤه بواسطة LLM في بيئة آمنة. لن يتم إضافة هذه الأداة إلى [`ReactJsonAgent`] إلا إذا استخدمت `add_base_tools=True`، نظرًا لأن الأدوات المستندة إلى التعليمات البرمجية يمكنها بالفعل تنفيذ كود Python
-لا تترجم النصوص الخاصة ولا الأكواد البرمجية ولا الروابط ولا رموز HTML وCSS:
-
-يمكنك استخدام أداة يدويًا عن طريق استدعاء دالة [`load_tool`] وتحديد مهمة لتنفيذها.
-
-```python
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-### إنشاء أداة جديدة
-
-يمكنك إنشاء أداتك الخاصة لتغطية حالات الاستخدام التي لا تغطيها الأدوات الافتراضية من Hugging Face.
-على سبيل المثال، دعنا نقوم بإنشاء أداة تعرض النموذج الأكثر تنزيلًا لمهمة معينة من Hub.
-
-سوف نبدأ بالكود التالي.
-
-```python
-from huggingface_hub import list_models
-
-task = "text-classification"
-
-model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-print(model.id)
-```
-
-يمكن تحويل هذه الشيفرة إلى فئة ترث من الفئة العليا [`Tool`].
-
-تحتاج الأداة المخصصة إلى:
-
-- اسم `name`، والتي تمثل اسم الأداة نفسها. عادةً ما يصف الاسم وظيفتها. بما أن الكود يعيد النموذج الأكثر تنزيلًا لمهمة ما، فلنسمها `model_download_counter`.
-- تستخدم خاصية `description` لملء موجه نظام الوكيل.
-- خاصية `inputs`، والتي هي عبارة عن قاموس بمفاتيح "type" و"description". يحتوي على معلومات تساعد المفسر Python على اتخاذ خيارات مستنيرة بشأن المدخلات.
-- خاصية `output_type`، والتي تحدد نوع المخرج.
-- طريقة `forward` والتي تحتوي على الكود الذي سيتم تنفيذه للحصول على النتيجة النهائية.
-
-```python
-from transformers import Tool
-from huggingface_hub import list_models
-
-class HFModelDownloadsTool(Tool):
-    name = "model_download_counter"
-    description = (
-        "This is a tool that returns the most downloaded model of a given task on the Hugging Face Hub. "
-        "It returns the name of the checkpoint."
-    )
-
-    inputs = {
-        "task": {
-            "type": "text",
-            "description": "the task category (such as text-classification, depth-estimation, etc)",
-        }
-    }
-    output_type = "text"
-
-    def forward(self, task: str):
-        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-        return model.id
-```
-
-الآن بعد أن أصبحت فئة `HfModelDownloadsTool` المخصصة جاهزة، يمكنك حفظها في ملف باسم `model_downloads.py` واستيرادها للاستخدام.
-
-```python
-from model_downloads import HFModelDownloadsTool
-
-tool = HFModelDownloadsTool()
-```
-
-يمكنك أيضًا مشاركة أداتك المخصصة في Hub عن طريق استدعاء [`~Tool.push_to_hub`] على الأداة. تأكد من أنك قمت بإنشاء مستودع لها على Hub وأنك تستخدم رمز وصول للقراءة.
-
-```python
-tool.push_to_hub("{your_username}/hf-model-downloads")
-```
-
-قم بتحميل الأداة باستخدام دالة [`~Tool.load_tool`] ومررها إلى معلمة `tools` في الوكيل الخاص بك.
-
-```python
-from transformers import load_tool, CodeAgent
-
-model_download_tool = load_tool("m-ric/hf-model-downloads")
-agent = CodeAgent(tools=[model_download_tool], llm_engine=llm_engine)
-agent.run(
-    "Can you give me the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?"
-)
-```
-
-ستحصل على ما يلي:
-
-```text
-======== New task ========
-Can you give me the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?
-==== Agent is executing the code below:
-most_downloaded_model = model_download_counter(task="text-to-video")
-print(f"The most downloaded model for the 'text-to-video' task is {most_downloaded_model}.")
-====
-```
-
-والناتج:
-
-`"النموذج الأكثر تنزيلًا لمهمة `text-to-video` هو ByteDance/AnimateDiff-Lightning."`
-
-### إدارة صندوق أدوات الوكيل الخاص بك
-
-إذا كنت قد قمت بتهيئة وكيل، فمن غير الملائم إعادة تهيئته من البداية لإضافة أداة جديدة ترغب في استخدامها. باستخدام مكتبة Transformers، يمكنك إدارة صندوق أدوات الوكيل بإضافة أو استبدال أداة موجودة.
-
-دعنا نضيف الأداة `model_download_tool` إلى وكيل تم تهيئته مسبقًا باستخدام صندوق الأدوات الافتراضي.
-
-```python
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-agent.toolbox.add_tool(model_download_tool)
-```
-
-الآن يمكننا الاستفادة من الأداة الجديدة وأداة تحويل النص إلى كلام السابقة:
-
-```python
-    agent.run(
-        "Can you read out loud the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub and return the audio?"
-    )
-```
-
-| **Audio**                                                                                                                                            |
-|------------------------------------------------------------------------------------------------------------------------------------------------------|
-| <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/damo.wav" type="audio/wav"/> |
-
-> [!WARNING]
-> احترس عند إضافة أدوات إلى وكيل يعمل بالفعل لأنه يمكن أن يؤثر على اختيار الأداة لصالح أداتك أو اختيار أداة أخرى غير المحددة بالفعل.
-
-استخدم طريقة `agent.toolbox.update_tool()` لاستبدال أداة موجودة في صندوق أدوات الوكيل.
-هذا مفيد إذا كانت أداتك الجديدة بديلاً مباشرًا للأداة الموجودة لأن الوكيل يعرف بالفعل كيفية تنفيذ تلك المهمة المحددة.
-تأكد فقط من اتباع الأداة الجديدة لنفس واجهة برمجة التطبيقات (API) للأداة المستبدلة أو قم بتكييف قالب موجه النظام لضمان تحديث جميع الأمثلة التي تستخدم الأداة المستبدلة.
-
-### استخدام مجموعة من الأدوات
-
-يمكنك الاستفادة من مجموعات الأدوات باستخدام كائن ToolCollection، مع تحديد مجموعة الأدوات التي تريد استخدامها.
-ثم قم بتمريرها كقائمة لتهيئة الوكيل الخاص بك، وبدء استخدامها!
-
-```py
-from transformers import ToolCollection, ReactCodeAgent
-
-image_tool_collection = ToolCollection(collection_slug="huggingface-tools/diffusion-tools-6630bb19a942c2306a2cdb6f")
-agent = ReactCodeAgent(tools=[*image_tool_collection.tools], add_base_tools=True)
-
-agent.run("Please draw me a picture of rivers and lakes.")
-```
-
-لتسريع البداية، يتم تحميل الأدوات فقط إذا استدعاها الوكيل.
-
-ستحصل على هذه الصورة:
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" />
-
-### استخدام gradio-tools
-
-[gradio-tools](https://github.com/freddyaboulton/gradio-tools) هي مكتبة قوية تتيح استخدام Hugging
-Face Spaces كأدوات. تدعم العديد من المساحات الموجودة بالإضافة إلى مساحات مخصصة.
-
-تدعم مكتبة Transformers `gradio_tools` باستخدام طريقة [`Tool.from_gradio`] في الفئة. على سبيل المثال، دعنا نستخدم [`StableDiffusionPromptGeneratorTool`](https://github.com/freddyaboulton/gradio-tools/blob/main/gradio_tools/tools/prompt_generator.py) من مجموعة أدوات `gradio-tools` لتحسين المطالبات لإنشاء صور أفضل.
-
-استورد وقم بتهيئة الأداة، ثم مررها إلى طريقة `Tool.from_gradio`:
-
-```python
-from gradio_tools import StableDiffusionPromptGeneratorTool
-from transformers import Tool, load_tool, CodeAgent
-
-gradio_prompt_generator_tool = StableDiffusionPromptGeneratorTool()
-prompt_generator_tool = Tool.from_gradio(gradio_prompt_generator_tool)
-```
-
-الآن يمكنك استخدامه مثل أي أداة أخرى. على سبيل المثال، دعنا نحسن الموجه `a rabbit wearing a space suit`.
-
-```python
-image_generation_tool = load_tool('huggingface-tools/text-to-image')
-agent = CodeAgent(tools=[prompt_generator_tool, image_generation_tool], llm_engine=llm_engine)
-
-agent.run(
-    "Improve this prompt, then generate an image of it.", prompt='A rabbit wearing a space suit'
-)
-```
-
-يستفيد النموذج بشكل كافٍ من الأداة:
-
-```text
-======== New task ========
-Improve this prompt, then generate an image of it.
-You have been provided with these initial arguments: {'prompt': 'A rabbit wearing a space suit'}.
-==== Agent is executing the code below:
-improved_prompt = StableDiffusionPromptGenerator(query=prompt)
-while improved_prompt == "QUEUE_FULL":
-    improved_prompt = StableDiffusionPromptGenerator(query=prompt)
-print(f"The improved prompt is {improved_prompt}.")
-image = image_generator(prompt=improved_prompt)
-====
-```
-
-قبل إنشاء الصورة أخيرًا:
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit_spacesuit_flux.webp" />
-
-> [!WARNING]
-> تتطلب gradio-tools إدخالات وإخراجات *نصية* حتى عند العمل مع طرائق مختلفة مثل كائنات الصور والصوت. الإدخالات والإخراجات الصورية والصوتية غير متوافقة حاليًا.
-
-### استخدام أدوات LangChain
-
-نحن نحب Langchain ونعتقد أنها تحتوي على مجموعة أدوات قوية للغاية.
-لاستيراد أداة من LangChain، استخدم الطريقة `from_langchain()`.
-
-فيما يلي كيفية استخدامها لإعادة إنشاء نتيجة البحث في المقدمة باستخدام أداة بحث الويب LangChain.
-
-```python
-from langchain.agents import load_tools
-from transformers import Tool, ReactCodeAgent
-
-search_tool = Tool.from_langchain(load_tools(["serpapi"])[0])
-
-agent = ReactCodeAgent(tools=[search_tool])
-
-agent.run("How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?")
-```
-
-## واجهة Gradio
-
-يمكنك الاستفادة من `gradio.Chatbot` لعرض أفكار الوكيل الخاص بك باستخدام `stream_to_gradio`، إليك مثال:
-
-```py
-import gradio as gr
-from transformers import (
-    load_tool,
-    ReactCodeAgent,
-    HfEngine,
-    stream_to_gradio,
-)
-
-# Import tool from Hub
-image_generation_tool = load_tool("m-ric/text-to-image")
-
-llm_engine = HfEngine("meta-llama/Meta-Llama-3-70B-Instruct")
-
-# Initialize the agent with the image generation tool
-agent = ReactCodeAgent(tools=[image_generation_tool], llm_engine=llm_engine)
-
-
-def interact_with_agent(task):
-    messages = []
-    messages.append(gr.ChatMessage(role="user", content=task))
-    yield messages
-    for msg in stream_to_gradio(agent, task):
-        messages.append(msg)
-        yield messages + [
-            gr.ChatMessage(role="assistant", content="⏳ Task not finished yet!")
-        ]
-    yield messages
-
-
-with gr.Blocks() as demo:
-    text_input = gr.Textbox(lines=1, label="Chat Message", value="Make me a picture of the Statue of Liberty.")
-    submit = gr.Button("Run illustrator agent!")
-    chatbot = gr.Chatbot(
-        label="Agent",
-        type="messages",
-        avatar_images=(
-            None,
-            "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png",
-        ),
-    )
-    submit.click(interact_with_agent, [text_input], [chatbot])
-
-if __name__ == "__main__":
-    demo.launch()
-```

docs/source/ar/gguf.md

@@ -77,7 +77,7 @@ model = AutoModelForCausalLM.from_pretrained(model_id, gguf_file=filename)
 
 الآن لديك إمكانية الوصول إلى النسخة الكامل غير المكممة للنموذج في بيئة PyTorch، حيث يمكنك دمجه مع مجموعة كبيرة من الأدوات الأخرى.
 
-لإعادة التحويل إلى ملف `gguf`، نوصي باستخدام ملف [`convert-hf-to-gguf.py`](https://github.com/ggerganov/llama.cpp/blob/master/convert-hf-to-gguf.py) من llama.cpp.
+لإعادة التحويل إلى ملف `gguf`، نوصي باستخدام ملف [`convert-hf-to-gguf.py`](https://github.com/ggerganov/llama.cpp/blob/master/convert_hf_to_gguf.py) من llama.cpp.
 
 فيما يلي كيفية إكمال البرنامج النصي أعلاه لحفظ النموذج وإعادة تصديره مرة أخرى إلى `gguf`:
 

docs/source/ar/trainer.md

@@ -674,29 +674,7 @@ use_cpu: false
 ```
 
 </hfoption>
-<hfoption id="Tensor Parallelism with PyTorch 2">
 
-```yml
-compute_environment: LOCAL_MACHINE
-tp_config:
-  tp_size: 4
-distributed_type: TP
-downcast_bf16: 'no'
-machine_rank: 0
-main_training_function: main
-mixed_precision: 'no'
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
-
-```
-
-</hfoption>
 </hfoptions>
 يُعد أمر  [`accelerate_launch`](https://huggingface.co/docs/accelerate/package_reference/cli#accelerate-launch) هو الطريقة المُوصى بها لتشغيل نص البرمجى للتدريب على نظام موزع باستخدام Accelerate و [`Trainer`] مع المعلمات المحددة في `config_file.yaml`. يتم حفظ هذا الملف في مجلد ذاكرة التخزين المؤقت لـ Accelerate ويتم تحميله تلقائيًا عند تشغيل `accelerate_launch`.
 

docs/source/de/_toctree.yml

@@ -23,8 +23,6 @@
     title: Laden und Trainieren von Adaptern mit 🤗 PEFT
   - local: model_sharing
     title: Ein Modell teilen
-  - local: transformers_agents
-    title: Agents
   - local: llm_tutorial
     title: Generation with LLMs
   title: Tutorials
@@ -39,4 +37,4 @@
     title: Testen
   - local: pr_checks
     title: Überprüfung einer Pull Request
-  title: Contribute
+  title: Contribute

docs/source/de/transformers_agents.md

@@ -1,323 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Transformers Agents
-
-<Tip warning={true}>
-
-Transformers Agents ist eine experimentelle API, die jederzeit geändert werden kann. Die von den Agenten zurückgegebenen Ergebnisse
-zurückgegeben werden, können variieren, da sich die APIs oder die zugrunde liegenden Modelle ändern können.
-
-</Tip>
-
-Transformers Version v4.29.0, die auf dem Konzept von *Tools* und *Agenten* aufbaut. Sie können damit spielen in
-[dieses Colab](https://colab.research.google.com/drive/1c7MHD-T1forUPGcC_jlwsIptOzpG3hSj).
-
-Kurz gesagt, es bietet eine API für natürliche Sprache auf der Grundlage von Transformers: Wir definieren eine Reihe von kuratierten Tools und entwerfen einen 
-Agenten, um natürliche Sprache zu interpretieren und diese Werkzeuge zu verwenden. Es ist von vornherein erweiterbar; wir haben einige relevante Tools kuratiert, 
-aber wir werden Ihnen zeigen, wie das System einfach erweitert werden kann, um jedes von der Community entwickelte Tool zu verwenden.
-
-Beginnen wir mit einigen Beispielen dafür, was mit dieser neuen API erreicht werden kann. Sie ist besonders leistungsfähig, wenn es um 
-Sie ist besonders leistungsstark, wenn es um multimodale Aufgaben geht. Lassen Sie uns also eine Runde drehen, um Bilder zu erzeugen und Text vorzulesen.
-
-```py
-agent.run("Caption the following image", image=image)
-```
-
-| **Input**                                                                                                                   | **Output**                        |
-|-----------------------------------------------------------------------------------------------------------------------------|-----------------------------------|
-| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/beaver.png" width=200> | A beaver is swimming in the water |
-
----
-
-```py
-agent.run("Read the following text out loud", text=text)
-```
-| **Input**                                                                                                               | **Output**                                   |
-|-------------------------------------------------------------------------------------------------------------------------|----------------------------------------------|
-| A beaver is swimming in the water | <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tts_example.wav" type="audio/wav"> your browser does not support the audio element. </audio>
-
----
-
-```py
-agent.run(
-    "In the following `document`, where will the TRRF Scientific Advisory Council Meeting take place?",
-    document=document,
-)
-```
-| **Input**                                                                                                                   | **Output**     |
-|-----------------------------------------------------------------------------------------------------------------------------|----------------|
-| <img src="https://datasets-server.huggingface.co/assets/hf-internal-testing/example-documents/--/hf-internal-testing--example-documents/test/0/image/image.jpg" width=200> | ballroom foyer |
-
-## Schnellstart
-
-Bevor Sie `agent.run` verwenden können, müssen Sie einen Agenten instanziieren, der ein großes Sprachmodell (LLM) ist. 
-Wir bieten Unterstützung für openAI-Modelle sowie für OpenSource-Alternativen von BigCode und OpenAssistant. Die openAI
-Modelle sind leistungsfähiger (erfordern aber einen openAI-API-Schlüssel, können also nicht kostenlos verwendet werden); Hugging Face
-bietet kostenlosen Zugang zu Endpunkten für BigCode- und OpenAssistant-Modelle.
-
-To start with, please install the `agents` extras in order to install all default dependencies.
-```bash
-pip install transformers[agents]
-```
-
-Um openAI-Modelle zu verwenden, instanziieren Sie einen [`OpenAiAgent`], nachdem Sie die `openai`-Abhängigkeit installiert haben:
-
-```bash
-pip install openai
-```
-
-
-```py
-from transformers import OpenAiAgent
-
-agent = OpenAiAgent(model="text-davinci-003", api_key="<your_api_key>")
-```
-
-Um BigCode oder OpenAssistant zu verwenden, melden Sie sich zunächst an, um Zugriff auf die Inference API zu erhalten:
-
-```py
-from huggingface_hub import login
-
-login("<YOUR_TOKEN>")
-```
-
-Dann instanziieren Sie den Agenten
-
-```py
-from transformers import HfAgent
-
-# Starcoder
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
-# StarcoderBase
-# agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoderbase")
-# OpenAssistant
-# agent = HfAgent(url_endpoint="https://api-inference.huggingface.co/models/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5")
-```
-
-Dies geschieht mit der Inferenz-API, die Hugging Face derzeit kostenlos zur Verfügung stellt. Wenn Sie Ihren eigenen Inferenz
-Endpunkt für dieses Modell (oder einen anderen) haben, können Sie die obige URL durch Ihren URL-Endpunkt ersetzen.
-
-<Tip>
-
-StarCoder und OpenAssistant sind kostenlos und leisten bei einfachen Aufgaben bewundernswert gute Arbeit. Allerdings halten die Kontrollpunkte
-nicht, wenn es um komplexere Aufforderungen geht. Wenn Sie mit einem solchen Problem konfrontiert sind, empfehlen wir Ihnen, das OpenAI
-Modell auszuprobieren, das zwar leider nicht quelloffen ist, aber zur Zeit eine bessere Leistung erbringt.
-
-</Tip>
-
-Sie sind jetzt startklar! Lassen Sie uns in die beiden APIs eintauchen, die Ihnen jetzt zur Verfügung stehen.
-
-### Einzelne Ausführung (run)
-
-Die Methode der einmaligen Ausführung ist die Verwendung der [`~Agent.run`] Methode des Agenten:
-
-```py
-agent.run("Draw me a picture of rivers and lakes.")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200>
-
-Es wählt automatisch das (oder die) Werkzeug(e) aus, das (die) für die von Ihnen gewünschte Aufgabe geeignet ist (sind) und führt es (sie) entsprechend aus. Es
-kann eine oder mehrere Aufgaben in der gleichen Anweisung ausführen (je komplexer Ihre Anweisung ist, desto wahrscheinlicher ist ein
-der Agent scheitern).
-
-```py
-agent.run("Draw me a picture of the sea then transform the picture to add an island")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/sea_and_island.png" width=200>
-
-<br/>
-
-
-Jede [`~Agent.run`] Operation ist unabhängig, so dass Sie sie mehrmals hintereinander mit unterschiedlichen Aufgaben ausführen können.
-
-Beachten Sie, dass Ihr `Agent` nur ein großsprachiges Modell ist, so dass kleine Variationen in Ihrer Eingabeaufforderung völlig unterschiedliche Ergebnisse liefern können.
-unterschiedliche Ergebnisse liefern. Es ist wichtig, dass Sie die Aufgabe, die Sie ausführen möchten, so genau wie möglich erklären. Wir gehen noch weiter ins Detail
-wie man gute Prompts schreibt [hier](custom_tools#writing-good-user-inputs).
-
-Wenn Sie einen Status über Ausführungszeiten hinweg beibehalten oder dem Agenten Nicht-Text-Objekte übergeben möchten, können Sie dies tun, indem Sie
-Variablen, die der Agent verwenden soll. Sie könnten zum Beispiel das erste Bild von Flüssen und Seen erzeugen, 
-und das Modell bitten, dieses Bild zu aktualisieren und eine Insel hinzuzufügen, indem Sie Folgendes tun:
-
-```python
-picture = agent.run("Generate a picture of rivers and lakes.")
-updated_picture = agent.run("Transform the image in `picture` to add an island to it.", picture=picture)
-```
-
-<Tip>
-
-Dies kann hilfreich sein, wenn das Modell Ihre Anfrage nicht verstehen kann und die Werkzeuge verwechselt. Ein Beispiel wäre:
-
-```py
-agent.run("Draw me the picture of a capybara swimming in the sea")
-```
-
-Hier könnte das Modell auf zwei Arten interpretieren:
-- Die Funktion `Text-zu-Bild` erzeugt ein Wasserschwein, das im Meer schwimmt.
-- Oder Sie lassen das `Text-zu-Bild` ein Wasserschwein erzeugen und verwenden dann das Werkzeug `Bildtransformation`, um es im Meer schwimmen zu lassen.
-
-Falls Sie das erste Szenario erzwingen möchten, können Sie dies tun, indem Sie die Eingabeaufforderung als Argument übergeben:
-
-```py
-agent.run("Draw me a picture of the `prompt`", prompt="a capybara swimming in the sea")
-```
-
-</Tip>
-
-
-### Chat-basierte Ausführung (Chat)
-
-Der Agent verfügt auch über einen Chat-basierten Ansatz, der die Methode [`~Agent.chat`] verwendet:
-
-```py
-agent.chat("Generate a picture of rivers and lakes")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200> 
-
-```py
-agent.chat("Transform the picture so that there is a rock in there")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_and_beaver.png" width=200>
-
-<br/>
-
-Dies ist ein interessanter Ansatz, wenn Sie den Zustand über Anweisungen hinweg beibehalten möchten. Er ist besser für Experimente geeignet, 
-eignet sich aber eher für einzelne Anweisungen als für komplexe Anweisungen (die die [`~Agent.run`]
-Methode besser verarbeiten kann).
-
-Diese Methode kann auch Argumente entgegennehmen, wenn Sie Nicht-Text-Typen oder bestimmte Aufforderungen übergeben möchten.
-
-### ⚠️ Fernausführung
-
-Zu Demonstrationszwecken und damit es mit allen Setups verwendet werden kann, haben wir Remote-Executors für mehrere 
-der Standard-Tools erstellt, auf die der Agent in dieser Version Zugriff hat. Diese werden erstellt mit 
-[inference endpoints](https://huggingface.co/inference-endpoints).
-
-Wir haben diese vorerst deaktiviert, aber um zu sehen, wie Sie selbst Remote Executors Tools einrichten können,
-empfehlen wir die Lektüre des [custom tool guide](./custom_tools).
-
-### Was passiert hier? Was sind Tools und was sind Agenten?
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/diagram.png">
-
-#### Agenten
-
-Der "Agent" ist hier ein großes Sprachmodell, das wir auffordern, Zugang zu einem bestimmten Satz von Tools zu erhalten.
-
-LLMs sind ziemlich gut darin, kleine Codeproben zu erzeugen. Diese API macht sich das zunutze, indem sie das 
-LLM ein kleines Codebeispiel gibt, das eine Aufgabe mit einer Reihe von Werkzeugen ausführt. Diese Aufforderung wird dann ergänzt durch die 
-Aufgabe, die Sie Ihrem Agenten geben, und die Beschreibung der Werkzeuge, die Sie ihm geben. Auf diese Weise erhält er Zugriff auf die Dokumentation der 
-Tools, insbesondere die erwarteten Eingaben und Ausgaben, und kann den entsprechenden Code generieren.
-
-#### Tools
-
-Tools sind sehr einfach: Sie bestehen aus einer einzigen Funktion mit einem Namen und einer Beschreibung. Wir verwenden dann die Beschreibungen dieser Tools 
-um den Agenten aufzufordern. Anhand der Eingabeaufforderung zeigen wir dem Agenten, wie er die Tools nutzen kann, um das zu tun, was in der 
-in der Abfrage angefordert wurde.
-
-Dies geschieht mit brandneuen Tools und nicht mit Pipelines, denn der Agent schreibt besseren Code mit sehr atomaren Tools. 
-Pipelines sind stärker refaktorisiert und fassen oft mehrere Aufgaben in einer einzigen zusammen. Tools sind dafür gedacht, sich auf
-eine einzige, sehr einfache Aufgabe konzentrieren.
-
-#### Code-Ausführung?!
-
-Dieser Code wird dann mit unserem kleinen Python-Interpreter auf den mit Ihren Tools übergebenen Eingaben ausgeführt. 
-Wir hören Sie schon schreien "Willkürliche Codeausführung!", aber lassen Sie uns erklären, warum das nicht der Fall ist.
-
-Die einzigen Funktionen, die aufgerufen werden können, sind die von Ihnen zur Verfügung gestellten Tools und die Druckfunktion, so dass Sie bereits eingeschränkt sind 
-eingeschränkt, was ausgeführt werden kann. Sie sollten sicher sein, wenn es sich auf die Werkzeuge für das Umarmungsgesicht beschränkt. 
-
-Dann lassen wir keine Attributsuche oder Importe zu (die ohnehin nicht benötigt werden, um die 
-Inputs/Outputs an eine kleine Gruppe von Funktionen), so dass alle offensichtlichen Angriffe (und Sie müssten den LLM 
-dazu auffordern, sie auszugeben) kein Problem darstellen sollten. Wenn Sie auf Nummer sicher gehen wollen, können Sie die 
-run()-Methode mit dem zusätzlichen Argument return_code=True ausführen. In diesem Fall gibt der Agent nur den auszuführenden Code 
-zur Ausführung zurück und Sie können entscheiden, ob Sie ihn ausführen möchten oder nicht.
-
-Die Ausführung bricht bei jeder Zeile ab, in der versucht wird, eine illegale Operation auszuführen, oder wenn ein regulärer Python-Fehler 
-mit dem vom Agenten generierten Code.
-
-### Ein kuratierter Satz von Tools
-
-Wir haben eine Reihe von Tools identifiziert, die solche Agenten unterstützen können. Hier ist eine aktualisierte Liste der Tools, die wir integriert haben 
-in `transformers` integriert haben:
-
-- **Beantwortung von Fragen zu Dokumenten**: Beantworten Sie anhand eines Dokuments (z.B. PDF) im Bildformat eine Frage zu diesem Dokument ([Donut](./model_doc/donut))
-- Beantworten von Textfragen**: Geben Sie einen langen Text und eine Frage an, beantworten Sie die Frage im Text ([Flan-T5](./model_doc/flan-t5))
-- **Unbedingte Bildunterschriften**: Beschriften Sie das Bild! ([BLIP](./model_doc/blip))
-- **Bildfragebeantwortung**: Beantworten Sie bei einem Bild eine Frage zu diesem Bild ([VILT](./model_doc/vilt))
-- **Bildsegmentierung**: Geben Sie ein Bild und einen Prompt an und geben Sie die Segmentierungsmaske dieses Prompts aus ([CLIPSeg](./model_doc/clipseg))
-- **Sprache in Text**: Geben Sie eine Audioaufnahme einer sprechenden Person an und transkribieren Sie die Sprache in Text ([Whisper](./model_doc/whisper))
-- **Text in Sprache**: wandelt Text in Sprache um ([SpeechT5](./model_doc/speecht5))
-- **Zero-Shot-Textklassifizierung**: Ermitteln Sie anhand eines Textes und einer Liste von Bezeichnungen, welcher Bezeichnung der Text am ehesten entspricht ([BART](./model_doc/bart))
-- **Textzusammenfassung**: fassen Sie einen langen Text in einem oder wenigen Sätzen zusammen ([BART](./model_doc/bart))
-- **Übersetzung**: Übersetzen des Textes in eine bestimmte Sprache ([NLLB](./model_doc/nllb))
-
-Diese Tools sind in Transformatoren integriert und können auch manuell verwendet werden, zum Beispiel:
-
-```py
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-### Benutzerdefinierte Tools
-
-Wir haben zwar eine Reihe von Tools identifiziert, sind aber der festen Überzeugung, dass der Hauptwert dieser Implementierung darin besteht 
-die Möglichkeit, benutzerdefinierte Tools schnell zu erstellen und weiterzugeben.
-
-Indem Sie den Code eines Tools in einen Hugging Face Space oder ein Modell-Repository stellen, können Sie das Tool 
-direkt mit dem Agenten nutzen. Wir haben ein paar neue Funktionen hinzugefügt 
-**transformers-agnostic** Tools zur [`huggingface-tools` Organisation](https://huggingface.co/huggingface-tools) hinzugefügt:
-
-- **Text-Downloader**: zum Herunterladen eines Textes von einer Web-URL
-- **Text zu Bild**: erzeugt ein Bild nach einer Eingabeaufforderung und nutzt dabei stabile Diffusion
-- **Bildtransformation**: verändert ein Bild anhand eines Ausgangsbildes und einer Eingabeaufforderung, unter Ausnutzung der stabilen pix2pix-Diffusion
-- **Text zu Video**: Erzeugen eines kleinen Videos nach einer Eingabeaufforderung, unter Verwendung von damo-vilab
-
-Das Text-zu-Bild-Tool, das wir von Anfang an verwendet haben, ist ein Remote-Tool, das sich in 
-[*huggingface-tools/text-to-image*](https://huggingface.co/spaces/huggingface-tools/text-to-image)! Wir werden
-weiterhin solche Tools für diese und andere Organisationen veröffentlichen, um diese Implementierung weiter zu verbessern.
-
-Die Agenten haben standardmäßig Zugriff auf die Tools, die sich auf [*huggingface-tools*](https://huggingface.co/huggingface-tools) befinden.
-Wie Sie Ihre eigenen Tools schreiben und freigeben können und wie Sie jedes benutzerdefinierte Tool, das sich auf dem Hub befindet, nutzen können, erklären wir in [folgender Anleitung](custom_tools).
-
-### Code-Erzeugung
-
-Bisher haben wir gezeigt, wie Sie die Agenten nutzen können, um Aktionen für Sie durchzuführen. Der Agent generiert jedoch nur Code
-den wir dann mit einem sehr eingeschränkten Python-Interpreter ausführen. Falls Sie den generierten Code in einer anderen Umgebung verwenden möchten 
-einer anderen Umgebung verwenden möchten, können Sie den Agenten auffordern, den Code zusammen mit einer Tooldefinition und genauen Importen zurückzugeben.
-
-Zum Beispiel die folgende Anweisung
-```python
-agent.run("Draw me a picture of rivers and lakes", return_code=True)
-```
-
-gibt den folgenden Code zurück
-
-```python
-from transformers import load_tool
-
-image_generator = load_tool("huggingface-tools/text-to-image")
-
-image = image_generator(prompt="rivers and lakes")
-```
-
-die Sie dann selbst ändern und ausführen können.

docs/source/en/_toctree.yml

@@ -161,6 +161,10 @@
   sections:
   - local: quantization/overview
     title: Overview
+  - local: quantization/selecting
+    title: Selecting a quantization method
+  - local: quantization/concept_guide
+    title: Quantization concepts
   - local: quantization/aqlm
     title: AQLM
   - local: quantization/awq
@@ -279,6 +283,8 @@
         title: Image-text-to-text
       - local: tasks/video_text_to_text
         title: Video-text-to-text
+      - local: tasks/visual_document_retrieval
+        title: Visual Document Retrieval
       title: Multimodal
     title: Task recipes
   - local: run_scripts
@@ -306,8 +312,6 @@
 - isExpanded: false
   sections:
   - sections:
-    - local: main_classes/agent
-      title: Agents and Tools
     - local: model_doc/auto
       title: Auto Classes
     - local: main_classes/backbones
@@ -461,6 +465,8 @@
         title: Gemma2
       - local: model_doc/glm
         title: GLM
+      - local: model_doc/glm4
+        title: glm4
       - local: model_doc/openai-gpt
         title: GPT
       - local: model_doc/gpt_neo
@@ -735,6 +741,8 @@
         title: Mask2Former
       - local: model_doc/maskformer
         title: MaskFormer
+      - local: model_doc/mlcd
+        title: MLCD
       - local: model_doc/mobilenet_v1
         title: MobileNetV1
       - local: model_doc/mobilenet_v2
@@ -819,6 +827,8 @@
         title: EnCodec
       - local: model_doc/fastspeech2_conformer
         title: FastSpeech2Conformer
+      - local: model_doc/granite_speech
+        title: GraniteSpeech
       - local: model_doc/hubert
         title: Hubert
       - local: model_doc/mctct
@@ -943,6 +953,8 @@
         title: InstructBLIP
       - local: model_doc/instructblipvideo
         title: InstructBlipVideo
+      - local: model_doc/janus
+        title: Janus
       - local: model_doc/kosmos-2
         title: KOSMOS-2
       - local: model_doc/layoutlm
@@ -989,6 +1001,8 @@
         title: Pix2Struct
       - local: model_doc/pixtral
         title: Pixtral
+      - local: model_doc/qwen2_5_omni
+        title: Qwen2.5-Omni
       - local: model_doc/qwen2_5_vl
         title: Qwen2.5-VL
       - local: model_doc/qwen2_audio
@@ -1049,6 +1063,8 @@
         title: PatchTST
       - local: model_doc/time_series_transformer
         title: Time Series Transformer
+      - local: model_doc/timesfm
+        title: TimesFM
       title: Time series models
     - sections:
       - local: model_doc/graphormer
@@ -1074,6 +1090,8 @@
       title: Utilities for Audio processing
     - local: internal/file_utils
       title: General Utilities
+    - local: internal/import_utils
+      title: Importing Utilities
     - local: internal/time_series_utils
       title: Utilities for Time Series
     title: Internal helpers

docs/source/en/agents.md

@@ -15,283 +15,4 @@ rendered properly in your Markdown viewer.
 -->
 
 > [!WARNING]
-> Agents and tools are being spun out into the standalone [smolagents](https://huggingface.co/docs/smolagents/index) library. These docs will be deprecated in the future!
-
-# Agents
-
-[[open-in-colab]]
-
-An agent is a system where a large language model (LLM) can execute more complex tasks through *planning* and using *tools*.
-
-- Planning helps a LLM reason its way through a task by breaking it down into smaller subtasks. For example, [`CodeAgent`] plans a series of actions to take and then generates Python code to execute all the actions at once.
-
-    Another planning method is by self-reflection and refinement of its previous actions to improve its performance. The [`ReactJsonAgent`] is an example of this type of planning, and it's based on the [ReAct](https://hf.co/papers/2210.03629) framework. This agent plans and executes actions one at a time based on the feedback it receives from each action.
-
-- Tools give a LLM access to external functions or APIs that it can use to help it complete a task. For example, [gradio-tools](https://github.com/freddyaboulton/gradio-tools) gives a LLM access to any of the [Gradio](https://www.gradio.app/) apps available on Hugging Face [Spaces](https://hf.co/spaces). These apps can be used for a wide range of tasks such as image generation, video generation, audio transcription, and more.
-
-To use agents in Transformers, make sure you have the extra `agents` dependencies installed.
-
-```bash
-!pip install transformers[agents]
-```
-
-Create an agent instance (refer to the [Agents](./main_classes/agent#agents) API for supported agents in Transformers) and a list of tools available for it to use, then [`~ReactAgent.run`] the agent on your task. The example below demonstrates how a ReAct agent reasons through a task.
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[])
-agent.run(
-    "How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?",
-)
-```
-
-```bash
-======== New task ========
-How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?
-==== Agent is executing the code below:
-bert_layers = 12  # BERT base encoder has 12 layers
-attention_layers = 6  # Encoder in Attention is All You Need has 6 layers
-layer_diff = bert_layers - attention_layers
-print("The difference in layers between BERT base encoder and Attention is All You Need is", layer_diff)
-====
-Print outputs:
-The difference in layers between BERT base encoder and Attention is All You Need is 6
-
-==== Agent is executing the code below:
-final_answer("BERT base encoder has {} more layers than the encoder from Attention is All You Need.".format(layer_diff))
-====
-Print outputs:
-
->>> Final answer:
-BERT base encoder has 6 more layers than the encoder from Attention is All You Need.
-```
-
-This guide will walk you through in more detail how to initialize an agent.
-
-## LLM
-
-An agent uses a LLM to plan and execute a task; it is the engine that powers the agent. To choose and build your own LLM engine, you need a method that:
-
-1. the input uses the [chat template](./chat_templating) format, `List[Dict[str, str]]`, and it returns a string
-2. the LLM stops generating outputs when it encounters the sequences in `stop_sequences`
-
-```py
-def llm_engine(messages, stop_sequences=["Task"]) -> str:
-    response = client.chat_completion(messages, stop=stop_sequences, max_tokens=1000)
-    answer = response.choices[0].message.content
-    return answer
-```
-
-Next, initialize an engine to load a model. To run an agent locally, create a [`TransformersEngine`] to load a preinitialized [`Pipeline`].
-
-However, you could also leverage Hugging Face's powerful inference infrastructure, [Inference API](https://hf.co/docs/api-inference/index) or [Inference Endpoints](https://hf.co/docs/inference-endpoints/index), to run your model. This is useful for loading larger models that are typically required for agentic behavior. In this case, load the [`HfApiEngine`] to run the agent.
-
-The agent requires a list of tools it can use to complete a task. If you aren't using any additional tools, pass an empty list. The default tools provided by Transformers are loaded automatically, but you can optionally set `add_base_tools=True` to explicitly enable them.
-
-<hfoptions id="engine">
-<hfoption id="TransformersEngine">
-
-```py
-from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TransformersEngine, CodeAgent
-
-tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
-model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.1-8B-Instruct").to("cuda")
-pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
-llm_engine = TransformersEngine(pipeline)
-agent = CodeAgent(tools=[], llm_engine=llm_engine)
-agent.run(
-    "What causes bread to rise?",
-)
-```
-
-</hfoption>
-<hfoption id="HfApiEngine">
-
-```py
-from transformers import CodeAgent, HfApiEngine
-
-llm_engine = HfApiEngine(model="meta-llama/Meta-Llama-3-70B-Instruct")
-agent = CodeAgent(tools=[], llm_engine=llm_engine)
-agent.run(
-    "Could you translate this sentence from French, say it out loud and return the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-</hfoption>
-</hfoptions>
-
-The agent supports [constrained generation](https://hf.co/docs/text-generation-inference/conceptual/guidance) for generating outputs according to a specific structure with the `grammar` parameter. The `grammar` parameter should be specified in the `llm_engine` method or you can set it when initializing an agent.
-
-Lastly, an agent accepts additional inputs such as text and audio. In the [`HfApiEngine`] example above, the agent accepted a sentence to translate. But you could also pass a path to a local or remote file for the agent to access. The example below demonstrates how to pass a path to an audio file.
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], llm_engine=llm_engine)
-agent.run("Why doesn't he know many people in New York?", audio="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/recording.mp3")
-```
-
-## System prompt
-
-A system prompt describes how an agent should behave, a description of the available tools, and the expected output format.
-
-Tools are defined by the `<<tool_descriptions>>` token which is dynamically replaced during runtime with the actual tool. The tool description is derived from the tool name, description, inputs, output type, and a Jinja2 template. Refer to the [Tools](./tools) guide for more information about how to describe tools.
-
-The example below is the system prompt for [`ReactCodeAgent`].
-
-```py
-You will be given a task to solve as best you can.
-You have access to the following tools:
-<<tool_descriptions>>
-
-To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
-
-At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task, then the tools that you want to use.
-Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '/End code' sequence.
-During each intermediate step, you can use 'print()' to save whatever important information you will then need.
-These print outputs will then be available in the 'Observation:' field, for using this information as input for the next step.
-
-In the end you have to return a final answer using the `final_answer` tool.
-
-Here are a few examples using notional tools:
----
-{examples}
-
-Above example were using notional tools that might not exist for you. You only have access to those tools:
-<<tool_names>>
-You also can perform computations in the python code you generate.
-
-Always provide a 'Thought:' and a 'Code:\n```py' sequence ending with '```<end_code>' sequence. You MUST provide at least the 'Code:' sequence to move forward.
-
-Remember to not perform too many operations in a single code block! You should split the task into intermediate code blocks.
-Print results at the end of each step to save the intermediate results. Then use final_answer() to return the final result.
-
-Remember to make sure that variables you use are all defined.
-
-Now Begin!
-```
-
-The system prompt can be tailored to the intended task. For example, you can add a better explanation of the output format or you can overwrite the system prompt template entirely with your own custom system prompt as shown below.
-
-> [!WARNING]
-> If you're writing a custom system prompt, make sure to include `<<tool_descriptions>>` in the template so the agent is aware of the available tools.
-
-```py
-from transformers import ReactJsonAgent
-from transformers.agents import PythonInterpreterTool
-
-agent = ReactJsonAgent(tools=[PythonInterpreterTool()], system_prompt="{your_custom_prompt}")
-```
-
-## Code execution
-
-For safety, only the tools you provide (and the default Transformers tools) and the `print` function are executed. The interpreter doesn't allow importing modules that aren't on a safe list.
-
-To import modules that aren't on the list, add them as a list to the `additional_authorized_imports` parameter when initializing an agent.
-
-```py
-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'?")
-```
-
-Code execution stops if a tool isn't on the safe list, it isn't authorized, or if the code generated by the agent returns a Python error.
-
-> [!WARNING]
-> A LLM can generate any arbitrary code that can be executed, so don't add any unsafe imports!
-
-## Multi-agent
-
-[Multi-agent](https://hf.co/papers/2308.08155) refers to multiple agents working together to solve a task. Performance is typically better because each agent is specialized for a particular subtask.
-
-Multi-agents are created through a [`ManagedAgent`] class, where a *manager agent* oversees how other agents work together. The manager agent requires an agent and their name and description. These are added to the manager agents system prompt which lets it know how to call and use them.
-
-The multi-agent example below creates a web search agent that is managed by another [`ReactCodeAgent`].
-
-```py
-from transformers.agents import ReactCodeAgent, HfApiEngine, DuckDuckGoSearchTool, ManagedAgent
-
-llm_engine = HfApiEngine()
-web_agent = ReactCodeAgent(tools=[DuckDuckGoSearchTool()], llm_engine=llm_engine)
-managed_web_agent = ManagedAgent(
-    agent=web_agent,
-    name="web_search",
-    description="Runs web searches for you. Give it your query as an argument."
-)
-manager_agent = ReactCodeAgent(
-    tools=[], llm_engine=llm_engine, managed_agents=[managed_web_agent]
-)
-manager_agent.run("Who is the CEO of Hugging Face?")
-```
-
-## Gradio integration
-
-[Gradio](https://www.gradio.app/) is a library for quickly creating and sharing machine learning apps. The [gradio.Chatbot](https://www.gradio.app/docs/gradio/chatbot) supports chatting with a Transformers agent with the [`stream_to_gradio`] function.
-
-Load a tool and LLM with an agent, and then create a Gradio app. The key is to use [`stream_to_gradio`] to stream the agents messages and display how it's reasoning through a task.
-
-```py
-import gradio as gr
-from transformers import (
-    load_tool,
-    ReactCodeAgent,
-    HfApiEngine,
-    stream_to_gradio,
-)
-
-# Import tool from Hub
-image_generation_tool = load_tool("m-ric/text-to-image")
-llm_engine = HfApiEngine("meta-llama/Meta-Llama-3-70B-Instruct")
-
-# Initialize the agent with the image generation tool
-agent = ReactCodeAgent(tools=[image_generation_tool], llm_engine=llm_engine)
-
-def interact_with_agent(task):
-    messages = []
-    messages.append(gr.ChatMessage(role="user", content=task))
-    yield messages
-    for msg in stream_to_gradio(agent, task):
-        messages.append(msg)
-        yield messages + [
-            gr.ChatMessage(role="assistant", content="⏳ Task not finished yet!")
-        ]
-    yield messages
-
-with gr.Blocks() as demo:
-    text_input = gr.Textbox(lines=1, label="Chat Message", value="Make me a picture of the Statue of Liberty.")
-    submit = gr.Button("Run illustrator agent!")
-    chatbot = gr.Chatbot(
-        label="Agent",
-        type="messages",
-        avatar_images=(
-            None,
-            "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png",
-        ),
-    )
-    submit.click(interact_with_agent, [text_input], [chatbot])
-
-if __name__ == "__main__":
-    demo.launch()
-```
-
-## Troubleshoot
-
-For a better idea of what is happening when you call an agent, it is always a good idea to check the system prompt template first.
-
-```py
-print(agent.system_prompt_template)
-```
-
-If the agent is behaving unexpectedly, remember to explain the task you want to perform as clearly as possible. Every [`~Agent.run`] is different and minor variations in your system prompt may yield completely different results.
-
-To find out what happened after a run, check the following agent attributes.
-
-- `agent.logs` stores the finegrained agent logs. At every step of the agents run, everything is stored in a dictionary and appended to `agent.logs`.
-- `agent.write_inner_memory_from_logs` only stores a high-level overview of the agents run. For example, at each step, it stores the LLM output as a message and the tool call output as a separate message. Not every detail from a step is transcripted by `write_inner_memory_from_logs`.
-
-## Resources
-
-Learn more about ReAct agents in the [Open-source LLMs as LangChain Agents](https://hf.co/blog/open-source-llms-as-agents) blog post.
+> Agents and tools were spun out into the standalone [smolagents](https://huggingface.co/docs/smolagents/index) library. They were removed from `transformers` in v4.52.

docs/source/en/chat_templating_multimodal.md

@@ -181,35 +181,6 @@ processed_chat = processor.apply_chat_template(
 print(processed_chat.keys())
 ```
 
-</hfoption>
-<hfoption id="custom frame sampling">
-
-Some models don't sample frames *uniformly* and require more complex logic to determine which frames to use. For example, the model may have an *adaptive frame selection* or if the model prioritizes *key moments* in a video rather than evenly spaced frames.
-
-If a model has a different sampling strategy, you can write a function that customizes frame selection. The function should include the following requirements.
-
-- Use the `sample_indices_fn` parameter to pass a callable function for sampling.
-- If provided, this function *overrides* the standard `num_frames` and `fps` parameters.
-- The function receives all the parameters passed to `load_video` and must return valid frame indices to sample from.
-
-An example function is shown below. This gives you full control over frame selection, making the model more adaptable to different video scenarios.
-
-```py
-def sample_indices_fn(metadata, **kwargs):
-    # samples only the first and the second frame
-    return [0, 1]
-
-processed_chat = processor.apply_chat_template(
-    messages,
-    add_generation_prompt=True,
-    tokenize=True,
-    return_dict=True,
-    sample_indices_fn=sample_indices_fn,
-    video_load_backend="decord",
-)
-print(processed_chat.keys())
-```
-
 </hfoption>
 <hfoption id="list of image frames">
 

docs/source/en/installation.md

@@ -20,7 +20,7 @@ rendered properly in your Markdown viewer.
 
 # Installation
 
-Transformers works with [PyTorch](https://pytorch.org/get-started/locally/), [TensorFlow 2.0](https://www.tensorflow.org/install/pip), and [Flax](https://flax.readthedocs.io/en/latest/). It has been tested on Python 3.9+, PyTorch 2.0+, TensorFlow 2.6+, and Flax 0.4.1+.
+Transformers works with [PyTorch](https://pytorch.org/get-started/locally/), [TensorFlow 2.0](https://www.tensorflow.org/install/pip), and [Flax](https://flax.readthedocs.io/en/latest/). It has been tested on Python 3.9+, PyTorch 2.1+, TensorFlow 2.6+, and Flax 0.4.1+.
 
 ## Virtual environment
 

docs/source/en/internal/import_utils.md

@@ -0,0 +1,91 @@
+<!--Copyright 2025 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.
+
+-->
+
+# Import Utilities
+
+This page goes through the transformers utilities to enable lazy and fast object import.
+While we strive for minimal dependencies, some models have specific dependencies requirements that cannot be
+worked around. We don't want for all users of `transformers` to have to install those dependencies to use other models,
+we therefore mark those as soft dependencies rather than hard dependencies.
+
+The transformers toolkit is not made to error-out on import of a model that has a specific dependency; instead, an
+object for which you are lacking a dependency will error-out when calling any method on it. As an example, if 
+`torchvision` isn't installed, the fast image processors will not be available. 
+
+This object is still importable:
+
+```python
+>>> from transformers import DetrImageProcessorFast
+>>> print(DetrImageProcessorFast)
+<class 'DetrImageProcessorFast'>
+```
+
+However, no method can be called on that object:
+
+```python
+>>> DetrImageProcessorFast.from_pretrained()
+ImportError: 
+DetrImageProcessorFast requires the Torchvision library but it was not found in your environment. Checkout the instructions on the
+installation page: https://pytorch.org/get-started/locally/ and follow the ones that match your environment.
+Please note that you may need to restart your runtime after installation.
+```
+
+Let's see how to specify specific object dependencies.
+
+## Specifying Object Dependencies
+
+### Filename-based
+
+All objects under a given filename have an automatic dependency to the tool linked to the filename
+
+**TensorFlow**: All files starting with `modeling_tf_` have an automatic TensorFlow dependency.
+
+**Flax**: All files starting with `modeling_flax_` have an automatic Flax dependency
+
+**PyTorch**: All files starting with `modeling_` and not valid with the above (TensorFlow and Flax) have an automatic 
+PyTorch dependency
+
+**Tokenizers**: All files starting with `tokenization_` and ending with `_fast` have an automatic `tokenizers` dependency
+
+**Vision**: All files starting with `image_processing_` have an automatic dependency to the `vision` dependency group; 
+at the time of writing, this only contains the `pillow` dependency.
+
+**Vision + Torch + Torchvision**: All files starting with `image_processing_` and ending with `_fast` have an automatic
+dependency to `vision`, `torch`, and `torchvision`.
+
+All of these automatic dependencies are added on top of the explicit dependencies that are detailed below.
+
+### Explicit Object Dependencies
+
+We add a method called `requires` that is used to explicitly specify the dependencies of a given object. As an
+example, the `Trainer` class has two hard dependencies: `torch` and `accelerate`. Here is how we specify these 
+required dependencies:
+
+```python
+from .utils.import_utils import requires
+
+@requires(backends=("torch", "accelerate"))
+class Trainer:
+    ...
+```
+
+Backends that can be added here are all the backends that are available in the `import_utils.py` module.
+
+## Methods
+
+[[autodoc]] utils.import_utils.define_import_structure
+
+[[autodoc]] utils.import_utils.requires

docs/source/en/main_classes/agent.md

@@ -1,167 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Agents & Tools
-
-<Tip warning={true}>
-
-Transformers Agents is an experimental API which is subject to change at any time. Results returned by the agents
-can vary as the APIs or underlying models are prone to change.
-
-</Tip>
-
-To learn more about agents and tools make sure to read the [introductory guide](../transformers_agents). This page
-contains the API docs for the underlying classes.
-
-## Agents
-
-We provide two types of agents, based on the main [`Agent`] class:
-- [`CodeAgent`] acts in one shot, generating code to solve the task, then executes it at once.
-- [`ReactAgent`] acts step by step, each step consisting of one thought, then one tool call and execution. It has two classes:
-  - [`ReactJsonAgent`] writes its tool calls in JSON.
-  - [`ReactCodeAgent`] writes its tool calls in Python code.
-
-### Agent
-
-[[autodoc]] Agent
-
-### CodeAgent
-
-[[autodoc]] CodeAgent
-
-### React agents
-
-[[autodoc]] ReactAgent
-
-[[autodoc]] ReactJsonAgent
-
-[[autodoc]] ReactCodeAgent
-
-### ManagedAgent
-
-[[autodoc]] ManagedAgent
-
-## Tools
-
-### load_tool
-
-[[autodoc]] load_tool
-
-### tool
-
-[[autodoc]] tool
-
-### Tool
-
-[[autodoc]] Tool
-
-### Toolbox
-
-[[autodoc]] Toolbox
-
-### PipelineTool
-
-[[autodoc]] PipelineTool
-
-### launch_gradio_demo
-
-[[autodoc]] launch_gradio_demo
-
-### stream_to_gradio
-
-[[autodoc]] stream_to_gradio
-
-### ToolCollection
-
-[[autodoc]] ToolCollection
-
-## Engines
-
-You're free to create and use your own engines to be usable by the Agents framework.
-These engines have the following specification:
-1. Follow the [messages format](../chat_templating.md) for its input (`List[Dict[str, str]]`) and return a string.
-2. Stop generating outputs *before* the sequences passed in the argument `stop_sequences`
-
-### TransformersEngine
-
-For convenience, we have added a `TransformersEngine` that implements the points above, taking a pre-initialized `Pipeline` as input.
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TransformersEngine
-
->>> model_name = "HuggingFaceTB/SmolLM-135M-Instruct"
->>> tokenizer = AutoTokenizer.from_pretrained(model_name)
->>> model = AutoModelForCausalLM.from_pretrained(model_name)
-
->>> pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
-
->>> engine = TransformersEngine(pipe)
->>> engine([{"role": "user", "content": "Ok!"}], stop_sequences=["great"])
-
-"What a "
-```
-
-[[autodoc]] TransformersEngine
-
-### HfApiEngine
-
-The `HfApiEngine` is an engine that wraps an [HF Inference API](https://huggingface.co/docs/api-inference/index) client for the execution of the LLM.
-
-```python
->>> from transformers import HfApiEngine
-
->>> messages = [
-...   {"role": "user", "content": "Hello, how are you?"},
-...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
-...   {"role": "user", "content": "No need to help, take it easy."},
-... ]
-
->>> HfApiEngine()(messages, stop_sequences=["conversation"])
-
-"That's very kind of you to say! It's always nice to have a relaxed "
-```
-
-[[autodoc]] HfApiEngine
-
-
-## Agent Types
-
-Agents can handle any type of object in-between tools; tools, being completely multimodal, can accept and return
-text, image, audio, video, among other types. In order to increase compatibility between tools, as well as to 
-correctly render these returns in ipython (jupyter, colab, ipython notebooks, ...), we implement wrapper classes
-around these types.
-
-The wrapped objects should continue behaving as initially; a text object should still behave as a string, an image
-object should still behave as a `PIL.Image`.
-
-These types have three specific purposes:
-
-- Calling `to_raw` on the type should return the underlying object
-- Calling `to_string` on the type should return the object as a string: that can be the string in case of an `AgentText`
-  but will be the path of the serialized version of the object in other instances
-- Displaying it in an ipython kernel should display the object correctly
-
-### AgentText
-
-[[autodoc]] transformers.agents.agent_types.AgentText
-
-### AgentImage
-
-[[autodoc]] transformers.agents.agent_types.AgentImage
-
-### AgentAudio
-
-[[autodoc]] transformers.agents.agent_types.AgentAudio

docs/source/en/model_doc/bit.md

@@ -58,6 +58,11 @@ If you're interested in submitting a resource to be included here, please feel f
 [[autodoc]] BitImageProcessor
     - preprocess
 
+## BitImageProcessorFast
+
+[[autodoc]] BitImageProcessorFast
+    - preprocess
+
 ## BitModel
 
 [[autodoc]] BitModel

docs/source/en/model_doc/blip.md

@@ -88,6 +88,11 @@ The original code can be found [here](https://github.com/salesforce/BLIP).
 [[autodoc]] BlipTextModel
     - forward
 
+## BlipTextLMHeadModel
+
+[[autodoc]] BlipTextLMHeadModel
+- forward
+
 ## BlipVisionModel
 
 [[autodoc]] BlipVisionModel
@@ -123,6 +128,11 @@ The original code can be found [here](https://github.com/salesforce/BLIP).
 [[autodoc]] TFBlipTextModel
     - call
 
+## TFBlipTextLMHeadModel
+
+[[autodoc]] TFBlipTextLMHeadModel
+- forward
+
 ## TFBlipVisionModel
 
 [[autodoc]] TFBlipVisionModel

docs/source/en/model_doc/bridgetower.md

@@ -147,6 +147,11 @@ Tips:
 [[autodoc]] BridgeTowerImageProcessor
     - preprocess
 
+## BridgeTowerImageProcessorFast
+
+[[autodoc]] BridgeTowerImageProcessorFast
+    - preprocess
+
 ## BridgeTowerProcessor
 
 [[autodoc]] BridgeTowerProcessor

docs/source/en/model_doc/chinese_clip.md

@@ -90,6 +90,11 @@ Currently, following scales of pretrained Chinese-CLIP models are available on
 [[autodoc]] ChineseCLIPImageProcessor
     - preprocess
 
+## ChineseCLIPImageProcessorFast
+
+[[autodoc]] ChineseCLIPImageProcessorFast
+    - preprocess
+
 ## ChineseCLIPFeatureExtractor
 
 [[autodoc]] ChineseCLIPFeatureExtractor

docs/source/en/model_doc/colpali.md

@@ -1,5 +1,4 @@
 <!--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
 
@@ -9,76 +8,134 @@ Unless required by applicable law or agreed to in writing, software distributed
 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
+⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
 rendered properly in your Markdown viewer.
-
 -->
 
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
+</div>
+
 # ColPali
 
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-</div>
+[ColPali](https://huggingface.co/papers/2407.01449) is a model designed to retrieve documents by analyzing their visual features. Unlike traditional systems that rely heavily on text extraction and OCR, ColPali treats each page as an image. It uses [Paligemma-3B](./paligemma) to capture not only text, but also the layout, tables, charts, and other visual elements to create detailed embeddings. This offers a more comprehensive understanding of documents and enables more efficient and accurate retrieval.
 
-## Overview
+You can find all the original ColPali checkpoints under the [ColPali](https://huggingface.co/collections/vidore/hf-native-colvision-models-6755d68fc60a8553acaa96f7) collection.
 
-The *ColPali* model was proposed in [ColPali: Efficient Document Retrieval with Vision Language Models](https://doi.org/10.48550/arXiv.2407.01449) by **Manuel Faysse***, **Hugues Sibille***, **Tony Wu***, Bilel Omrani, Gautier Viaud, Céline Hudelot, Pierre Colombo (* denotes equal contribution). Work lead by ILLUIN Technology.
+> [!TIP]
+> Click on the ColPali models in the right sidebar for more examples of how to use ColPali for image retrieval.
 
-In our proposed *ColPali* approach, we leverage VLMs to construct efficient multi-vector embeddings directly from document images (“screenshots”) for document retrieval. We train the model to maximize the similarity between these document embeddings and the corresponding query embeddings, using the late interaction method introduced in ColBERT.
+<hfoptions id="usage">
+<hfoption id="image retrieval">
 
-Using *ColPali* removes the need for potentially complex and brittle layout recognition and OCR pipelines with a single model that can take into account both the textual and visual content (layout, charts, etc.) of a document.
-
-## Resources
-
-- The *ColPali* arXiv paper can be found [here](https://doi.org/10.48550/arXiv.2407.01449). 📄
-- The official blog post detailing ColPali can be found [here](https://huggingface.co/blog/manu/colpali). 📝
-- The original model implementation code for the ColPali model and for the `colpali-engine` package can be found [here](https://github.com/illuin-tech/colpali). 🌎
-- Cookbooks for learning to use the transformers-native version of *ColPali*, fine-tuning, and similarity maps generation can be found [here](https://github.com/tonywu71/colpali-cookbooks). 📚
-
-This model was contributed by [@tonywu71](https://huggingface.co/tonywu71) and [@yonigozlan](https://huggingface.co/yonigozlan).
-
-## Usage
-
-This example demonstrates how to use *ColPali* to embed both queries and images, calculate their similarity scores, and identify the most relevant matches. For a specific query, you can retrieve the top-k most similar images by selecting the ones with the highest similarity scores.
-
-```python
+```py
+import requests
 import torch
 from PIL import Image
-
 from transformers import ColPaliForRetrieval, ColPaliProcessor
 
-model_name = "vidore/colpali-v1.2-hf"
-
+# Load model (bfloat16 support is limited; fallback to float32 if needed)
 model = ColPaliForRetrieval.from_pretrained(
-    model_name,
-    torch_dtype=torch.bfloat16,
-    device_map="cuda:0",  # or "mps" if on Apple Silicon
+    "vidore/colpali-v1.2-hf",
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto",  # "cpu", "cuda", or "mps" for Apple Silicon
 ).eval()
 
 processor = ColPaliProcessor.from_pretrained(model_name)
 
-# Your inputs (replace dummy images with screenshots of your documents)
+url1 = "https://upload.wikimedia.org/wikipedia/commons/8/89/US-original-Declaration-1776.jpg"
+url2 = "https://upload.wikimedia.org/wikipedia/commons/thumb/4/4c/Romeoandjuliet1597.jpg/500px-Romeoandjuliet1597.jpg"
+
 images = [
-    Image.new("RGB", (32, 32), color="white"),
-    Image.new("RGB", (16, 16), color="black"),
+    Image.open(requests.get(url1, stream=True).raw),
+    Image.open(requests.get(url2, stream=True).raw),
 ]
+
 queries = [
-    "What is the organizational structure for our R&D department?",
-    "Can you provide a breakdown of last year’s financial performance?",
+    "Who printed the edition of Romeo and Juliet?",
+    "When was the United States Declaration of Independence proclaimed?",
 ]
 
 # Process the inputs
-batch_images = processor(images=images).to(model.device)
-batch_queries = processor(text=queries).to(model.device)
+inputs_images = processor(images=images, return_tensors="pt").to(model.device)
+inputs_text = processor(text=queries, return_tensors="pt").to(model.device)
 
 # Forward pass
 with torch.no_grad():
-    image_embeddings = model(**batch_images).embeddings
-    query_embeddings = model(**batch_queries).embeddings
+    image_embeddings = model(**inputs_images).embeddings
+    query_embeddings = model(**inputs_text).embeddings
 
-# Score the queries against the images
 scores = processor.score_retrieval(query_embeddings, image_embeddings)
+
+print("Retrieval scores (query x image):")
+print(scores)
 ```
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [bitsandbytes](../quantization/bitsandbytes.md) to quantize the weights to int4.
+
+```py
+import requests
+import torch
+from PIL import Image
+from transformers import ColPaliForRetrieval, ColPaliProcessor
+from transformers import BitsAndBytesConfig
+
+# 4-bit quantization configuration
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.float16,
+)
+
+model_name = "vidore/colpali-v1.2-hf"
+
+# Load model 
+model = ColPaliForRetrieval.from_pretrained(
+    model_name,
+    quantization_config=bnb_config,
+    device_map="cuda"
+).eval()
+
+processor = ColPaliProcessor.from_pretrained(model_name)
+
+url1 = "https://upload.wikimedia.org/wikipedia/commons/8/89/US-original-Declaration-1776.jpg"
+url2 = "https://upload.wikimedia.org/wikipedia/commons/thumb/4/4c/Romeoandjuliet1597.jpg/500px-Romeoandjuliet1597.jpg"
+
+images = [
+    Image.open(requests.get(url1, stream=True).raw),
+    Image.open(requests.get(url2, stream=True).raw),
+]
+
+queries = [
+    "Who printed the edition of Romeo and Juliet?",
+    "When was the United States Declaration of Independence proclaimed?",
+]
+
+# Process the inputs
+inputs_images = processor(images=images, return_tensors="pt").to(model.device)
+inputs_text = processor(text=queries, return_tensors="pt").to(model.device)
+
+# Forward pass
+with torch.no_grad():
+    image_embeddings = model(**inputs_images).embeddings
+    query_embeddings = model(**inputs_text).embeddings
+
+scores = processor.score_retrieval(query_embeddings, image_embeddings)
+
+print("Retrieval scores (query x image):")
+print(scores)
+```
+
+## Notes
+
+- [`~ColPaliProcessor.score_retrieval`] returns a 2D tensor where the first dimension is the number of queries and the second dimension is the number of images. A higher score indicates more similarity between the query and image.
 
 ## ColPaliConfig
 

docs/source/en/model_doc/conditional_detr.md

@@ -48,6 +48,11 @@ This model was contributed by [DepuMeng](https://huggingface.co/DepuMeng). The o
 
 [[autodoc]] ConditionalDetrImageProcessor
     - preprocess
+
+## ConditionalDetrImageProcessorFast
+
+[[autodoc]] ConditionalDetrImageProcessorFast
+    - preprocess
     - post_process_object_detection
     - post_process_instance_segmentation
     - post_process_semantic_segmentation

docs/source/en/model_doc/dinov2.md

@@ -10,71 +10,134 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# DINOv2
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
-">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
 
-The DINOv2 model was proposed in [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by
-Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
-DINOv2 is an upgrade of [DINO](https://arxiv.org/abs/2104.14294), a self-supervised method applied on [Vision Transformers](vit). This method enables all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning.
+# DINOv2
 
-The abstract from the paper is the following:
+[DINOv2](https://huggingface.co/papers/2304.07193) is a vision foundation model that uses [ViT](./vit) as a feature extractor for multiple downstream tasks like image classification and depth estimation. It focuses on stabilizing and accelerating training through techniques like a faster memory-efficient attention, sequence packing, improved stochastic depth, Fully Sharded Data Parallel (FSDP), and model distillation.
 
-*The recent breakthroughs in natural language processing for model pretraining on large quantities of data have opened the way for similar foundation models in computer vision. These models could greatly simplify the use of images in any system by producing all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning. This work shows that existing pretraining methods, especially self-supervised methods, can produce such features if trained on enough curated data from diverse sources. We revisit existing approaches and combine different techniques to scale our pretraining in terms of data and model size. Most of the technical contributions aim at accelerating and stabilizing the training at scale. In terms of data, we propose an automatic pipeline to build a dedicated, diverse, and curated image dataset instead of uncurated data, as typically done in the self-supervised literature. In terms of models, we train a ViT model (Dosovitskiy et al., 2020) with 1B parameters and distill it into a series of smaller models that surpass the best available all-purpose features, OpenCLIP (Ilharco et al., 2021) on most of the benchmarks at image and pixel levels.*
+You can find all the original DINOv2 checkpoints under the [Dinov2](https://huggingface.co/collections/facebook/dinov2-6526c98554b3d2576e071ce3) collection.
 
-This model was contributed by [nielsr](https://huggingface.co/nielsr).
-The original code can be found [here](https://github.com/facebookresearch/dinov2).
+> [!TIP]
+> Click on the DINOv2 models in the right sidebar for more examples of how to apply DINOv2 to different vision tasks.
 
-## Usage tips
+The example below demonstrates how to obtain an image embedding with [`Pipeline`] or the [`AutoModel`] class.
 
-The model can be traced using `torch.jit.trace` which leverages JIT compilation to optimize the model making it faster to run. Note this still produces some mis-matched elements and the difference between the original model and the traced model is of the order of 1e-4.
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-```python
+```py
 import torch
-from transformers import AutoImageProcessor, AutoModel
-from PIL import Image
+from transformers import pipeline
+
+pipe = pipeline(
+    task="image-classification", 
+    model="facebook/dinov2-small-imagenet1k-1-layer",
+    torch_dtype=torch.float16,
+    device=0
+)
+
+pipe("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg")
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+
+```py
 import requests
+from transformers import AutoImageProcessor, AutoModelForImageClassification
+from PIL import Image
+
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+
+processor = AutoImageProcessor.from_pretrained("facebook/dinov2-small-imagenet1k-1-layer")
+model = AutoModelForImageClassification.from_pretrained(
+    "facebook/dinov2-small-imagenet1k-1-layer", 
+    torch_dtype=torch.float16, 
+    device_map="auto", 
+    attn_implementation="sdpa"
+)
+
+inputs = processor(images=image, return_tensors="pt")
+logits = model(**inputs).logits
+predicted_class_idx = logits.argmax(-1).item()
+print("Predicted class:", model.config.id2label[predicted_class_idx])
+```
+
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [torchao](../quantization/torchao) to only quantize the weights to int4.
+
+```py
+# pip install torchao
+import requests
+from transformers import TorchAoConfig, AutoImageProcessor, AutoModelForImageClassification
+from torchao.quantization import Int4WeightOnlyConfig
+from PIL import Image
 
 url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
 image = Image.open(requests.get(url, stream=True).raw)
 
-processor = AutoImageProcessor.from_pretrained('facebook/dinov2-base')
-model = AutoModel.from_pretrained('facebook/dinov2-base')
+processor = AutoImageProcessor.from_pretrained('facebook/dinov2-giant-imagenet1k-1-layer')
+
+quant_config = Int4WeightOnlyConfig(group_size=128)
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+model = AutoModelForImageClassification.from_pretrained(
+    'facebook/dinov2-giant-imagenet1k-1-layer',
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    quantization_config=quantization_config
+)
 
 inputs = processor(images=image, return_tensors="pt")
 outputs = model(**inputs)
-last_hidden_states = outputs[0]
-
-# We have to force return_dict=False for tracing
-model.config.return_dict = False
-
-with torch.no_grad():
-    traced_model = torch.jit.trace(model, [inputs.pixel_values])
-    traced_outputs = traced_model(inputs.pixel_values)
-
-print((last_hidden_states - traced_outputs[0]).abs().max())
+logits = outputs.logits
+predicted_class_idx = logits.argmax(-1).item()
+print("Predicted class:", model.config.id2label[predicted_class_idx])
 ```
 
-## Resources
+## Notes
 
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with DINOv2.
+- Use [torch.jit.trace](https://pytorch.org/docs/stable/generated/torch.jit.trace.html) to speedup inference. However, it will produce some mismatched elements. The difference between the original and traced model is 1e-4.
 
-- Demo notebooks for DINOv2 can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/DINOv2). 🌎
+    ```py
+    import torch
+    from transformers import AutoImageProcessor, AutoModel
+    from PIL import Image
+    import requests
 
-<PipelineTag pipeline="image-classification"/>
+    url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
+    image = Image.open(requests.get(url, stream=True).raw)
 
-- [`Dinov2ForImageClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb).
-- See also: [Image classification task guide](../tasks/image_classification)
+    processor = AutoImageProcessor.from_pretrained('facebook/dinov2-base')
+    model = AutoModel.from_pretrained('facebook/dinov2-base')
 
-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.
+    inputs = processor(images=image, return_tensors="pt")
+    outputs = model(**inputs)
+    last_hidden_states = outputs[0]
+
+    # We have to force return_dict=False for tracing
+    model.config.return_dict = False
+
+    with torch.no_grad():
+        traced_model = torch.jit.trace(model, [inputs.pixel_values])
+        traced_outputs = traced_model(inputs.pixel_values)
+
+    print((last_hidden_states - traced_outputs[0]).abs().max())
+    ```
 
 ## Dinov2Config
 

docs/source/en/model_doc/donut.md

@@ -13,180 +13,191 @@ rendered properly in your Markdown viewer.
 
 specific language governing permissions and limitations under the License. -->
 
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
+</div>
+
 # Donut
 
-## Overview
+[Donut (Document Understanding Transformer)](https://huggingface.co/papers2111.15664) is a visual document understanding model that doesn't require an Optical Character Recognition (OCR) engine. Unlike traditional approaches that extract text using OCR before processing, Donut employs an end-to-end Transformer-based architecture to directly analyze document images. This eliminates OCR-related inefficiencies making it more accurate and adaptable to diverse languages and formats. 
 
-The Donut model was proposed in [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by
-Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-Donut consists of an image Transformer encoder and an autoregressive text Transformer decoder to perform document understanding
-tasks such as document image classification, form understanding and visual question answering.
+Donut features vision encoder ([Swin](./swin)) and a text decoder ([BART](./bart)). Swin converts document images into embeddings and BART processes them into meaningful text sequences.
 
-The abstract from the paper is the following:
+You can find all the original Donut checkpoints under the [Naver Clova Information Extraction](https://huggingface.co/naver-clova-ix) organization.
 
-*Understanding document images (e.g., invoices) is a core but challenging task since it requires complex functions such as reading text and a holistic understanding of the document. Current Visual Document Understanding (VDU) methods outsource the task of reading text to off-the-shelf Optical Character Recognition (OCR) engines and focus on the understanding task with the OCR outputs. Although such OCR-based approaches have shown promising performance, they suffer from 1) high computational costs for using OCR; 2) inflexibility of OCR models on languages or types of document; 3) OCR error propagation to the subsequent process. To address these issues, in this paper, we introduce a novel OCR-free VDU model named Donut, which stands for Document understanding transformer. As the first step in OCR-free VDU research, we propose a simple architecture (i.e., Transformer) with a pre-training objective (i.e., cross-entropy loss). Donut is conceptually simple yet effective. Through extensive experiments and analyses, we show a simple OCR-free VDU model, Donut, achieves state-of-the-art performances on various VDU tasks in terms of both speed and accuracy. In addition, we offer a synthetic data generator that helps the model pre-training to be flexible in various languages and domains.*
+> [!TIP]
+> Click on the Donut models in the right sidebar for more examples of how to apply Donut to different language and vision tasks.
 
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/donut_architecture.jpg"
-alt="drawing" width="600"/>
+The examples below demonstrate how to perform document understanding tasks using Donut with [`Pipeline`] and [`AutoModel`]
 
-<small> Donut high-level overview. Taken from the <a href="https://arxiv.org/abs/2111.15664">original paper</a>. </small>
-
-This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code can be found
-[here](https://github.com/clovaai/donut).
-
-## Usage tips
-
-- The quickest way to get started with Donut is by checking the [tutorial
-  notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/Donut), which show how to use the model
-  at inference time as well as fine-tuning on custom data.
-- Donut is always used within the [VisionEncoderDecoder](vision-encoder-decoder) framework.
-
-## Inference examples
-
-Donut's [`VisionEncoderDecoder`] model accepts images as input and makes use of
-[`~generation.GenerationMixin.generate`] to autoregressively generate text given the input image.
-
-The [`DonutImageProcessor`] class is responsible for preprocessing the input image and
-[`XLMRobertaTokenizer`/`XLMRobertaTokenizerFast`] decodes the generated target tokens to the target string. The
-[`DonutProcessor`] wraps [`DonutImageProcessor`] and [`XLMRobertaTokenizer`/`XLMRobertaTokenizerFast`]
-into a single instance to both extract the input features and decode the predicted token ids.
-
-- Step-by-step Document Image Classification
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
 ```py
->>> import re
+# pip install datasets
+import torch
+from transformers import pipeline
+from PIL import Image
 
->>> from transformers import DonutProcessor, VisionEncoderDecoderModel
->>> from datasets import load_dataset
->>> import torch
+pipeline = pipeline(
+    task="document-question-answering",
+    model="naver-clova-ix/donut-base-finetuned-docvqa",
+    device=0,
+    torch_dtype=torch.float16
+)
+dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+image = dataset[0]["image"]
 
->>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
->>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
-
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
->>> model.to(device)  # doctest: +IGNORE_RESULT
-
->>> # load document image
->>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
->>> image = dataset[1]["image"]
-
->>> # prepare decoder inputs
->>> task_prompt = "<s_rvlcdip>"
->>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
-
->>> pixel_values = processor(image, return_tensors="pt").pixel_values
-
->>> outputs = model.generate(
-...     pixel_values.to(device),
-...     decoder_input_ids=decoder_input_ids.to(device),
-...     max_length=model.decoder.config.max_position_embeddings,
-...     pad_token_id=processor.tokenizer.pad_token_id,
-...     eos_token_id=processor.tokenizer.eos_token_id,
-...     use_cache=True,
-...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
-...     return_dict_in_generate=True,
-... )
-
->>> sequence = processor.batch_decode(outputs.sequences)[0]
->>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
->>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
->>> print(processor.token2json(sequence))
-{'class': 'advertisement'}
+pipeline(image=image, question="What time is the coffee break?")
 ```
 
-- Step-by-step Document Parsing
+</hfoption>
+<hfoption id="AutoModel">
 
 ```py
->>> import re
+# pip install datasets
+import torch
+from datasets import load_dataset
+from transformers import AutoProcessor, AutoModelForVision2Seq
 
->>> from transformers import DonutProcessor, VisionEncoderDecoderModel
->>> from datasets import load_dataset
->>> import torch
+processor = AutoProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
+model = AutoModelForVision2Seq.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
 
->>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
->>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
+dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+image = dataset[0]["image"]
+question = "What time is the coffee break?"
+task_prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
+inputs = processor(image, task_prompt, return_tensors="pt")
 
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
->>> model.to(device)  # doctest: +IGNORE_RESULT
-
->>> # load document image
->>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
->>> image = dataset[2]["image"]
-
->>> # prepare decoder inputs
->>> task_prompt = "<s_cord-v2>"
->>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
-
->>> pixel_values = processor(image, return_tensors="pt").pixel_values
-
->>> outputs = model.generate(
-...     pixel_values.to(device),
-...     decoder_input_ids=decoder_input_ids.to(device),
-...     max_length=model.decoder.config.max_position_embeddings,
-...     pad_token_id=processor.tokenizer.pad_token_id,
-...     eos_token_id=processor.tokenizer.eos_token_id,
-...     use_cache=True,
-...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
-...     return_dict_in_generate=True,
-... )
-
->>> sequence = processor.batch_decode(outputs.sequences)[0]
->>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
->>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
->>> print(processor.token2json(sequence))
-{'menu': {'nm': 'CINNAMON SUGAR', 'unitprice': '17,000', 'cnt': '1 x', 'price': '17,000'}, 'sub_total': {'subtotal_price': '17,000'}, 'total': {'total_price': '17,000', 'cashprice': '20,000', 'changeprice': '3,000'}}
+outputs = model.generate(
+    input_ids=inputs.input_ids,
+    pixel_values=inputs.pixel_values,
+    max_length=512
+)
+answer = processor.decode(outputs[0], skip_special_tokens=True)
+print(answer)
 ```
 
-- Step-by-step Document Visual Question Answering (DocVQA)
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [torchao](../quantization/torchao) to only quantize the weights to int4.
 
 ```py
->>> import re
+# pip install datasets torchao
+import torch
+from datasets import load_dataset
+from transformers import TorchAoConfig, AutoProcessor, AutoModelForVision2Seq
 
->>> from transformers import DonutProcessor, VisionEncoderDecoderModel
->>> from datasets import load_dataset
->>> import torch
+quantization_config = TorchAoConfig("int4_weight_only", group_size=128)
+processor = AutoProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
+model = AutoModelForVision2Seq.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa", quantization_config=quantization_config)
 
->>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
->>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
+dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+image = dataset[0]["image"]
+question = "What time is the coffee break?"
+task_prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
+inputs = processor(image, task_prompt, return_tensors="pt")
 
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
->>> model.to(device)  # doctest: +IGNORE_RESULT
-
->>> # load document image from the DocVQA dataset
->>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
->>> image = dataset[0]["image"]
-
->>> # prepare decoder inputs
->>> task_prompt = "<s_docvqa><s_question>{user_input}</s_question><s_answer>"
->>> question = "When is the coffee break?"
->>> prompt = task_prompt.replace("{user_input}", question)
->>> decoder_input_ids = processor.tokenizer(prompt, add_special_tokens=False, return_tensors="pt").input_ids
-
->>> pixel_values = processor(image, return_tensors="pt").pixel_values
-
->>> outputs = model.generate(
-...     pixel_values.to(device),
-...     decoder_input_ids=decoder_input_ids.to(device),
-...     max_length=model.decoder.config.max_position_embeddings,
-...     pad_token_id=processor.tokenizer.pad_token_id,
-...     eos_token_id=processor.tokenizer.eos_token_id,
-...     use_cache=True,
-...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
-...     return_dict_in_generate=True,
-... )
-
->>> sequence = processor.batch_decode(outputs.sequences)[0]
->>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
->>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
->>> print(processor.token2json(sequence))
-{'question': 'When is the coffee break?', 'answer': '11-14 to 11:39 a.m.'}
+outputs = model.generate(
+    input_ids=inputs.input_ids,
+    pixel_values=inputs.pixel_values,
+    max_length=512
+)
+answer = processor.decode(outputs[0], skip_special_tokens=True)
+print(answer)
 ```
 
-See the [model hub](https://huggingface.co/models?filter=donut) to look for Donut checkpoints.
+## Notes
 
-## Training
+- Use Donut for document image classification as shown below.
 
-We refer to the [tutorial notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/Donut).
+    ```py
+    >>> import re
+    >>> from transformers import DonutProcessor, VisionEncoderDecoderModel
+    >>> from datasets import load_dataset
+    >>> import torch
+
+    >>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
+    >>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
+
+    >>> device = "cuda" if torch.cuda.is_available() else "cpu"
+    >>> model.to(device)  # doctest: +IGNORE_RESULT
+
+    >>> # load document image
+    >>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+    >>> image = dataset[1]["image"]
+
+    >>> # prepare decoder inputs
+    >>> task_prompt = "<s_rvlcdip>"
+    >>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
+
+    >>> pixel_values = processor(image, return_tensors="pt").pixel_values
+
+    >>> outputs = model.generate(
+    ...     pixel_values.to(device),
+    ...     decoder_input_ids=decoder_input_ids.to(device),
+    ...     max_length=model.decoder.config.max_position_embeddings,
+    ...     pad_token_id=processor.tokenizer.pad_token_id,
+    ...     eos_token_id=processor.tokenizer.eos_token_id,
+    ...     use_cache=True,
+    ...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
+    ...     return_dict_in_generate=True,
+    ... )
+
+    >>> sequence = processor.batch_decode(outputs.sequences)[0]
+    >>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
+    >>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
+    >>> print(processor.token2json(sequence))
+    {'class': 'advertisement'}
+    ```
+
+- Use Donut for document parsing as shown below.
+
+    ```py
+    >>> import re
+    >>> from transformers import DonutProcessor, VisionEncoderDecoderModel
+    >>> from datasets import load_dataset
+    >>> import torch
+
+    >>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
+    >>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
+
+    >>> device = "cuda" if torch.cuda.is_available() else "cpu"
+    >>> model.to(device)  # doctest: +IGNORE_RESULT
+
+    >>> # load document image
+    >>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+    >>> image = dataset[2]["image"]
+
+    >>> # prepare decoder inputs
+    >>> task_prompt = "<s_cord-v2>"
+    >>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
+
+    >>> pixel_values = processor(image, return_tensors="pt").pixel_values
+
+    >>> outputs = model.generate(
+    ...     pixel_values.to(device),
+    ...     decoder_input_ids=decoder_input_ids.to(device),
+    ...     max_length=model.decoder.config.max_position_embeddings,
+    ...     pad_token_id=processor.tokenizer.pad_token_id,
+    ...     eos_token_id=processor.tokenizer.eos_token_id,
+    ...     use_cache=True,
+    ...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
+    ...     return_dict_in_generate=True,
+    ... )
+
+    >>> sequence = processor.batch_decode(outputs.sequences)[0]
+    >>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
+    >>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
+    >>> print(processor.token2json(sequence))
+    {'menu': {'nm': 'CINNAMON SUGAR', 'unitprice': '17,000', 'cnt': '1 x', 'price': '17,000'}, 'sub_total': {'subtotal_price': '17,000'}, 'total': 
+    {'total_price': '17,000', 'cashprice': '20,000', 'changeprice': '3,000'}}
+    ```
 
 ## DonutSwinConfig
 
@@ -197,6 +208,11 @@ We refer to the [tutorial notebooks](https://github.com/NielsRogge/Transformers-
 [[autodoc]] DonutImageProcessor
     - preprocess
 
+## DonutImageProcessorFast
+
+[[autodoc]] DonutImageProcessorFast
+    - preprocess
+
 ## DonutFeatureExtractor
 
 [[autodoc]] DonutFeatureExtractor
@@ -215,3 +231,8 @@ We refer to the [tutorial notebooks](https://github.com/NielsRogge/Transformers-
 
 [[autodoc]] DonutSwinModel
     - forward
+
+## DonutSwinForImageClassification
+
+[[autodoc]] transformers.DonutSwinForImageClassification
+    - forward

docs/source/en/model_doc/efficientnet.md

@@ -43,6 +43,11 @@ The original code can be found [here](https://github.com/tensorflow/tpu/tree/mas
 [[autodoc]] EfficientNetImageProcessor
     - preprocess
 
+## EfficientNetImageProcessorFast
+
+[[autodoc]] EfficientNetImageProcessorFast
+    - preprocess
+
 ## EfficientNetModel
 
 [[autodoc]] EfficientNetModel

docs/source/en/model_doc/falcon_mamba.md

@@ -14,95 +14,100 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# FalconMamba
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
+# FalconMamba
 
-The FalconMamba model was proposed by TII UAE (Technology Innovation Institute) in their release.
+[FalconMamba](https://huggingface.co/papers/2410.05355) is a 7B large language model, available as pretrained and instruction-tuned variants, based on the [Mamba](./mamba). This model implements a pure Mamba design that focuses on computational efficiency while maintaining strong performance. FalconMamba is significantly faster at inference and requires substantially less memory for long sequence generation. The models are pretrained on a diverse 5.8T token dataset including [RefinedWeb](https://huggingface.co/datasets/tiiuae/falcon-refinedweb), technical content, code, and mathematical data.
 
-The abstract from the paper is the following:
+You can find the official FalconMamba checkpoints in the [FalconMamba 7B](https://huggingface.co/collections/tiiuae/falconmamba-7b-66b9a580324dd1598b0f6d4a) collection.
 
-*We present FalconMamba, a new base large language model based on the novel Mamba architecture. FalconMamba is trained on 5.8 trillion tokens with carefully selected data mixtures. As a pure Mamba-based model, FalconMamba surpasses leading open-weight models based on Transformers, such as Mistral 7B, Llama3 8B, and Falcon2 11B. It is on par with Gemma 7B and outperforms models with different architecture designs, such as RecurrentGemma 9B. Currently, FalconMamba is the best-performing Mamba model in the literature at this scale, surpassing both existing Mamba and hybrid Mamba-Transformer models.
-Due to its architecture, FalconMamba is significantly faster at inference and requires substantially less memory for long sequence generation. Despite recent studies suggesting that hybrid Mamba-Transformer models outperform pure architecture designs, we argue and demonstrate that the pure Mamba design can achieve similar, even superior results compared to the hybrid design. We make the weights of our implementation of FalconMamba publicly available under a permissive license.*
+> [!TIP]
+> Click on the FalconMamba models in the right sidebar for more examples of how to apply FalconMamba to different language tasks.
 
-Tips:
+The examples below demonstrate how to generate text with [`Pipeline`], [`AutoModel`], and from the command line.
 
-- FalconMamba is mostly based on Mamba architecture, the same [tips and best practices](./mamba) would be relevant here.
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-The model has been trained on approximtely 6T tokens consisting a mixture of many data sources such as RefineWeb, Cosmopedia and Math data.
-
-For more details about the training procedure and the architecture, have a look at [the technical paper of FalconMamba]() (coming soon).
-
-# Usage
-
-Below we demonstrate how to use the model:
-
-```python 
-from transformers import FalconMambaForCausalLM, AutoTokenizer
+```py
 import torch
+from transformers import pipeline
 
-tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-mamba-7b")
-model = FalconMambaForCausalLM.from_pretrained("tiiuae/falcon-mamba-7b")
-
-input_ids = tokenizer("Hey how are you doing?", return_tensors= "pt")["input_ids"]
-
-out = model.generate(input_ids, max_new_tokens=10)
-print(tokenizer.batch_decode(out))
+pipeline = pipeline(
+    "text-generation", 
+    model="tiiuae/falcon-mamba-7b-instruct",
+    torch_dtype=torch.bfloat16,
+    device=0
+)
+pipeline(
+    "Explain the difference between transformers and SSMs",
+    max_length=100,
+    do_sample=True,
+    temperature=0.7
+)
 ```
 
-The architecture is also compatible with `torch.compile` for faster generation:
+</hfoption>
+<hfoption id="AutoModel">
 
-```python 
-from transformers import FalconMambaForCausalLM, AutoTokenizer
-import torch
-
-tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-mamba-7b")
-model = FalconMambaForCausalLM.from_pretrained("tiiuae/falcon-mamba-7b", torch_dtype=torch.bfloat16).to(0)
-model = torch.compile(model)
-
-input_ids = tokenizer("Hey how are you doing?", return_tensors= "pt")["input_ids"]
-
-out = model.generate(input_ids, max_new_tokens=10)
-print(tokenizer.batch_decode(out))
-```
-
-If you have access to a GPU that is compatible with `bitsandbytes`, you can also quantize the model in 4-bit precision:
-
-```python 
-from transformers import FalconMambaForCausalLM, AutoTokenizer, BitsAndBytesConfig
-import torch
-
-tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-mamba-7b")
-quantization_config = BitsAndBytesConfig(load_in_4bit=True)
-model = FalconMambaForCausalLM.from_pretrained("tiiuae/falcon-mamba-7b", quantization_config=quantization_config)
-
-input_ids = tokenizer("Hey how are you doing?", return_tensors= "pt")["input_ids"]
-
-out = model.generate(input_ids, max_new_tokens=10)
-print(tokenizer.batch_decode(out))
-```
-
-You can also play with the instruction fine-tuned model:
-
-```python 
-from transformers import FalconMambaForCausalLM, AutoTokenizer
+```py
 import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
 tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-mamba-7b-instruct")
-model = FalconMambaForCausalLM.from_pretrained("tiiuae/falcon-mamba-7b-instruct")
+model = AutoModelForCausalLM.from_pretrained(
+    "tiiuae/falcon-mamba-7b-instruct",
+    torch_dtype=torch.bfloat16,
+    device_map="auto"
+)
 
-# We use the tokenizer's chat template to format each message - see https://huggingface.co/docs/transformers/main/en/chat_templating
-messages = [
-    {"role": "user", "content": "How many helicopters can a human eat in one sitting?"},
-]
-input_ids = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True).input_ids
+input_ids = tokenizer("Explain the difference between transformers and SSMs", return_tensors="pt").to("cuda")
 
-outputs = model.generate(input_ids)
-print(tokenizer.decode(outputs[0]))
+output = model.generate(**input_ids, max_new_tokens=100, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+
+</hfoption>
+<hfoption id="transformers-cli">
+
+```bash
+transformers-cli chat --model_name_or_path tiiuae/falcon-mamba-7b-instruct --torch_dtype auto --device 0
+```
+
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to quantize the weights to 4-bits.
+
+```python
+import torch
+from transformers import AutoTokenizer, FalconMambaForCausalLM, BitsAndBytesConfig
+
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_use_double_quant=True,
+)
+
+tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-mamba-7b")
+model = FalconMambaForCausalLM.from_pretrained(
+    "tiiuae/falcon-mamba-7b",
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    quantization_config=quantization_config,
+)
+
+inputs = tokenizer("Explain the concept of state space models in simple terms", return_tensors="pt").to("cuda")
+outputs = model.generate(**inputs, max_new_tokens=100)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 
 ## FalconMambaConfig

docs/source/en/model_doc/flava.md

@@ -72,6 +72,11 @@ This model was contributed by [aps](https://huggingface.co/aps). The original co
 [[autodoc]] FlavaImageProcessor
     - preprocess
 
+## FlavaImageProcessorFast
+
+[[autodoc]] FlavaImageProcessorFast
+    - preprocess
+
 ## FlavaForPreTraining
 
 [[autodoc]] FlavaForPreTraining

docs/source/en/model_doc/gemma2.md

@@ -14,36 +14,133 @@ specific language governing permissions and limitations under the License.
 rendered properly in your Markdown viewer.
 
 -->
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
+        ">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
+</div>
 
 # Gemma2
 
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+[Gemma 2](https://huggingface.co/papers/2408.00118) is a family of language models with pretrained and instruction-tuned variants, available in 2B, 9B, 27B parameters. The architecture is similar to the previous Gemma, except it features interleaved local attention (4096 tokens) and global attention (8192 tokens) and grouped-query attention (GQA) to increase inference performance.
+
+The 2B and 9B models are trained with knowledge distillation, and the instruction-tuned variant was post-trained with supervised fine-tuning and reinforcement learning.
+
+You can find all the original Gemma 2 checkpoints under the [Gemma 2](https://huggingface.co/collections/google/gemma-2-release-667d6600fd5220e7b967f315) collection.
+
+> [!TIP]
+> Click on the Gemma 2 models in the right sidebar for more examples of how to apply Gemma to different language tasks.
+
+The example below demonstrates how to chat with the model with [`Pipeline`] or the [`AutoModel`] class, and from the command line.
+
+<hfoptions id="usage">
+<hfoption id="Pipeline">
+
+
+```python
+import torch
+from transformers import pipeline
+
+pipe = pipeline(
+    task="text-generation",
+    model="google/gemma-2-9b",
+    torch_dtype=torch.bfloat16,
+    device="cuda",
+)
+
+pipe("Explain quantum computing simply. ", max_new_tokens=50)
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+    
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-9b")
+model = AutoModelForCausalLM.from_pretrained(
+    "google/gemma-2-9b",
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    attn_implementation="sdpa"
+)
+
+input_text = "Explain quantum computing simply."
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+outputs = model.generate(**input_ids, max_new_tokens=32, cache_implementation="static")
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+
+```
+
+</hfoption>
+<hfoption id="transformers-cli">
+
+```
+echo -e "Explain quantum computing simply." | transformers-cli run --task text-generation --model google/gemma-2-2b --device 0
+```
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+	
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to only quantize the weights to int4.
+
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+
+quantization_config = BitsAndBytesConfig(load_in_4bit=True)
+tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-27b")
+model = AutoModelForCausalLM.from_pretrained(
+    "google/gemma-2-27b",
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    attn_implementation="sdpa"
+)
+
+input_text = "Explain quantum computing simply."
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+outputs = model.generate(**input_ids, max_new_tokens=32, cache_implementation="static")
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```
+
+Use the [AttentionMaskVisualizer](https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139) to better understand what tokens the model can and cannot attend to.
+
+
+```python
+from transformers.utils.attention_visualizer import AttentionMaskVisualizer
+visualizer = AttentionMaskVisualizer("google/gemma-2b")
+visualizer("You are an assistant. Make sure you print me") 
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/gemma-2-attn-mask.png"/>
 </div>
 
-## Overview
+## Notes
 
-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).
+- Use a [`HybridCache`] instance to enable caching in Gemma 2. Gemma 2 doesn't support kv-caching strategies like [`DynamicCache`] or tuples of tensors because it uses sliding window attention every second layer.
 
-The abstract from the blog post is the following:
+    ```python
+    from transformers import AutoTokenizer, AutoModelForCausalLM, HybridCache
 
-*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` 
-
-<Tip warning={true}>
-
-- Gemma2 uses sliding window attention every second layer, which makes it unsuitable for typical kv caching with [`~DynamicCache`] or tuples of tensors. To enable caching in Gemma2 forward call, you must initialize a [`~HybridCache`] instance and pass it as `past_key_values` to the forward call. Note, that you also have to prepare `cache_position` if the `past_key_values` already contains previous keys and values.
-
-</Tip>
-
-This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ), [Pedro Cuenca](https://huggingface.co/pcuenq) and [Tom Arsen]().
+    model = AutoModelForCausalLM.from_pretrained("google/gemma-2-2b")
+    tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-2b")
 
+    inputs = tokenizer(text="My name is Gemma", return_tensors="pt")
+    max_generated_length = inputs.input_ids.shape[1] + 10
+    past_key_values = HybridCache(config=model.config, max_batch_size=1, 
+    max_cache_len=max_generated_length, device=model.device, dtype=model.dtype)
+    outputs = model(**inputs, past_key_values=past_key_values, use_cache=True)
+    ```
 
 ## Gemma2Config
 

docs/source/en/model_doc/glm4.md

@@ -1,4 +1,4 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
+<!--Copyright 2025 The GLM & ZhipuAI team and The HuggingFace Team. All rights reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
 the License. You may obtain a copy of the License at
@@ -14,13 +14,32 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# エージェントとツール
+# Glm4
 
-<Tip warning={true}>
+## Overview
 
-The Agents framework has significantly changed in version v4.41.0.
-This document has been removed as it was referencing an older API.
+To be released with the official model launch.
 
-We eagerly welcome new contributions for the updated API.
+## Glm4Config
 
-</Tip>
+[[autodoc]] Glm4Config
+
+## Glm4Model
+
+[[autodoc]] Glm4Model
+    - forward
+
+## Glm4ForCausalLM
+
+[[autodoc]] Glm4ForCausalLM
+    - forward
+
+## Glm4ForSequenceClassification
+
+[[autodoc]] Glm4ForSequenceClassification
+    - forward
+
+## Glm4ForTokenClassification
+
+[[autodoc]] Glm4ForTokenClassification
+    - forward

docs/source/en/model_doc/gpt2.md

@@ -14,197 +14,97 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# OpenAI GPT2
-
-<div class="flex flex-wrap space-x-1">
-<a href="https://huggingface.co/models?filter=gpt2">
-<img alt="Models" src="https://img.shields.io/badge/All_model_pages-gpt2-blueviolet">
-</a>
-<a href="https://huggingface.co/spaces/docs-demos/gpt2">
-<img alt="Spaces" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue">
-</a>
+<div style="float: right;">
+  <div class="flex flex-wrap space-x-1">
+    <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+    <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+    <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+  </div>
 </div>
 
-## Overview
 
-OpenAI GPT-2 model was proposed in [Language Models are Unsupervised Multitask Learners](https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf) by Alec
-Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei and Ilya Sutskever from [OpenAI](https://huggingface.co/openai). It's a causal (unidirectional)
-transformer pretrained using language modeling on a very large corpus of ~40 GB of text data.
+# GPT-2
 
-The abstract from the paper is the following:
+[GPT-2](https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf) is a scaled up version of GPT, a causal transformer language model, with 10x more parameters and training data. The model was pretrained on a 40GB dataset to predict the next word in a sequence based on all the previous words. This approach enabled the model to perform many downstream tasks in a zero-shot setting.
 
-*GPT-2 is a large transformer-based language model with 1.5 billion parameters, trained on a dataset[1] of 8 million
-web pages. GPT-2 is trained with a simple objective: predict the next word, given all of the previous words within some
-text. The diversity of the dataset causes this simple goal to contain naturally occurring demonstrations of many tasks
-across diverse domains. GPT-2 is a direct scale-up of GPT, with more than 10X the parameters and trained on more than
-10X the amount of data.*
+The model architecture uses a unidirectional (causal) attention mechanism where each token can only attend to previous tokens, making it particularly effective for text generation tasks.
 
-[Write With Transformer](https://transformer.huggingface.co/doc/gpt2-large) is a webapp created and hosted by
-Hugging Face showcasing the generative capabilities of several models. GPT-2 is one of them and is available in five
-different sizes: small, medium, large, xl and a distilled version of the small checkpoint: *distilgpt-2*.
+You can find all the original GPT-2 checkpoints under the [OpenAI community](https://huggingface.co/openai-community?search_models=gpt) organization.
 
-This model was contributed by [thomwolf](https://huggingface.co/thomwolf). The original code can be found [here](https://openai.com/blog/better-language-models/).
+> [!TIP]
+> Click on the GPT-2 models in the right sidebar for more examples of how to apply GPT-2 to different language tasks.
 
-## Usage tips
+The example below demonstrates how to generate text with [`Pipeline`] or the [`AutoModel`], and from the command line.
 
-- GPT-2 is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather than
-  the left.
-- GPT-2 was trained with a causal language modeling (CLM) objective and is therefore powerful at predicting the next
-  token in a sequence. Leveraging this feature allows GPT-2 to generate syntactically coherent text as it can be
-  observed in the *run_generation.py* example script.
-- The model can take the *past_key_values* (for PyTorch) or *past* (for TF) as input, which is the previously computed
-  key/value attention pairs. Using this (*past_key_values* or *past*) value prevents the model from re-computing
-  pre-computed values in the context of text generation. For PyTorch, see *past_key_values* argument of the
-  [`GPT2Model.forward`] method, or for TF the *past* argument of the
-  [`TFGPT2Model.call`] method for more information on its usage.
-- Enabling the *scale_attn_by_inverse_layer_idx* and *reorder_and_upcast_attn* flags will apply the training stability
-  improvements from [Mistral](https://github.com/stanford-crfm/mistral/) (for PyTorch only).
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-## Usage example
+```py
+import torch
+from transformers import pipeline
 
-The `generate()` method can be used to generate text using GPT2 model.
+pipeline = pipeline(task="text-generation", model="openai-community/gpt2", torch_dtype=torch.float16, device=0)
+pipeline("Hello, I'm a language model")
+```
+</hfoption>
+<hfoption id="AutoModel">
 
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
+```py
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
->>> model = AutoModelForCausalLM.from_pretrained("gpt2")
->>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
+model = AutoModelForCausalLM.from_pretrained("openai-community/gpt2", torch_dtype=torch.float16, device_map="auto", attn_implementation="sdpa")
+tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")
 
->>> prompt = "GPT2 is a model developed by OpenAI."
+input_ids = tokenzier("Hello, I'm a language model". return_tensors="pt").to("cuda")
 
->>> input_ids = tokenizer(prompt, return_tensors="pt").input_ids
-
->>> gen_tokens = model.generate(
-...     input_ids,
-...     do_sample=True,
-...     temperature=0.9,
-...     max_length=100,
-... )
->>> gen_text = tokenizer.batch_decode(gen_tokens)[0]
+output = model.generate(**input_ids, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
 
-## Using Flash Attention 2
-
-Flash Attention 2 is a faster, optimized version of the attention scores computation which relies on `cuda` kernels.
-
-### Installation 
-
-First, check whether your hardware is compatible with Flash Attention 2. The latest list of compatible hardware can be found in the [official documentation](https://github.com/Dao-AILab/flash-attention#installation-and-features). If your hardware is not compatible with Flash Attention 2, you can still benefit from attention kernel optimisations through Better Transformer support covered [above](https://huggingface.co/docs/transformers/main/en/model_doc/bark#using-better-transformer).
-
-Next, [install](https://github.com/Dao-AILab/flash-attention#installation-and-features) the latest version of Flash Attention 2:
+</hfoption>
+<hfoption id="transformers-cli">
 
 ```bash
-pip install -U flash-attn --no-build-isolation
+echo -e "Hello, I'm a language model" | transformers-cli run --task text-generation --model openai-community/gpt2 --device 0
 ```
 
-### Usage
+</hfoption>
+</hfoptions>
 
-To load a model using Flash Attention 2, we can pass the argument `attn_implementation="flash_attention_2"` to [`.from_pretrained`](https://huggingface.co/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained). We'll also load the model in half-precision (e.g. `torch.float16`), since it results in almost no degradation to audio quality but significantly lower memory usage and faster inference:
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
 
-```python
->>> import torch
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
->>> device = "cuda" # the device to load the model onto
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to only quantize the weights to 4-bits.
 
->>> model = AutoModelForCausalLM.from_pretrained("gpt2", torch_dtype=torch.float16, attn_implementation="flash_attention_2")
->>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
+```py
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
 
->>> prompt = "def hello_world():"
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,  
+    bnb_4bit_quant_type="nf4",  
+    bnb_4bit_compute_dtype="float16",  
+    bnb_4bit_use_double_quant=True 
+)
 
->>> model_inputs = tokenizer([prompt], return_tensors="pt").to(device)
->>> model.to(device)
+model = AutoModelForCausalLM.from_pretrained(
+    "openai-community/gpt2-xl",
+    quantization_config=quantization_config,
+    device_map="auto"  
+)
 
->>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
->>> tokenizer.batch_decode(generated_ids)[0]
+tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2-xl")
+inputs = tokenizer("Once upon a time, there was a magical forest", return_tensors="pt").to("cuda")
+outputs = model.generate(**inputs, max_new_tokens=100)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 
+## Notes
 
-### Expected speedups
-
-Below is an expected speedup diagram that compares pure inference time between the native implementation in transformers using `gpt2` checkpoint and the Flash Attention 2 version of the model using a sequence length of 512.
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/EduardoPacheco/documentation-images/resolve/main/gpt2_flash_attention_2_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 AutoModelForCausalLM
-model = AutoModelForCausalLM.from_pretrained("gpt2", 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
-[gpt2-large](https://huggingface.co/openai-community/gpt2-large), 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 |                       0.039 |                     0.032 |      23.042 |             3482.32 |            3494.62 |            -0.352 |
-|          1 |     256 |                       0.073 |                     0.059 |       25.15 |             3546.66 |             3552.6 |            -0.167 |
-|          1 |     512 |                       0.155 |                     0.118 |       30.96 |              4230.1 |            3665.59 |              15.4 |
-|          1 |    1024 |                       0.316 |                     0.209 |      50.839 |             8682.26 |            4881.09 |            77.875 |
-|          2 |     128 |                        0.07 |                      0.06 |      15.324 |              3557.8 |            3545.91 |             0.335 |
-|          2 |     256 |                       0.143 |                     0.122 |       16.53 |              3901.5 |            3657.68 |             6.666 |
-|          2 |     512 |                       0.267 |                     0.213 |      25.626 |             7062.21 |            4876.47 |            44.822 |
-|          2 |    1024 |                         OOM |                     0.404 |           / |                 OOM |            8096.35 | SDPA does not OOM |
-|          4 |     128 |                       0.134 |                     0.128 |       4.412 |             3675.79 |            3648.72 |             0.742 |
-|          4 |     256 |                       0.243 |                     0.217 |      12.292 |             6129.76 |            4871.12 |            25.839 |
-|          4 |     512 |                       0.494 |                     0.406 |      21.687 |             12466.6 |            8102.64 |            53.858 |
-|          4 |    1024 |                         OOM |                     0.795 |           / |                 OOM |            14568.2 | 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 |                        7.991 |                       6.968 |      14.681 |         1685.2 |       1701.32 |        -0.947 |
-|          1 |     256 |                        8.462 |                       7.199 |      17.536 |        1745.49 |       1770.78 |        -1.428 |
-|          1 |     512 |                         8.68 |                       7.853 |      10.529 |        1907.69 |       1921.29 |        -0.708 |
-|          1 |     768 |                        9.101 |                       8.365 |       8.791 |        2032.93 |       2068.12 |        -1.701 |
-|          2 |     128 |                        9.169 |                       9.001 |       1.861 |        1803.84 |        1811.4 |        -0.418 |
-|          2 |     256 |                        9.907 |                        9.78 |       1.294 |        1907.72 |       1921.44 |        -0.714 |
-|          2 |     512 |                       11.519 |                      11.644 |      -1.071 |        2176.86 |       2197.75 |        -0.951 |
-|          2 |     768 |                       13.022 |                      13.407 |      -2.873 |         2464.3 |       2491.06 |        -1.074 |
-|          4 |     128 |                       10.097 |                       9.831 |       2.709 |        1942.25 |       1985.13 |         -2.16 |
-|          4 |     256 |                       11.599 |                      11.398 |       1.764 |        2177.28 |       2197.86 |        -0.937 |
-|          4 |     512 |                       14.653 |                       14.45 |       1.411 |        2753.16 |       2772.57 |          -0.7 |
-|          4 |     768 |                       17.846 |                      17.617 |       1.299 |        3327.04 |       3343.97 |        -0.506 |
-
-
-
-
-## Resources
-
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with GPT2. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
-
-<PipelineTag pipeline="text-generation"/>
-
-- A blog on how to [Finetune a non-English GPT-2 Model with Hugging Face](https://www.philschmid.de/fine-tune-a-non-english-gpt-2-model-with-huggingface).
-- A blog on [How to generate text: using different decoding methods for language generation with Transformers](https://huggingface.co/blog/how-to-generate) with GPT-2.
-- A blog on [Training CodeParrot 🦜 from Scratch](https://huggingface.co/blog/codeparrot), a large GPT-2 model.
-- A blog on [Faster Text Generation with TensorFlow and XLA](https://huggingface.co/blog/tf-xla-generate) with GPT-2.
-- A blog on [How to train a Language Model with Megatron-LM](https://huggingface.co/blog/megatron-training) with a GPT-2 model.
-- A notebook on how to [finetune GPT2 to generate lyrics in the style of your favorite artist](https://colab.research.google.com/github/AlekseyKorshuk/huggingartists/blob/master/huggingartists-demo.ipynb). 🌎
-- A notebook on how to [finetune GPT2 to generate tweets in the style of your favorite Twitter user](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb). 🌎
-- [Causal language modeling](https://huggingface.co/course/en/chapter7/6?fw=pt#training-a-causal-language-model-from-scratch) chapter of the 🤗 Hugging Face Course.
-- [`GPT2LMHeadModel`] is supported by this [causal language modeling example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling#gpt-2gpt-and-causal-language-modeling), [text generation example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-generation), and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb).
-- [`TFGPT2LMHeadModel`] is supported by this [causal language modeling example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling#run_clmpy) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling-tf.ipynb).
-- [`FlaxGPT2LMHeadModel`] is supported by this [causal language modeling example script](https://github.com/huggingface/transformers/tree/main/examples/flax/language-modeling#causal-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/causal_language_modeling_flax.ipynb).
-- [Text classification task guide](../tasks/sequence_classification)
-- [Token classification task guide](../tasks/token_classification)
-- [Causal language modeling task guide](../tasks/language_modeling)
+- Pad inputs on the right because GPT-2 uses absolute position embeddings.
+- GPT-2 can reuse previously computed key-value attention pairs. Access this feature with the [past_key_values](https://huggingface.co/docs/transformers//en/model_doc/gpt2#transformers.GPT2Model.forward.past_key_values) parameter in [`GPT2Model.forward`].
+- Enable the [scale_attn_by_inverse_layer_idx](https://huggingface.co/docs/transformers/en/model_doc/gpt2#transformers.GPT2Config.scale_attn_by_inverse_layer_idx) and [reorder_and_upcast_attn](https://huggingface.co/docs/transformers/en/model_doc/gpt2#transformers.GPT2Config.reorder_and_upcast_attn) parameters to apply the training stability improvements from [Mistral](./mistral).
 
 ## GPT2Config
 

docs/source/en/model_doc/granite_speech.md

@@ -0,0 +1,68 @@
+<!--Copyright 2025 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.
+
+-->
+
+# Granite Speech
+
+<div class="flex flex-wrap space-x-1">
+<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+</div>
+
+## Overview
+The Granite Speech model is a multimodal language model, consisting of a speech encoder, speech projector, large language model, and LoRA adapter(s). More details regarding each component for the current (Granite 3.2 Speech) model architecture may be found below.
+
+1. Speech Encoder: A [Conformer](https://arxiv.org/abs/2005.08100) encoder trained with Connectionist Temporal Classification (CTC) on character-level targets on ASR corpora. The encoder uses block-attention and self-conditioned CTC from the middle layer.
+
+2. Speech Projector: A query transformer (q-former) operating on the outputs of the last encoder block. The encoder and projector temporally downsample the audio features to be merged into the multimodal embeddings to be processed by the llm.
+
+3. Large Language Model: The Granite Speech model leverages Granite LLMs, which were originally proposed in [this paper](https://arxiv.org/abs/2408.13359).
+
+4. LoRA adapter(s): The Granite Speech model contains a modality specific LoRA, which will be enabled when audio features are provided, and disabled otherwise.
+
+
+Note that most of the aforementioned components are implemented generically to enable compatability and potential integration with other model architectures in transformers.
+
+
+This model was contributed by [Alexander Brooks](https://huggingface.co/abrooks9944), [Avihu Dekel](https://huggingface.co/Avihu), and [George Saon](https://huggingface.co/gsaon).
+
+## Usage tips
+- This model bundles its own LoRA adapter, which will be automatically loaded and enabled/disabled as needed during inference calls. Be sure to install [PEFT](https://github.com/huggingface/peft) to ensure the LoRA is correctly applied!
+
+<!-- TODO (@alex-jw-brooks) Add an example here once the model compatible with the transformers implementation is released -->
+
+## GraniteSpeechConfig
+
+[[autodoc]] GraniteSpeechConfig
+
+
+## GraniteSpeechEncoderConfig
+
+[[autodoc]] GraniteSpeechEncoderConfig
+
+
+## GraniteSpeechProcessor
+
+[[autodoc]] GraniteSpeechProcessor
+
+
+## GraniteSpeechFeatureExtractor
+
+[[autodoc]] GraniteSpeechFeatureExtractor
+
+
+## GraniteSpeechForConditionalGeneration
+
+[[autodoc]] GraniteSpeechForConditionalGeneration
+    - forward

docs/source/en/model_doc/grounding-dino.md

@@ -102,6 +102,11 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 
 [[autodoc]] GroundingDinoImageProcessor
     - preprocess
+
+## GroundingDinoImageProcessorFast
+
+[[autodoc]] GroundingDinoImageProcessorFast
+    - preprocess
     - post_process_object_detection
 
 ## GroundingDinoProcessor

docs/source/en/model_doc/jamba.md

@@ -14,96 +14,126 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# Jamba
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+  <div class="flex flex-wrap space-x-1">
+    <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+    <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+  </div>
 </div>
 
-## Overview
+# Jamba
 
-Jamba is a state-of-the-art, hybrid SSM-Transformer LLM. It is the first production-scale Mamba implementation, which opens up interesting research and application opportunities. While this initial experimentation shows encouraging gains, we expect these to be further enhanced with future optimizations and explorations.
+[Jamba](https://huggingface.co/papers/2403.19887) is a hybrid Transformer-Mamba mixture-of-experts (MoE) language model ranging from 52B to 398B total parameters. This model aims to combine the advantages of both model families, the performance of transformer models and the efficiency and longer context (256K tokens) of state space models (SSMs) like Mamba.
 
-For full details of this model please read the [release blog post](https://www.ai21.com/blog/announcing-jamba).
+Jamba's architecture features a blocks-and-layers approach that allows Jamba to successfully integrate Transformer and Mamba architectures altogether. Each Jamba block contains either an attention or a Mamba layer, followed by a multi-layer perceptron (MLP), producing an overall ratio of one Transformer layer out of every eight total layers. MoE layers are mixed in to increase model capacity.
 
-### Model Details
+You can find all the original Jamba checkpoints under the [AI21](https://huggingface.co/ai21labs) organization.
 
-Jamba is a pretrained, mixture-of-experts (MoE) generative text model, with 12B active parameters and an overall of 52B parameters across all experts. It supports a 256K context length, and can fit up to 140K tokens on a single 80GB GPU.
+> [!TIP]
+> Click on the Jamba models in the right sidebar for more examples of how to apply Jamba to different language tasks.
 
-As depicted in the diagram below, Jamba's architecture features a blocks-and-layers approach that allows Jamba to successfully integrate Transformer and Mamba architectures altogether. Each Jamba block contains either an attention or a Mamba layer, followed by a multi-layer perceptron (MLP), producing an overall ratio of one Transformer layer out of every eight total layers.
+The example below demonstrates how to generate text with [`Pipeline`], [`AutoModel`], and from the command line.
 
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/jamba_architecture.png"
-alt="drawing" width="600"/>
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-## Usage
+```py
+# install optimized Mamba implementations
+# !pip install mamba-ssm causal-conv1d>=1.2.0
+import torch
+from transformers import pipeline
 
-### Prerequisites
-
-Jamba requires you use `transformers` version 4.39.0 or higher:
-```bash
-pip install transformers>=4.39.0
+pipeline = pipeline(
+    task="text-generation",
+    model="ai21labs/AI21-Jamba-Mini-1.6",
+    torch_dtype=torch.float16,
+    device=0
+)
+pipeline("Plants create energy through a process known as")
 ```
 
-In order to run optimized Mamba implementations, you first need to install `mamba-ssm` and `causal-conv1d`:
-```bash
-pip install mamba-ssm causal-conv1d>=1.2.0
-```
-You also have to have the model on a CUDA device.
+</hfoption>
+<hfoption id="AutoModel">
 
-You can run the model not using the optimized Mamba kernels, but it is **not** recommended as it will result in significantly lower latencies. In order to do that, you'll need to specify `use_mamba_kernels=False` when loading the model.
-
-### Run the model
-```python
+```py
+import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
-model = AutoModelForCausalLM.from_pretrained("ai21labs/Jamba-v0.1")
-tokenizer = AutoTokenizer.from_pretrained("ai21labs/Jamba-v0.1")
+tokenizer = AutoTokenizer.from_pretrained(
+    "ai21labs/AI21-Jamba-Large-1.6",
+)
+model = AutoModelForCausalLM.from_pretrained(
+    "ai21labs/AI21-Jamba-Large-1.6",
+    torch_dtype=torch.float16,
+    device_map="auto",
+    attn_implementation="sdpa"
+)
+input_ids = tokenizer("Plants create energy through a process known as", return_tensors="pt").to("cuda")
 
-input_ids = tokenizer("In the recent Super Bowl LVIII,", return_tensors='pt').to(model.device)["input_ids"]
+output = model.generate(**input_ids, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+<hfoption id="transformers-cli">
+
+```bash
+echo -e "Plants create energy through a process known as" | transformers-cli run --task text-generation --model ai21labs/AI21-Jamba-Mini-1.6 --device 0
+```
+
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to only quantize the weights to 8-bits.
+
+```py
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+
+quantization_config = BitsAndBytesConfig(load_in_8bit=True,
+                                         llm_int8_skip_modules=["mamba"])
+
+# a device map to distribute the model evenly across 8 GPUs
+device_map = {'model.embed_tokens': 0, 'model.layers.0': 0, 'model.layers.1': 0, 'model.layers.2': 0, 'model.layers.3': 0, 'model.layers.4': 0, 'model.layers.5': 0, 'model.layers.6': 0, 'model.layers.7': 0, 'model.layers.8': 0, 'model.layers.9': 1, 'model.layers.10': 1, 'model.layers.11': 1, 'model.layers.12': 1, 'model.layers.13': 1, 'model.layers.14': 1, 'model.layers.15': 1, 'model.layers.16': 1, 'model.layers.17': 1, 'model.layers.18': 2, 'model.layers.19': 2, 'model.layers.20': 2, 'model.layers.21': 2, 'model.layers.22': 2, 'model.layers.23': 2, 'model.layers.24': 2, 'model.layers.25': 2, 'model.layers.26': 2, 'model.layers.27': 3, 'model.layers.28': 3, 'model.layers.29': 3, 'model.layers.30': 3, 'model.layers.31': 3, 'model.layers.32': 3, 'model.layers.33': 3, 'model.layers.34': 3, 'model.layers.35': 3, 'model.layers.36': 4, 'model.layers.37': 4, 'model.layers.38': 4, 'model.layers.39': 4, 'model.layers.40': 4, 'model.layers.41': 4, 'model.layers.42': 4, 'model.layers.43': 4, 'model.layers.44': 4, 'model.layers.45': 5, 'model.layers.46': 5, 'model.layers.47': 5, 'model.layers.48': 5, 'model.layers.49': 5, 'model.layers.50': 5, 'model.layers.51': 5, 'model.layers.52': 5, 'model.layers.53': 5, 'model.layers.54': 6, 'model.layers.55': 6, 'model.layers.56': 6, 'model.layers.57': 6, 'model.layers.58': 6, 'model.layers.59': 6, 'model.layers.60': 6, 'model.layers.61': 6, 'model.layers.62': 6, 'model.layers.63': 7, 'model.layers.64': 7, 'model.layers.65': 7, 'model.layers.66': 7, 'model.layers.67': 7, 'model.layers.68': 7, 'model.layers.69': 7, 'model.layers.70': 7, 'model.layers.71': 7, 'model.final_layernorm': 7, 'lm_head': 7}
+model = AutoModelForCausalLM.from_pretrained("ai21labs/AI21-Jamba-Large-1.6",
+                                             torch_dtype=torch.bfloat16,
+                                             attn_implementation="flash_attention_2",
+                                             quantization_config=quantization_config,
+                                             device_map=device_map)
+
+tokenizer = AutoTokenizer.from_pretrained("ai21labs/AI21-Jamba-Large-1.6")
+
+messages = [
+   {"role": "system", "content": "You are an ancient oracle who speaks in cryptic but wise phrases, always hinting at deeper meanings."},
+   {"role": "user", "content": "Hello!"},
+]
+
+input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors='pt').to(model.device)
 
 outputs = model.generate(input_ids, max_new_tokens=216)
 
-print(tokenizer.batch_decode(outputs))
-# ["<|startoftext|>In the recent Super Bowl LVIII, the Kansas City Chiefs emerged victorious, defeating the San Francisco 49ers in a thrilling overtime showdown. The game was a nail-biter, with both teams showcasing their skills and determination.\n\nThe Chiefs, led by their star quarterback Patrick Mahomes, displayed their offensive prowess, while the 49ers, led by their strong defense, put up a tough fight. The game went into overtime, with the Chiefs ultimately securing the win with a touchdown.\n\nThe victory marked the Chiefs' second Super Bowl win in four years, solidifying their status as one of the top teams in the NFL. The game was a testament to the skill and talent of both teams, and a thrilling end to the NFL season.\n\nThe Super Bowl is not just about the game itself, but also about the halftime show and the commercials. This year's halftime show featured a star-studded lineup, including Usher, Alicia Keys, and Lil Jon. The show was a spectacle of music and dance, with the performers delivering an energetic and entertaining performance.\n"]
+# Decode the output
+conversation = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+# Split the conversation to get only the assistant's response
+assistant_response = conversation.split(messages[-1]['content'])[1].strip()
+print(assistant_response)
+# Output: Seek and you shall find. The path is winding, but the journey is enlightening. What wisdom do you seek from the ancient echoes?
 ```
 
-<details>
-<summary><strong>Loading the model in half precision</strong></summary>
+## Notes
 
-The published checkpoint is saved in BF16. In order to load it into RAM in BF16/FP16, you need to specify `torch_dtype`:
+- Don't quantize the Mamba blocks to prevent model performance degradation.
+- It is not recommended to use Mamba without the optimized Mamba kernels as it results in significantly lower latencies. If you still want to use Mamba without the kernels, then set `use_mamba_kernels=False` in [`~AutoModel.from_pretrained`].
 
-```python
-from transformers import AutoModelForCausalLM
-import torch
-model = AutoModelForCausalLM.from_pretrained("ai21labs/Jamba-v0.1", torch_dtype=torch.bfloat16)
-# you can also use torch_dtype=torch.float16
-```
-
-When using half precision, you can enable the [FlashAttention2](https://github.com/Dao-AILab/flash-attention) implementation of the Attention blocks. In order to use it, you also need the model on a CUDA device. Since in this precision the model is to big to fit on a single 80GB GPU, you'll also need to parallelize it using [accelerate](https://huggingface.co/docs/accelerate/index):
-```python
-from transformers import AutoModelForCausalLM
-import torch
-model = AutoModelForCausalLM.from_pretrained("ai21labs/Jamba-v0.1",
-                                             torch_dtype=torch.bfloat16,
-                                             attn_implementation="flash_attention_2",
-                                             device_map="auto")
-```
-
-</details>
-<details><summary><strong>Load the model in 8-bit</strong></summary>
-
-**Using 8-bit precision, it is possible to fit up to 140K sequence lengths on a single 80GB GPU.** You can easily quantize the model to 8-bit using [bitsandbytes](https://huggingface.co/docs/bitsandbytes/index). In order to not degrade model quality, we recommend to exclude the Mamba blocks from the quantization:
-
-```python
-from transformers import AutoModelForCausalLM, BitsAndBytesConfig
-quantization_config = BitsAndBytesConfig(load_in_8bit=True, llm_int8_skip_modules=["mamba"])
-model = AutoModelForCausalLM.from_pretrained(
-    "ai21labs/Jamba-v0.1", torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2", quantization_config=quantization_config
-)
-```
-</details>
+    ```py
+    import torch
+    from transformers import AutoModelForCausalLM
+    model = AutoModelForCausalLM.from_pretrained("ai21labs/AI21-Jamba-1.5-Large",
+                                                 use_mamba_kernels=False)
+    ```
 
 ## JambaConfig
 

docs/source/en/model_doc/janus.md

@@ -0,0 +1,230 @@
+<!--Copyright 2025 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.
+
+-->
+
+# Janus
+
+## Overview
+
+The Janus Model was originally proposed in [Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation](https://arxiv.org/abs/2410.13848) by DeepSeek AI team and later refined in [Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling](https://arxiv.org/abs/2501.17811). Janus is a vision-language model that can generate both image and text output, it can also take both images and text as input.
+
+> [!NOTE]
+> The model doesn't generate both images and text in an interleaved format. The user has to pass a parameter indicating whether to generate text or image.
+
+The abstract from the original paper is the following:
+
+*In this paper, we introduce Janus, an autoregressive framework that unifies multimodal understanding and generation. Prior research often relies on a single visual encoder for both tasks, such as Chameleon. However, due to the differing levels of information granularity required by multimodal understanding and generation, this approach can lead to suboptimal performance, particularly in multimodal understanding. To address this issue, we decouple visual encoding into separate pathways, while still leveraging a single, unified transformer architecture for processing. The decoupling not only alleviates the conflict between the visual encoder's roles in understanding and generation, but also enhances the framework's flexibility. For instance, both the multimodal understanding and generation components can independently select their most suitable encoding methods. Experiments show that Janus surpasses previous unified model and matches or exceeds the performance of task-specific models. The simplicity, high flexibility, and effectiveness of Janus make it a strong candidate for next-generation unified multimodal models.*
+
+The abstract from the aforementioned `Janus-Pro` paper, released afterwards, is the following:
+
+*In this work, we introduce Janus-Pro, an advanced version of the previous work Janus. Specifically, Janus-Pro incorporates (1) an optimized training strate (2) expanded training data, and (3) scaling to larger model size. With these improvements, Janus-Pro achieves significant advancements in both multimodal understanding and text-to-image instruction-following capabilities, while also enhancing the stability of text-to-image generation. We hope this work will inspire further exploration in the field. Code and models are publicly available.*
+
+This model was contributed by [Yaswanth Gali](https://huggingface.co/yaswanthgali) and [Hugo Silva](https://huggingface.co/hugosilva664).
+The original code can be found [here](https://github.com/deepseek-ai/Janus).
+
+## Usage Example
+
+### Single image inference
+
+Here is the example of visual understanding with a single image.
+
+> [!NOTE]
+> Note that the model has been trained with a specific prompt format for chatting. Use `processor.apply_chat_template(my_conversation_dict)` to correctly format your prompts.
+
+```python
+import torch  
+from PIL import Image  
+import requests  
+
+from transformers import JanusForConditionalGeneration, JanusProcessor  
+
+model_id = "deepseek-community/Janus-Pro-1B"
+# Prepare Input for generation.
+messages = [
+    {
+        "role": "user",
+        "content": [
+            {'type':'image', 'url': 'http://images.cocodataset.org/val2017/000000039769.jpg'},
+            {'type':"text", "text":"What do you see in this image?."}
+        ]
+    },
+]
+
+# Set generation mode to `text` to perform text generation.
+processor = JanusProcessor.from_pretrained(model_id)
+model = JanusForConditionalGeneration.from_pretrained(model_id,     
+        torch_dtype=torch.bfloat16,
+        device_map="auto")
+
+inputs = processor.apply_chat_template(
+    messages,
+    add_generation_prompt=True,
+    generation_mode="text",
+    tokenize=True,
+    return_dict=True,
+    return_tensors="pt",
+).to(model.device, dtype=torch.bfloat16)
+
+output = model.generate(**inputs, max_new_tokens=40,generation_mode='text',do_sample=True)
+text = processor.decode(output[0], skip_special_tokens=True)
+print(text)
+```
+
+### Multi image inference
+
+Janus can perform inference with multiple images as input, where images can belong to the same prompt or different prompts in batched inference, where the model processes many conversations in parallel. Here is how you can do it:
+
+```python
+import torch
+from PIL import Image
+import requests
+
+from transformers import JanusForConditionalGeneration, JanusProcessor
+
+model_id = "deepseek-community/Janus-Pro-1B"
+
+image_urls = [
+    "http://images.cocodataset.org/val2017/000000039769.jpg",
+    "https://www.ilankelman.org/stopsigns/australia.jpg",
+    "https://huggingface.co/microsoft/kosmos-2-patch14-224/resolve/main/snowman.jpg"
+]
+
+messages = [
+    [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "What’s the difference between"},
+                {"type": "image", "url": image_urls[0]},
+                {"type": "text", "text": " and "},
+                {"type": "image", "url": image_urls[1]}
+            ]
+        }
+    ],
+    [
+        {
+            "role": "user",
+            "content": [
+                {"type": "image", "url": image_urls[2]},
+                {"type": "text", "text": "What do you see in this image?"}
+            ]
+        }
+    ]
+]
+
+# Load model and processor
+processor = JanusProcessor.from_pretrained(model_id)
+model = JanusForConditionalGeneration.from_pretrained(
+    model_id, torch_dtype=torch.bfloat16, device_map="auto"
+)
+
+inputs = processor.apply_chat_template(
+    messages,
+    add_generation_prompt=True,
+    generation_mode="text",
+    tokenize=True,
+    padding=True,
+    return_dict=True,
+    return_tensors="pt"
+).to(model.device, dtype=torch.bfloat16)
+
+# Generate response
+output = model.generate(**inputs, max_new_tokens=40, generation_mode='text', do_sample=False)
+text = processor.batch_decode(output, skip_special_tokens=True)
+print(text)
+```
+
+## Text to Image generation
+
+Janus can also generate images given a prompt.
+
+```python
+import torch
+from transformers import JanusForConditionalGeneration, JanusProcessor
+
+# Set generation mode to `image` to prepare inputs for image generation..
+
+model_id = "deepseek-community/Janus-Pro-1B"
+processor = JanusProcessor.from_pretrained(model_id)
+model = JanusForConditionalGeneration.from_pretrained(model_id,
+        torch_dtype=torch.bfloat16,
+        device_map="auto")
+
+messages = [
+    {
+        "role": "user",
+        "content": [
+            {"type": "text", "text": "A dog running under the rain."},
+        ],
+     }
+]
+
+prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
+inputs = processor(text=prompt,generation_mode="image",return_tensors="pt").to(model.device, dtype=torch.bfloat16)
+
+# Set num_return_sequence parameter to generate multiple images per prompt.
+model.generation_config.num_return_sequences = 2
+outputs = model.generate(**inputs,
+                         generation_mode="image",
+                         do_sample=True,
+                         use_cache=True,
+                         )
+# Perform post-processing on the generated token ids.
+decoded_image = model.decode_image_tokens(outputs)
+images = processor.postprocess(list(decoded_image.float()),return_tensors="PIL.Image.Image")
+# Save the image
+for i, image in enumerate(images['pixel_values']):
+    image.save(f"result{i}.png")
+```
+
+## JanusConfig
+
+[[autodoc]] JanusConfig
+
+## JanusVisionConfig
+
+[[autodoc]] JanusVisionConfig
+
+## JanusVQVAEConfig
+
+[[autodoc]] JanusVQVAEConfig
+
+## JanusProcessor
+
+[[autodoc]] JanusProcessor
+
+## JanusImageProcessor
+
+[[autodoc]] JanusImageProcessor
+
+## JanusVisionModel
+
+[[autodoc]] JanusVisionModel
+    - forward
+
+## JanusVQVAE
+
+[[autodoc]] JanusVQVAE
+    - forward
+
+## JanusModel
+
+[[autodoc]] JanusModel
+    - forward
+
+## JanusForConditionalGeneration
+
+[[autodoc]] JanusForConditionalGeneration
+    - forward

docs/source/en/model_doc/layoutlmv2.md

@@ -310,6 +310,11 @@ print(encoding.keys())
 [[autodoc]] LayoutLMv2ImageProcessor
     - preprocess
 
+## LayoutLMv2ImageProcessorFast
+
+[[autodoc]] LayoutLMv2ImageProcessorFast
+    - preprocess
+
 ## LayoutLMv2Tokenizer
 
 [[autodoc]] LayoutLMv2Tokenizer

docs/source/en/model_doc/layoutlmv3.md

@@ -88,6 +88,11 @@ LayoutLMv3 is nearly identical to LayoutLMv2, so we've also included LayoutLMv2
 [[autodoc]] LayoutLMv3ImageProcessor
     - preprocess
 
+## LayoutLMv3ImageProcessorFast
+
+[[autodoc]] LayoutLMv3ImageProcessorFast
+    - preprocess
+
 ## LayoutLMv3Tokenizer
 
 [[autodoc]] LayoutLMv3Tokenizer

docs/source/en/model_doc/levit.md

@@ -94,6 +94,11 @@ If you're interested in submitting a resource to be included here, please feel f
   [[autodoc]] LevitImageProcessor
     - preprocess
 
+## LevitImageProcessorFast
+
+  [[autodoc]] LevitImageProcessorFast
+    - preprocess
+
 ## LevitModel
 
 [[autodoc]] LevitModel

docs/source/en/model_doc/mistral.md

@@ -14,74 +14,55 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# Mistral
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
-">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
+        ">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
+# Mistral
 
-Mistral was introduced in the [this blogpost](https://mistral.ai/news/announcing-mistral-7b/) by Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
+[Mistral](https://huggingface.co/papers/2310.06825) is a 7B parameter language model, available as a pretrained and instruction-tuned variant, focused on balancing 
+the scaling costs of large models with performance and efficient inference. This model uses sliding window attention (SWA) trained with a 8K context length and a fixed cache size to handle longer sequences more effectively. Grouped-query attention (GQA) speeds up inference and reduces memory requirements. Mistral also features a byte-fallback BPE tokenizer to improve token handling and efficiency by ensuring characters are never mapped to out-of-vocabulary tokens.
 
-The introduction of the blog post says:
+You can find all the original Mistral checkpoints under the [Mistral AI_](https://huggingface.co/mistralai) organization.
 
-*Mistral AI team is proud to release Mistral 7B, the most powerful language model for its size to date.*
+> [!TIP]
+> Click on the Mistral models in the right sidebar for more examples of how to apply Mistral to different language tasks.
 
-Mistral-7B is the first large language model (LLM) released by [mistral.ai](https://mistral.ai/).
+The example below demonstrates how to chat with [`Pipeline`] or the [`AutoModel`], and from the command line.
 
-### Architectural details
-
-Mistral-7B is a decoder-only Transformer with the following architectural choices:
-
-- Sliding Window Attention - Trained with 8k context length and fixed cache size, with a theoretical attention span of 128K tokens
-- GQA (Grouped Query Attention) - allowing faster inference and lower cache size.
-- Byte-fallback BPE tokenizer - ensures that characters are never mapped to out of vocabulary tokens.
-
-For more details refer to the [release blog post](https://mistral.ai/news/announcing-mistral-7b/).
-
-### License
-
-`Mistral-7B` is released under the Apache 2.0 license.
-
-## Usage tips
-
-The Mistral team has released 3 checkpoints:
-
-- a base model, [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1), which has been pre-trained to predict the next token on internet-scale data.
-- an instruction tuned model, [Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1), which is the base model optimized for chat purposes using supervised fine-tuning (SFT) and direct preference optimization (DPO).
-- an improved instruction tuned model, [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2), which improves upon v1.
-
-The base model can be used as follows:
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
 ```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
+>>> import torch
+>>> from transformers import pipeline
 
->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1", device_map="auto")
->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
+>>> messages = [
+...     {"role": "user", "content": "What is your favourite condiment?"},
+...     {"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"},
+...     {"role": "user", "content": "Do you have mayonnaise recipes?"}
+... ]
 
->>> prompt = "My favourite condiment is"
-
->>> model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
->>> model.to(device)
-
->>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
->>> tokenizer.batch_decode(generated_ids)[0]
-"My favourite condiment is to ..."
+>>> chatbot = pipeline("text-generation", model="mistralai/Mistral-7B-Instruct-v0.3", torch_dtype=torch.bfloat16, device=0)
+>>> chatbot(messages)
 ```
 
-The instruction tuned model can be used as follows:
+</hfoption>
+<hfoption id="AutoModel">
 
 ```python
+>>> import torch
 >>> from transformers import AutoModelForCausalLM, AutoTokenizer
 
->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2", device_map="auto")
->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
+>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3", torch_dtype=torch.bfloat16, attn_implementation="sdpa", device_map="auto")
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3")
 
 >>> messages = [
 ...     {"role": "user", "content": "What is your favourite condiment?"},
@@ -96,59 +77,20 @@ The instruction tuned model can be used as follows:
 "Mayonnaise can be made as follows: (...)"
 ```
 
-As can be seen, the instruction-tuned model requires a [chat template](../chat_templating) to be applied to make sure the inputs are prepared in the right format.
-
-## Speeding up Mistral by using Flash Attention
-
-The code snippets above showcase inference without any optimization tricks. However, one can drastically speed up the model by leveraging [Flash Attention](../perf_train_gpu_one#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 to include the sliding window attention feature.
-
-```bash
-pip install -U flash-attn --no-build-isolation
-```
-
-Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of the [flash attention repository](https://github.com/Dao-AILab/flash-attention). 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:
+</hfoption>
+<hfoption id="transformers-cli">
 
 ```python
->>> import torch
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
-
->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1", torch_dtype=torch.float16, attn_implementation="flash_attention_2", device_map="auto")
->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
-
->>> prompt = "My favourite condiment is"
-
->>> model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
->>> model.to(device)
-
->>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
->>> tokenizer.batch_decode(generated_ids)[0]
-"My favourite condiment is to (...)"
+echo -e "My favorite condiment is" | transformers-cli chat --model_name_or_path mistralai/Mistral-7B-v0.3 --torch_dtype auto --device 0 --attn_implementation flash_attention_2
 ```
 
-### Expected speedups
+</hfoption>
+</hfoptions>
 
-Below is a expected speedup diagram that compares pure inference time between the native implementation in transformers using `mistralai/Mistral-7B-v0.1` checkpoint and the Flash Attention 2 version of the model.
 
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/mistral-7b-inference-large-seqlen.png">
-</div>
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
 
-### Sliding window Attention
-
-The current implementation supports the sliding window attention mechanism and memory efficient cache management. 
-To enable sliding window attention, just make sure to have a `flash-attn` version that is compatible with sliding window attention (`>=2.3.0`). 
-
-The Flash Attention-2 model uses also a more memory efficient cache slicing mechanism - as recommended per the official implementation of Mistral model that use rolling cache mechanism we keep the cache size fixed (`self.config.sliding_window`), support batched generation only for `padding_side="left"` and use the absolute position of the current token to compute the positional embedding.
-
-## Shrinking down Mistral using quantization
-
-As the Mistral model has 7 billion parameters, that would require about 14GB of GPU RAM in half precision (float16), since each parameter is stored in 2 bytes. However, one can shrink down the size of the model using [quantization](../quantization.md). If the model is quantized to 4 bits (or half a byte per parameter),that requires only about 3.5GB of RAM.
-
-Quantizing a model is as simple as passing a `quantization_config` to the model. Below, we'll leverage the BitsAndyBytes quantization (but refer to [this page](../quantization.md) for other quantization methods):
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to only quantize the weights to 4-bits.
 
 ```python
 >>> import torch
@@ -161,8 +103,8 @@ Quantizing a model is as simple as passing a `quantization_config` to the model.
 ...         bnb_4bit_compute_dtype="torch.float16",
 ... )
 
->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2", quantization_config=True, device_map="auto")
->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
+>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3", quantization_config=True, torch_dtype=torch.bfloat16, device_map="auto")
+>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3")
 
 >>> prompt = "My favourite condiment is"
 
@@ -179,19 +121,18 @@ Quantizing a model is as simple as passing a `quantization_config` to the model.
 "The expected output"
 ```
 
-This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArthurZ) .
-The original code can be found [here](https://github.com/mistralai/mistral-src).
+Use the [AttentionMaskVisualizer](https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139) to better understand what tokens the model can and cannot attend to.
 
-## Resources
+```py
+>>> from transformers.utils.attention_visualizer import AttentionMaskVisualizer
 
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Mistral. 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.
+>>> visualizer = AttentionMaskVisualizer("mistralai/Mistral-7B-Instruct-v0.3")
+>>> visualizer("Do you have mayonnaise recipes?")
+```
 
-<PipelineTag pipeline="text-generation"/>
-
-- A demo notebook to perform supervised fine-tuning (SFT) of Mistral-7B can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Mistral/Supervised_fine_tuning_(SFT)_of_an_LLM_using_Hugging_Face_tooling.ipynb). 🌎
-- A [blog post](https://www.philschmid.de/fine-tune-llms-in-2024-with-trl) on how to fine-tune LLMs in 2024 using Hugging Face tooling. 🌎
-- The [Alignment Handbook](https://github.com/huggingface/alignment-handbook) by Hugging Face includes scripts and recipes to perform supervised fine-tuning (SFT) and direct preference optimization with Mistral-7B. This includes scripts for full fine-tuning, QLoRa on a single GPU as well as multi-GPU fine-tuning.
-- [Causal language modeling task guide](../tasks/language_modeling)
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/mistral-attn-mask.png"/>
+</div>
 
 ## MistralConfig
 
@@ -245,4 +186,4 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 ## TFMistralForSequenceClassification
 
 [[autodoc]] TFMistralForSequenceClassification
-    - call
+    - call

docs/source/en/model_doc/mlcd.md

@@ -0,0 +1,81 @@
+<!--Copyright 2025 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.
+
+-->
+
+# MLCD
+
+<div class="flex flex-wrap space-x-1">
+<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+</div>
+
+## Overview
+
+The MLCD models were released by the DeepGlint-AI team in [unicom](https://github.com/deepglint/unicom), which focuses on building foundational visual models for large multimodal language models using large-scale datasets such as LAION400M and COYO700M, and employs sample-to-cluster contrastive learning to optimize performance. MLCD models are primarily used for multimodal visual large language models, such as LLaVA.
+
+🔥**MLCD-ViT-bigG**🔥 series is the state-of-the-art vision transformer model enhanced with 2D Rotary Position Embedding (RoPE2D), achieving superior performance on document understanding and visual question answering tasks. Developed by DeepGlint AI, this model demonstrates exceptional capabilities in processing complex visual-language interactions.
+
+Tips:
+
+- We adopted the official [LLaVA-NeXT](https://github.com/LLaVA-VL/LLaVA-NeXT) and the official training dataset [LLaVA-NeXT-Data](https://huggingface.co/datasets/lmms-lab/LLaVA-NeXT-Data) for evaluating the foundational visual models.
+
+- The language model is [Qwen2.5-7B](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct). 
+
+Result: 
+
+| Vision Tower                                                                                  | RoPE2D | ChartQA   | DocVQA    | InfoVQA   | OCRBench   | MMMU      |
+| :-------------------------------------------------------------------------------------------- | :----: | :-------- | :-------- | :-------- | :--------- | :-------- |
+| CLIP (ViT-L-14-336px)                                                                         |   ×    | 66.52     | 75.21     | 38.88     | 525.00     | 44.20     |
+| SigLIP (ViT-SO400M-384px)                                                                     |   ×    | 69.28     | 76.71     | 41.38     | 554.00     | 46.78     |
+| DFN5B (ViT-H-14-378px)                                                                        |   ×    | 64.36     | 70.87     | 38.59     | 473.00     | **48.00** |
+| **[MLCD (ViT-L-14-336px)](https://huggingface.co/DeepGlint-AI/mlcd-vit-large-patch14-336)**   |   ×    | 67.84     | 76.46     | 43.48     | 531.00     | 44.30     |
+| **[MLCD (ViT-bigG-14-336px)](https://huggingface.co/DeepGlint-AI/mlcd-vit-bigG-patch14-336)** |   √    | 71.07     | 79.63     | 44.38     | 572.00     | 46.78     |
+| **[MLCD (ViT-bigG-14-448px)](https://huggingface.co/DeepGlint-AI/mlcd-vit-bigG-patch14-448)** |   √    | **73.80** | **83.34** | **46.59** | **582.00** | 46.00     |
+
+
+## Usage
+
+```python
+import requests
+from PIL import Image
+from transformers import AutoProcessor, MLCDVisionModel
+
+# Load model and processor
+model = MLCDVisionModel.from_pretrained("DeepGlint-AI/mlcd-vit-bigG-patch14-448")
+processor = AutoProcessor.from_pretrained("DeepGlint-AI/mlcd-vit-bigG-patch14-448")
+
+# Process single image
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+inputs = processor(images=image, return_tensors="pt")
+
+# Generate outputs
+with torch.no_grad():
+    outputs = model(**inputs)
+
+# Get visual features
+features = outputs.last_hidden_state
+
+print(f"Extracted features shape: {features.shape}")
+```
+
+## MLCDVisionConfig
+
+[[autodoc]] MLCDVisionConfig
+
+## MLCDVisionModel
+
+[[autodoc]] MLCDVisionModel
+    - forward

docs/source/en/model_doc/mobilenet_v2.md

@@ -84,6 +84,11 @@ If you're interested in submitting a resource to be included here, please feel f
 
 [[autodoc]] MobileNetV2ImageProcessor
     - preprocess
+
+## MobileNetV2ImageProcessorFast
+
+[[autodoc]] MobileNetV2ImageProcessorFast
+    - preprocess
     - post_process_semantic_segmentation
 
 ## MobileNetV2Model

docs/source/en/model_doc/owlvit.md

@@ -94,6 +94,11 @@ A demo notebook on using OWL-ViT for zero- and one-shot (image-guided) object de
 
 [[autodoc]] OwlViTImageProcessor
     - preprocess
+
+## OwlViTImageProcessorFast
+
+[[autodoc]] OwlViTImageProcessorFast
+    - preprocess
     - post_process_object_detection
     - post_process_image_guided_detection
 

docs/source/en/model_doc/perceiver.md

@@ -132,6 +132,11 @@ audio classification, video classification, etc.
 [[autodoc]] PerceiverImageProcessor
     - preprocess
 
+## PerceiverImageProcessorFast
+
+[[autodoc]] PerceiverImageProcessorFast
+    - preprocess
+
 ## PerceiverTextPreprocessor
 
 [[autodoc]] models.perceiver.modeling_perceiver.PerceiverTextPreprocessor

docs/source/en/model_doc/phi4_multimodal.md

@@ -64,7 +64,7 @@ inputs = processor.apply_chat_template(
     tokenize=True,
     return_dict=True,
     return_tensors="pt",
-).to(device, torch.float16)
+).to(device)
 
 # Generate response
 generate_ids = model.generate(
@@ -98,8 +98,7 @@ inputs = processor.apply_chat_template(
     tokenize=True,
     return_dict=True,
     return_tensors="pt",
-    sample_rate=sample_rate,
-).to(device, torch.float16)
+).to(device)
 
 generate_ids = model.generate(
     **inputs,

docs/source/en/model_doc/qwen2_5_omni.md

@@ -0,0 +1,400 @@
+<!--Copyright 2025 The Qwen Team and The HuggingFace Inc. 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.
+
+-->
+
+# Qwen2.5-Omni
+
+<div class="flex flex-wrap space-x-1">
+<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+</div>
+
+## Overview
+
+The [Qwen2.5-Omni](https://qwenlm.github.io/blog/) model is a unified multiple modalities model proposed in [Qwen2.5-Omni Technical Report]() from Qwen team, Alibaba Group.
+
+The abstract from the technical report is the following:
+
+*We present Qwen2.5-Omni, an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. To enable the streaming of multimodal information inputs, both audio and visual encoders utilize a block-wise processing approach. This strategy effectively decouples the handling of long sequences of multimodal data, assigning the perceptual responsibilities to the multimodal encoder and entrusting the modeling of extended sequences to a large language model. Such a division of labor enhances the fusion of different modalities via the shared attention mechanism. To synchronize the timestamps of video inputs with audio, we organized the audio and video sequentially in an interleaved manner and propose a novel position embedding approach, named TMRoPE (Time-aligned Multimodal RoPE). To concurrently generate text and speech while avoiding interference between the two modalities, we propose Thinker-Talker architecture. In this framework, Thinker functions as a large language model tasked with text generation, while Talker is a dual-track autoregressive model that directly utilizes the hidden representations from the Thinker to produce audio tokens as output. Both the Thinker and Talker models are designed to be trained and inferred in an end-to-end manner. For decoding audio tokens in a streaming manner, we introduce a sliding-window DiT that restricts the receptive field, aiming to reduce the initial package delay. Qwen2.5-Omni outperforms the similarly sized Qwen2-VL and Qwen2-Audio in both image and audio capabilities. Furthermore, Qwen2.5-Omni achieves state-of-the-art performance on multimodal benchmarks like Omni-Bench. Notably, Qwen2.5-Omni is the first open-source model to achieve a level of performance in end-to-end speech instruction following that is comparable to its capabilities with text inputs, as evidenced by benchmarks such as MMLU and GSM8K. As for speech generation, Qwen2.5-Omni’s streaming Talker outperform most existing streaming and non-streaming alternatives in robustness and naturalness.*
+
+
+
+## Notes
+
+- Use [`Qwen2_5OmniForConditionalGeneration`] to generate audio and text output. To generate only one output type, use [`Qwen2_5OmniThinkerForConditionalGeneration`] for text-only and [`Qwen2_5OmniTalkersForConditionalGeneration`] for audio-only outputs.
+- Audio generation with [`Qwen2_5OmniForConditionalGeneration`] supports only single batch size at the moment.
+- In case out out-of-memory errors hwen working with video input, decrease `processor.max_pixels`. By default the maximum is set to a very arge value and high resolution visuals will not be resized, unless resolution exceeds `processor.max_pixels`.
+- The processor has its own [`~ProcessorMixin.apply_chat_template`] method to convert chat messages to model inputs.
+
+
+## Usage example
+
+`Qwen2.5-Omni` can be found on the [Huggingface Hub](https://huggingface.co/Qwen).
+
+### Single Media inference
+
+The model can accept text, images, audio and videos as input. Here's an example code for inference.
+
+```python
+import soundfile as sf
+from transformers import Qwen2_5OmniForConditionalGeneration, Qwen2_5OmniProcessor
+
+model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
+    "Qwen/Qwen2.5-Omni-7B",
+    torch_dtype="auto",
+    device_map="auto"
+)
+processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")
+
+conversation = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
+        ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "video", "video": "/path/to/video.mp4"},
+            {"type": "text", "text": "What cant you hear and see in this video?"},
+        ],
+    },
+]
+
+inputs = processor.apply_chat_template(
+    conversations,
+    load_audio_from_video=True,
+    add_generation_prompt=True,
+    tokenize=True,
+    return_dict=True,
+    return_tensors="pt",
+    video_fps=1,
+
+    # kwargs to be passed to `Qwen2-5-OmniProcessor`
+    padding=True,
+    use_audio_in_video=True,
+).to(model.device)
+
+# Generation params for audio or text can be different and have to be prefixed with `thinker_` or `talker_`
+text_ids, audio = model.generate(**inputs, use_audio_in_video=True, thinker_do_sample=False, talker_do_sample=True)
+text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
+
+sf.write(
+    "output.wav",
+    audio.reshape(-1).detach().cpu().numpy(),
+    samplerate=24000,
+)
+print(text)
+```
+
+### Text-only generation
+
+To generate only text output and save compute by not loading the audio generation model, we can use `Qwen2_5OmniThinkerForConditionalGeneration` model.  
+
+```python
+from transformers import Qwen2_5OmniThinkerForConditionalGeneration, Qwen2_5OmniProcessor
+
+model = Qwen2_5OmniThinkerForConditionalGeneration.from_pretrained(
+    "Qwen/Qwen2.5-Omni-7B",
+    torch_dtype="auto",
+    device_map="auto",
+)
+processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")
+
+conversation = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
+        ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "video", "video": "/path/to/video.mp4"},
+            {"type": "text", "text": "What cant you hear and see in this video?"},
+        ],
+    },
+]
+
+inputs = processor.apply_chat_template(
+    conversations,
+    load_audio_from_video=True,
+    add_generation_prompt=True,
+    tokenize=True,
+    return_dict=True,
+    return_tensors="pt",
+    video_fps=1,
+
+    # kwargs to be passed to `Qwen2-5-OmniProcessor`
+    padding=True,
+    use_audio_in_video=True,
+).to(model.device)
+
+
+text_ids = model.generate(**inputs, use_audio_in_video=True)
+text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
+
+sf.write(
+    "output.wav",
+    audio.reshape(-1).detach().cpu().numpy(),
+    samplerate=24000,
+)
+print(text)
+```
+
+### Batch Mixed Media Inference
+
+The model can batch inputs composed of mixed samples of various types such as text, images, audio and videos as input when using `Qwen2_5OmniThinkerForConditionalGeneration` model. Here is an example.
+
+```python
+import soundfile as sf
+from transformers import Qwen2_5OmniForConditionalGeneration, Qwen2_5OmniProcessor
+
+model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
+    "Qwen/Qwen2.5-Omni-7B",
+    torch_dtype="auto",
+    device_map="auto"
+)
+processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")
+
+# Conversation with video only
+conversation1 = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
+        ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "video", "path": "/path/to/video.mp4"},
+        ]
+    }
+]
+
+# Conversation with audio only
+conversation2 = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
+        ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "audio", "path": "/path/to/audio.wav"},
+        ]
+    }
+]
+
+# Conversation with pure text
+conversation3 = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
+        ],
+    },
+    {
+        "role": "user",
+        "content": [{"type": "text", "text": "who are you?"}],
+    }
+]
+
+
+# Conversation with mixed media
+conversation4 = [
+    {
+        "role": "system",
+        "content": [
+            {"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
+        ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "image", "path": "/path/to/image.jpg"},
+            {"type": "video", "path": "/path/to/video.mp4"},
+            {"type": "audio", "path": "/path/to/audio.wav"},
+            {"type": "text", "text": "What are the elements can you see and hear in these medias?"},
+        ],
+    }
+]
+
+conversations = [conversation1, conversation2, conversation3, conversation4]
+
+inputs = processor.apply_chat_template(
+    conversations,
+    load_audio_from_video=True,
+    add_generation_prompt=True,
+    tokenize=True,
+    return_dict=True,
+    return_tensors="pt",
+    video_fps=1,
+
+    # kwargs to be passed to `Qwen2-5-OmniProcessor`
+    padding=True,
+    use_audio_in_video=True,
+).to(model.thinker.device)
+
+text_ids = model.generate(**inputs, use_audio_in_video=True)
+text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
+
+print(text)
+```
+
+### Usage Tips
+
+#### Image Resolution trade-off
+
+The model supports a wide range of resolution inputs. By default, it uses the native resolution for input, but higher resolutions can enhance performance at the cost of more computation. Users can set the minimum and maximum number of pixels to achieve an optimal configuration for their needs.
+
+```python
+min_pixels = 128*28*28
+max_pixels = 768*28*28
+processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B", min_pixels=min_pixels, max_pixels=max_pixels)
+```
+
+#### Prompt for audio output
+If users need audio output, the system prompt must be set as "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.", otherwise the audio output may not work as expected.
+```
+{
+    "role": "system",
+    "content": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.",
+}
+```
+
+#### Use audio output or not
+
+The model supports both text and audio outputs, if users do not need audio outputs, they can set `enable_audio_output` in the `from_pretrained` function. This option will save about `~2GB` of GPU memory but the `return_audio` option for `generate` function will only allow to be set at `False`.
+```python
+model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
+    "Qwen/Qwen2.5-Omni-7B",
+    torch_dtype="auto",
+    device_map="auto",
+    enable_audio_output=False,
+)
+```
+
+In order to obtain a flexible experience, we recommend that users set `enable_audio_output` at `True` when initializing the model through `from_pretrained` function, and then decide whether to return audio when `generate` function is called. When `return_audio` is set to `False`, the model will only return text outputs to get text responses faster.
+
+```python
+model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
+    "Qwen/Qwen2.5-Omni-7B",
+    torch_dtype="auto",
+    device_map="auto",
+    enable_audio_output=True,
+)
+...
+text_ids = model.generate(**inputs, return_audio=False)
+```
+
+#### Change voice type of output audio
+Qwen2.5-Omni supports the ability to change the voice of the output audio. Users can use the `spk` parameter of `generate` function to specify the voice type. The `"Qwen/Qwen2.5-Omni-7B"` checkpoint support two voice types: `Chelsie` and `Ethan`, while `Chelsie` is a female voice and `Ethan` is a male voice. By defalut, if `spk` is not specified, the default voice type is `Chelsie`.
+
+```python
+text_ids, audio = model.generate(**inputs, spk="Chelsie")
+```
+
+```python
+text_ids, audio = model.generate(**inputs, spk="Ethan")
+```
+
+#### Flash-Attention 2 to speed up generation
+
+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 hardware that is compatible with FlashAttention 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 FlashAttention-2, add `attn_implementation="flash_attention_2"` when loading the model:
+
+```python
+from transformers import Qwen2_5OmniForConditionalGeneration
+
+model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
+    "Qwen/Qwen2.5-Omni-7B",
+    device_map="auto",
+    torch_dtype=torch.bfloat16,
+    attn_implementation="flash_attention_2",
+)
+```
+
+
+
+## Qwen2_5OmniConfig
+
+[[autodoc]] Qwen2_5OmniConfig
+
+## Qwen2_5OmniProcessor
+
+[[autodoc]] Qwen2_5OmniProcessor
+
+## Qwen2_5OmniForConditionalGeneration
+
+[[autodoc]] Qwen2_5OmniForConditionalGeneration
+    - forward
+
+## Qwen2_5OmniPreTrainedModelForConditionalGeneration
+
+[[autodoc]] Qwen2_5OmniPreTrainedModelForConditionalGeneration
+
+## Qwen2_5OmniThinkerConfig
+
+[[autodoc]] Qwen2_5OmniThinkerConfig
+
+## Qwen2_5OmniThinkerForConditionalGeneration
+
+[[autodoc]] Qwen2_5OmniThinkerForConditionalGeneration
+
+## Qwen2_5OmniThinkerTextModel
+
+[[autodoc]] Qwen2_5OmniThinkerTextModel
+
+## Qwen2_5OmniTalkerConfig
+
+[[autodoc]] Qwen2_5OmniTalkerConfig
+
+## Qwen2_5OmniTalkerForConditionalGeneration
+
+[[autodoc]] Qwen2_5OmniTalkerForConditionalGeneration
+
+## Qwen2_5OmniTalkerModel
+
+[[autodoc]] Qwen2_5OmniTalkerModel
+
+## Qwen2_5OmniToken2WavConfig
+
+[[autodoc]] Qwen2_5OmniToken2WavConfig
+
+## Qwen2_5OmniToken2WavModel
+
+[[autodoc]] Qwen2_5OmniToken2WavModel
+
+## Qwen2_5OmniToken2WavDiTModel
+
+[[autodoc]] Qwen2_5OmniToken2WavDiTModel
+
+## Qwen2_5OmniToken2WavBigVGANModel
+
+[[autodoc]] Qwen2_5OmniToken2WavBigVGANModel

docs/source/en/model_doc/qwen2_5_vl.md

@@ -232,10 +232,15 @@ model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
 
 [[autodoc]] Qwen2_5_VLConfig
 
+## Qwen2_5_VLTextConfig
+
+[[autodoc]] Qwen2_5_VLTextConfig
+
 ## Qwen2_5_VLProcessor
 
 [[autodoc]] Qwen2_5_VLProcessor
 
+
 ## Qwen2_5_VLModel
 
 [[autodoc]] Qwen2_5_VLModel

docs/source/en/model_doc/qwen2_vl.md

@@ -278,6 +278,10 @@ model = Qwen2VLForConditionalGeneration.from_pretrained(
 
 [[autodoc]] Qwen2VLConfig
 
+## Qwen2VLTextConfig
+
+[[autodoc]] Qwen2VLTextConfig
+
 ## Qwen2VLImageProcessor
 
 [[autodoc]] Qwen2VLImageProcessor

docs/source/en/model_doc/timesfm.md

@@ -0,0 +1,88 @@
+<!--Copyright 2025 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.
+
+-->
+
+# TimesFM
+
+<div class="flex flex-wrap space-x-1">
+<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+</div>
+
+## Overview
+
+TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model proposed in [A decoder-only foundation model for time-series forecasting](https://huggingface.co/papers/2310.10688) by Abhimanyu Das, Weihao Kong, Rajat Sen, and  Yichen Zhou. It is a decoder only model that uses non-overlapping patches of time-series data as input and outputs some output patch length prediction in an autoregressive fashion.
+
+
+The abstract from the paper is the following:
+
+*Motivated by recent advances in large language models for Natural Language Processing (NLP), we design a time-series foundation model for forecasting whose out-of-the-box zero-shot performance on a variety of public datasets comes close to the accuracy of state-of-the-art supervised forecasting models for each individual dataset. Our model is based on pretraining a patched-decoder style attention model on a large time-series corpus, and can work well across different forecasting history lengths, prediction lengths and temporal granularities.*
+
+
+This model was contributed by [kashif](https://huggingface.co/kashif).
+The original code can be found [here](https://github.com/google-research/timesfm).
+
+
+To use the model:
+
+```python
+import torch
+from transformers import TimesFmModelForPrediction
+
+
+model = TimesFmModelForPrediction.from_pretrained(
+    "google/timesfm-2.0-500m-pytorch",
+    torch_dtype=torch.bfloat16,
+    attn_implementation="sdpa",
+    device_map="cuda" if torch.cuda.is_available() else None
+)
+
+
+ # Create dummy inputs
+forecast_input = [
+    np.sin(np.linspace(0, 20, 100)),
+    np.sin(np.linspace(0, 20, 200)),
+    np.sin(np.linspace(0, 20, 400)),
+]
+frequency_input = [0, 1, 2]
+
+# Convert inputs to sequence of tensors
+forecast_input_tensor = [
+    torch.tensor(ts, dtype=torch.bfloat16).to("cuda" if torch.cuda.is_available() else "cpu")
+    for ts in forecast_input
+]
+frequency_input_tensor = torch.tensor(frequency_input, dtype=torch.long).to(
+    "cuda" if torch.cuda.is_available() else "cpu"
+)
+
+# Get predictions from the pre-trained model
+with torch.no_grad():
+    outputs = model(past_values=forecast_input_tensor, freq=frequency_input_tensor, return_dict=True)
+    point_forecast_conv = outputs.mean_predictions.float().cpu().numpy()
+    quantile_forecast_conv = outputs.full_predictions.float().cpu().numpy()
+```
+
+## TimesFmConfig
+
+[[autodoc]] TimesFmConfig
+
+## TimesFmModel
+
+[[autodoc]] TimesFmModel
+    - forward
+
+## TimesFmModelForPrediction
+
+[[autodoc]] TimesFmModelForPrediction
+    - forward

docs/source/en/model_doc/vits.md

@@ -7,168 +7,139 @@ 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.
--->
+specific language governing permissions and limitations under the License.-->
+
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
+</div>
 
 # VITS
 
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-</div>
+[VITS (Variational Inference with adversarial learning for end-to-end Text-to-Speech)](https://hf.co/papers/2106.06103) is a end-to-end speech synthesis model, simplifying the traditional two-stage text-to-speech (TTS) systems. It's unique because it directly synthesizes speech from text using variational inference, adversarial learning, and normalizing flows to produce natural and expressive speech with diverse rhythms and intonations.
 
-## Overview
+You can find all the original VITS checkpoints under the [AI at Meta](https://huggingface.co/facebook?search_models=mms-tts) organization.
 
-The VITS model was proposed in [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
+> [!TIP]
+> Click on the VITS models in the right sidebar for more examples of how to apply VITS.
 
-VITS (**V**ariational **I**nference with adversarial learning for end-to-end **T**ext-to-**S**peech) is an end-to-end 
-speech synthesis model that predicts a speech waveform conditional on an input text sequence. It is a conditional variational 
-autoencoder (VAE) comprised of a posterior encoder, decoder, and conditional prior.
+The example below demonstrates how to generate text based on an image with [`Pipeline`] or the [`AutoModel`] class.
 
-A set of spectrogram-based acoustic features are predicted by the flow-based module, which is formed of a Transformer-based
-text encoder and multiple coupling layers. The spectrogram is decoded using a stack of transposed convolutional layers,
-much in the same style as the HiFi-GAN vocoder. Motivated by the one-to-many nature of the TTS problem, where the same text 
-input can be spoken in multiple ways, the model also includes a stochastic duration predictor, which allows the model to 
-synthesise speech with different rhythms from the same input text. 
-
-The model is trained end-to-end with a combination of losses derived from variational lower bound and adversarial training. 
-To improve the expressiveness of the model, normalizing flows are applied to the conditional prior distribution. During 
-inference, the text encodings are up-sampled based on the duration prediction module, and then mapped into the 
-waveform using a cascade of the flow module and HiFi-GAN decoder. Due to the stochastic nature of the duration predictor,
-the model is non-deterministic, and thus requires a fixed seed to generate the same speech waveform.
-
-The abstract from the paper is the following:
-
-*Several recent end-to-end text-to-speech (TTS) models enabling single-stage training and parallel sampling have been proposed, but their sample quality does not match that of two-stage TTS systems. In this work, we present a parallel end-to-end TTS method that generates more natural sounding audio than current two-stage models. Our method adopts variational inference augmented with normalizing flows and an adversarial training process, which improves the expressive power of generative modeling. We also propose a stochastic duration predictor to synthesize speech with diverse rhythms from input text. With the uncertainty modeling over latent variables and the stochastic duration predictor, our method expresses the natural one-to-many relationship in which a text input can be spoken in multiple ways with different pitches and rhythms. A subjective human evaluation (mean opinion score, or MOS) on the LJ Speech, a single speaker dataset, shows that our method outperforms the best publicly available TTS systems and achieves a MOS comparable to ground truth.*
-
-This model can also be used with TTS checkpoints from [Massively Multilingual Speech (MMS)](https://arxiv.org/abs/2305.13516) 
-as these checkpoints use the same architecture and a slightly modified tokenizer.
-
-This model was contributed by [Matthijs](https://huggingface.co/Matthijs) and [sanchit-gandhi](https://huggingface.co/sanchit-gandhi). The original code can be found [here](https://github.com/jaywalnut310/vits).
-
-## Usage examples
-
-Both the VITS and MMS-TTS checkpoints can be used with the same API. Since the flow-based model is non-deterministic, it 
-is good practice to set a seed to ensure reproducibility of the outputs. For languages with a Roman alphabet, 
-such as English or French, the tokenizer can be used directly to pre-process the text inputs. The following code example 
-runs a forward pass using the MMS-TTS English checkpoint:
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
 ```python
 import torch
-from transformers import VitsTokenizer, VitsModel, set_seed
+from transformers import pipeline, set_seed
+from scipy.io.wavfile import write
 
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
-model = VitsModel.from_pretrained("facebook/mms-tts-eng")
+set_seed(555)
 
-inputs = tokenizer(text="Hello - my dog is cute", return_tensors="pt")
+pipe = pipeline(
+    task="text-to-speech",
+    model="facebook/mms-tts-eng",
+    torch_dtype=torch.float16,
+    device=0
+)
 
-set_seed(555)  # make deterministic
+speech = pipe("Hello, my dog is cute")
+
+# Extract audio data and sampling rate
+audio_data = speech["audio"]
+sampling_rate = speech["sampling_rate"]
+
+# Save as WAV file
+write("hello.wav", sampling_rate, audio_data.squeeze())
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+
+```python
+import torch
+import scipy
+from IPython.display import Audio
+from transformers import AutoTokenizer, VitsModel, set_seed
+
+tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
+model = VitsModel.from_pretrained("facebook/mms-tts-eng", torch_dtype=torch.float16).to("cuda")
+inputs = tokenizer("Hello, my dog is cute", return_tensors="pt").to("cuda")
+
+set_seed(555)
 
 with torch.no_grad():
-   outputs = model(**inputs)
+    outputs = model(**inputs)
 
 waveform = outputs.waveform[0]
-```
-
-The resulting waveform can be saved as a `.wav` file:
-
-```python
-import scipy
-
-scipy.io.wavfile.write("techno.wav", rate=model.config.sampling_rate, data=waveform)
-```
-
-Or displayed in a Jupyter Notebook / Google Colab:
-
-```python
-from IPython.display import Audio
+scipy.io.wavfile.write("hello.wav", rate=model.config.sampling_rate, data=waveform)
 
+# display in Colab notebook
 Audio(waveform, rate=model.config.sampling_rate)
 ```
 
-For certain languages with a non-Roman alphabet, such as Arabic, Mandarin or Hindi, the [`uroman`](https://github.com/isi-nlp/uroman) 
-perl package is required to pre-process the text inputs to the Roman alphabet.
+</hfoption>
+</hfoptions>
 
-You can check whether you require the `uroman` package for your language by inspecting the `is_uroman` attribute of 
-the pre-trained `tokenizer`:
+## Notes
 
-```python
-from transformers import VitsTokenizer
+- Set a seed for reproducibility because VITS synthesizes speech non-deterministically.
+- For languages with non-Roman alphabets (Korean, Arabic, etc.), install the [uroman](https://github.com/isi-nlp/uroman) package to preprocess the text inputs to the Roman alphabet. You can check if the tokenizer requires uroman as shown below.
 
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
-print(tokenizer.is_uroman)
-```
-If the is_uroman attribute is `True`, the tokenizer will automatically apply the `uroman` package to your text inputs, but you need to install uroman if not already installed using:  
-```
-pip install --upgrade uroman
-```
-Note: Python version required to use `uroman` as python package should be >= `3.10`. 
-You can use the tokenizer as usual without any additional preprocessing steps:
-```python
-import torch
-from transformers import VitsTokenizer, VitsModel, set_seed
-import os
-import subprocess
+   ```py
+   # pip install -U uroman
+   from transformers import VitsTokenizer
 
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-kor")
-model = VitsModel.from_pretrained("facebook/mms-tts-kor")
-text = "이봐 무슨 일이야"
-inputs = tokenizer(text=text, return_tensors="pt")
+   tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-eng")
+   print(tokenizer.is_uroman)
+   ```
 
-set_seed(555)  # make deterministic
-with torch.no_grad():
-   outputs = model(inputs["input_ids"])
+   If your language requires uroman, the tokenizer automatically applies it to the text inputs. Python >= 3.10 doesn't require any additional preprocessing steps. For Python < 3.10, follow the steps below.
 
-waveform = outputs.waveform[0]
-```
-If you don't want to upgrade to python >= `3.10`, then you can use the `uroman` perl package to pre-process the text inputs to the Roman alphabet.
-To do this, first clone the uroman repository to your local machine and set the bash variable `UROMAN` to the local path:
+   ```bash
+   git clone https://github.com/isi-nlp/uroman.git
+   cd uroman
+   export UROMAN=$(pwd)
+   ```
 
+   Create a function to preprocess the inputs. You can either use the bash variable `UROMAN` or pass the directory path directly to the function.
 
-```bash
-git clone https://github.com/isi-nlp/uroman.git
-cd uroman
-export UROMAN=$(pwd)
-```
+   ```py
+   import torch
+   from transformers import VitsTokenizer, VitsModel, set_seed
+   import os
+   import subprocess
 
-You can then pre-process the text input using the following code snippet. You can either rely on using the bash variable 
-`UROMAN` to point to the uroman repository, or you can pass the uroman directory as an argument to the `uromanize` function:
+   tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-kor")
+   model = VitsModel.from_pretrained("facebook/mms-tts-kor")
 
-```python
-import torch
-from transformers import VitsTokenizer, VitsModel, set_seed
-import os
-import subprocess
+   def uromanize(input_string, uroman_path):
+       """Convert non-Roman strings to Roman using the `uroman` perl package."""
+       script_path = os.path.join(uroman_path, "bin", "uroman.pl")
 
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-kor")
-model = VitsModel.from_pretrained("facebook/mms-tts-kor")
+       command = ["perl", script_path]
 
-def uromanize(input_string, uroman_path):
-    """Convert non-Roman strings to Roman using the `uroman` perl package."""
-    script_path = os.path.join(uroman_path, "bin", "uroman.pl")
+       process = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+       # Execute the perl command
+       stdout, stderr = process.communicate(input=input_string.encode())
 
-    command = ["perl", script_path]
+       if process.returncode != 0:
+           raise ValueError(f"Error {process.returncode}: {stderr.decode()}")
 
-    process = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-    # Execute the perl command
-    stdout, stderr = process.communicate(input=input_string.encode())
+       # Return the output as a string and skip the new-line character at the end
+       return stdout.decode()[:-1]
 
-    if process.returncode != 0:
-        raise ValueError(f"Error {process.returncode}: {stderr.decode()}")
+   text = "이봐 무슨 일이야"
+   uromanized_text = uromanize(text, uroman_path=os.environ["UROMAN"])
 
-    # Return the output as a string and skip the new-line character at the end
-    return stdout.decode()[:-1]
+   inputs = tokenizer(text=uromanized_text, return_tensors="pt")
 
-text = "이봐 무슨 일이야"
-uromanized_text = uromanize(text, uroman_path=os.environ["UROMAN"])
+   set_seed(555)  # make deterministic
+   with torch.no_grad():
+      outputs = model(inputs["input_ids"])
 
-inputs = tokenizer(text=uromanized_text, return_tensors="pt")
-
-set_seed(555)  # make deterministic
-with torch.no_grad():
-   outputs = model(inputs["input_ids"])
-
-waveform = outputs.waveform[0]
-```
+   waveform = outputs.waveform[0]
+   ```
 
 ## VitsConfig
 
@@ -177,10 +148,11 @@ waveform = outputs.waveform[0]
 ## VitsTokenizer
 
 [[autodoc]] VitsTokenizer
-    - __call__
-    - save_vocabulary
+- __call__
+- save_vocabulary
 
 ## VitsModel
 
 [[autodoc]] VitsModel
-    - forward
+- forward
+

docs/source/en/model_doc/yolos.md

@@ -92,6 +92,11 @@ Use [`YolosImageProcessor`] for preparing images (and optional targets) for the
 
 [[autodoc]] YolosImageProcessor
     - preprocess
+
+## YolosImageProcessorFast
+
+[[autodoc]] YolosImageProcessorFast
+    - preprocess
     - pad
     - post_process_object_detection
 

docs/source/en/perf_train_gpu_many.md

@@ -111,7 +111,7 @@ This approach optimizes parallel data processing by reducing idle GPU utilizatio
 
 Data, pipeline and model parallelism combine to form [3D parallelism](https://www.microsoft.com/en-us/research/blog/deepspeed-extreme-scale-model-training-for-everyone/) to optimize memory and compute efficiency.
 
-Memory effiiciency is achieved by splitting the model across GPUs and also dividing it into stages to create a pipeline. This allows GPUs to work in parallel on micro-batches of data, reducing the memory usage of the model, optimizer, and activations.
+Memory efficiency is achieved by splitting the model across GPUs and also dividing it into stages to create a pipeline. This allows GPUs to work in parallel on micro-batches of data, reducing the memory usage of the model, optimizer, and activations.
 
 Compute efficiency is enabled by ZeRO data parallelism where each GPU only stores a slice of the model, optimizer, and activations. This allows higher communication bandwidth between data parallel nodes because communication can occur independently or in parallel with the other pipeline stages.
 

docs/source/en/pipeline_webserver.md

@@ -52,10 +52,10 @@ async def homepage(request):
     return JSONResponse(output)
 
 async def server_loop(q):
-    pipeline = pipeline(task="fill-mask",model="google-bert/bert-base-uncased")
+    pipe = pipeline(task="fill-mask",model="google-bert/bert-base-uncased")
     while True:
         (string, response_q) = await q.get()
-        out = pipeline(string)
+        out = pipe(string)
         await response_q.put(out)
 
 app = Starlette(
@@ -81,6 +81,10 @@ Query the server with a POST request.
 
 ```bash
 curl -X POST -d "Paris is the [MASK] of France." http://localhost:8000/
+```
+This should return the output below.
+
+```bash
 [{'score': 0.9969332218170166,
   'token': 3007,
   'token_str': 'capital',
@@ -112,23 +116,27 @@ The example below is written in pseudocode for readability rather than performan
 1. There is no batch size limit.
 2. The timeout is reset on every queue fetch, so you could end up waiting much longer than the `timeout` value before processing a request. This would also delay the first inference request by that amount of time. The web server always waits 1ms even if the queue is empty, which is inefficient, because that time can be used to start inference. It could make sense though if batching is essential to your use case.
 
-    It would be better to have a single 1ms deadline, instead of resetting it on every fetch.
+    It would be better to have a single 1ms deadline, instead of resetting it on every fetch, as shown below.
 
 ```py
-(string, rq) = await q.get()
-strings = []
-queues = []
-while True:
-    try:
-        (string, rq) = await asyncio.wait_for(q.get(), timeout=0.001)
-    except asyncio.exceptions.TimeoutError:
-        break
-    strings.append(string)
-    queues.append(rq)
-strings
-outs = pipeline(strings, batch_size=len(strings))
-for rq, out in zip(queues, outs):
-    await rq.put(out)
+async def server_loop(q):
+    pipe = pipeline(task="fill-mask", model="google-bert/bert-base-uncased")
+    while True:
+        (string, rq) = await q.get()
+        strings = []
+        queues = []
+        strings.append(string)
+        queues.append(rq)
+        while True:
+            try:
+                (string, rq) = await asyncio.wait_for(q.get(), timeout=1)
+            except asyncio.exceptions.TimeoutError:
+                break
+            strings.append(string)
+            queues.append(rq)
+        outs = pipe(strings, batch_size=len(strings))
+        for rq, out in zip(queues, outs):
+            await rq.put(out)
 ```
 
 ## Error checking

docs/source/en/quantization/concept_guide.md

@@ -0,0 +1,178 @@
+<!--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.
+
+-->
+
+# Quantization concepts
+
+Quantization reduces the memory footprint and computational cost of large machine learning models like those found in the Transformers library. It achieves this by representing the model's weights and or activations with lower-precision data types (like 8-bit integers or int8) instead of the standard 32-bit floating-point (float32).
+
+
+Reducing a model's precision offers several significant benefits:
+
+-  Smaller model size: Lower-precision data types require less storage space. An int8 model, for example, is roughly 4 times smaller than its float32 counterpart.
+-  Faster inference: Operations on lower-precision data types, especially integers, can be significantly faster on compatible hardware (CPUs and GPUs often have specialized instructions for int8 operations). This leads to lower latency.
+-  Reduced energy consumption: Faster computations and smaller memory transfers often translate to lower power usage.
+
+The primary trade-off in quantization is *efficiency* vs. *accuracy*. Reducing precision saves resources but inevitably introduces small errors (quantization noise). The goal is to minimize this error using appropriate schemes (affine/symmetric), granularity (per-tensor/channel), and techniques (PTQ/QAT) so that the model's performance on its target task degrades as little as possible.
+
+The sections below cover quantization schemes, granularity, and techniques.
+
+## Quantization schemes
+
+The core idea is to map the range of values found in the original float32 weights and activations to the much smaller range represented by int8 (typically \\([-128, 127]\\)).
+
+This section covers how some quantization techniques work.
+
+<div class="flex justify-center">
+    <img width="606" alt="quant_visual" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/quant_visual.png" />
+</div>
+
+### Affine quantization
+
+The most common method is *affine quantization*. For a given float32 tensor (like a layer's weights), it finds the minimum \\(val_{min}\\) and maximum \\(val_{max}\\) values. This range \\([val_{min}, val_{max}]\\) is mapped to the int8 range \\([q_{min}, q_{max}]\\), which is typically \\([-128, 127]\\).
+
+There are two main ways to perform this mapping, *symmetric* and *asymmetric*. The choice between symmetric and asymmetric quantization determines how the float32 range is mapped to the int8 range.
+
+- Symmetric: This method assumes the original float32 range is symmetric around zero ( \\([ -a, a ]\\) ). This range is mapped symmetrically to the int8 range, for example, \\([-127, 127]\\). A key characteristic is that the float32 value \\(0.0\\) maps directly to the int8 value \\(0\\). This only requires one parameter, the **scale ( \\(S\\) )**, to define the mapping. It can simplify computations, but it might be less accurate if the original data distribution isn't naturally centered around zero.
+- Asymmetric (Affine): This method does not assume the data is centered around zero. It maps the exact range \\([val_{min}, val_{max}]\\) from float32 to the full int8 range, like \\([-128, 127]\\). This requires two parameters, a **scale ( \\(S\\) )** and a **zero-point ( \\(Z\\) )**. 
+
+
+    scale ( \\(S\\) ): A positive float32 number representing the ratio between the float32 and the int8 range.
+
+$$
+S = \frac{val_{max} - val_{min}}{q_{max} - q_{min}}
+$$
+
+zero-Point ( \\(Z\\) ): An int8 value that corresponds to the float32 value \\(0.0\\).
+
+$$
+Z = q_{min} - round\left(\frac{val_{min}}{S}\right)
+$$
+
+> [!TIP]
+> In symmetric quantization, Z would typically be fixed at 0.
+
+With these parameters, a float32 value, \\(x\\). can be quantized to int8 ( \\(q\\) ) with the formula below.
+
+$$
+q = round\left(\frac{x}{S} + Z\right)
+$$
+
+The int8 value, \\(q\\), can be dequantized back to approximate float32 with the formula below.
+
+$$
+x \approx S \cdot (q - Z)
+$$
+
+<div class="flex justify-center">
+    <img width="606" alt="dequant" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/dequant.png" />
+</div>
+
+During inference, computations like matrix multiplication are performed using the int8 values ( \\(q\\) ), and the result is dequantized back to float32 (often using a higher-precision accumulation type like int32 internally) before it is passed to the next layer.
+
+### int4 and weight packing
+
+<div class="flex justify-center">
+    <img width="606" alt="weight packing" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/weight_packing.png" />
+</div>
+
+int4 quantization further reduces the model size and memory usage (halving it compared to int8). The same affine or symmetric quantization principles apply, mapping the float32 range to the 16 possible values representable by int4 ( \\([-8, 7]\\) for signed int4).
+
+A key aspect of int4 quantization is **weight packing**. Since most hardware can't natively handle 4-bit data types in memory, two int4 values are typically packed together into a single int8 byte for storage and transfer. For example, the first value might occupy the lower 4 bits and the second value the upper 4 bits of the byte (`packed_byte = (val1 & 0x0F) | (val2 << 4)`).
+
+int4 is still beneficial even without native int4 compute because the primary benefit comes from reduced memory bandwidth. Loading packed int4 weights (stored as int8) from memory (RAM or VRAM) to the compute units is twice as fast as loading int8 weights. For large models, memory access is often a significant bottleneck. The speed up from faster data transfer can outweigh the computational overhead of unpacking and dequantizing on the fly, leading to overall faster inference, especially in memory-bound scenarios.
+
+However, int4 quantization typically results in a larger accuracy drop compared to int8. Advanced quantization techniques like [GPTQ](./gptq) or [AWQ](./awq) are often necessary for good performance with int4.
+
+### FP8 Quantization (A8W8)
+
+<div class="flex justify-center">
+    <img width="606" alt="fp8" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/fp8.png" />
+</div>
+A newer datatype, 8-bit floating-point (FP8), offers another way to reduce precision while retaining more accuracy than int8 in certain scenarios. FP8 keeps the floating-point structure (sign, exponent, mantissa) but uses fewer bits.
+
+There are two common FP8 variants.
+
+- E4M3: 1 sign bit, 4 exponent bits, 3 mantissa bits. Offers higher precision (more mantissa bits) but a smaller dynamic range (fewer exponent bits).
+- E5M2: 1 sign bit, 5 exponent bits, 2 mantissa bits. Offers a wider dynamic range but lower precision.
+
+FP8 is used in the *A8W8* quantization scheme, which quantizes both activations (A) and weights (W) to 8-bit precision.
+
+While int8 has broad support, efficient FP8 computation requires specific hardware capabilities found in newer GPUs like NVIDIA H100/H200/B100 and AMD Instinct MI300 series. Without native hardware acceleration, the benefits of FP8 might not be fully realized.
+
+Transformers supports FP8 through specific backends like [FBGEMM](./fbgemm_fp8), [FineGrainedFP8](./finegrained_fp8), and [compressed-tensors](./compressed_tensors). These backends handle the underlying FP8 conversion and computation when the appropriate hardware and configurations are used.
+
+## Granularity
+
+Quantization parameters ( \\(S\\) and \\(Z\\)) can be calculated in one of two ways.
+
+- Per-Tensor: One set of \\(S\\) and \\(Z\\) for the entire tensor. Simpler, but less accurate if data values vary greatly within the tensor.
+- Per-Channel (or Per-Group/Block): Separate \\(S\\) and \\(Z\\) for each channel or group. More accurate and better performance at the cost of slightly more complexity and memory.
+
+<div class="flex justify-center">
+    <img width="625" alt="Granularities" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/Granularities.png" />
+</div>
+
+## Quantization techniques
+
+There are two main types of quantization techniques.
+
+- Post-Training Quantization (PTQ): Quantization is applied  *after* the model is fully trained.
+- Quantization-Aware Training (QAT): Quantization effects are simulated *during* training by inserting "fake quantization" ops that simulate the rounding errors of quantization. This lets the model adapt to quantization, and usually results in better accuracy, especially at lower bit-widths.
+
+## Quantization in Transformers
+
+Transformers integrates several quantization backends such as bitsandbytes, torchao, compressed-tensors, and more (refer to the quantization [overview](./overview) for more backends). 
+
+
+All backends are unified under the [`HfQuantizer`] API and associated [`QuantizationConfig`] classes. You can integrate your own custom quantization backends by implementing a custom [`HfQuantizer`] and [`QuantizationConfig`], as shown in the [Contribution](./contribute) guide.
+
+The typical workflow for quantization in Transformers is to:
+
+1. Choose a quantization method suitable for your hardware and use case (see the [Overview](./overview) or [Selecting a quantization method](./selecting) guide to help you).
+2. Load a pre-quantized model from the Hugging Face Hub or load a float32/float16/bfloat16 model and apply a specific quantization method with [`QuantizationConfig`].
+
+The example below demonstrates loading a 8B parameter model and quantizing it to 4-bits with bitsandbytes.
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+
+model_id = "meta-llama/Llama-3.1-8B-Instruct"
+
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    quantization_config=quantization_config,
+    torch_dtype=torch.bfloat16,
+    device_map="auto"
+)
+```
+
+
+## Resources
+
+To explore quantization and related performance optimization concepts more deeply, check out the following resources.
+
+- [Quantization Fundamentals with Hugging Face](https://www.deeplearning.ai/short-courses/quantization-fundamentals-with-hugging-face/)
+- [Quantization in Depth](https://www.deeplearning.ai/short-courses/quantization-in-depth)
+- [Introduction to Quantization cooked in 🤗 with 💗🧑‍🍳](https://huggingface.co/blog/merve/quantization)
+- [EfficientML.ai Lecture 5 - Quantization Part I](https://www.youtube.com/watch?v=RP23-dRVDWM)
+- [Making Deep Learning Go Brrrr From First Principles](https://horace.io/brrr_intro.html)
+- [Accelerating Generative AI with PyTorch Part 2: LLM Optimizations](https://pytorch.org/blog/accelerating-generative-ai-2/)

docs/source/en/quantization/selecting.md

@@ -0,0 +1,157 @@
+<!--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.
+
+-->
+
+# Selecting a quantization method
+
+There are many quantization methods available in Transformers for inference and fine-tuning. This guide helps you choose the most common and production-ready quantization techniques depending on your use case, and presents the advantages and disadvantages of each technique.
+
+For a comprehensive overview of all supported methods and their features, refer back to the table in the [Overview](./overview).
+
+## Inference
+
+Consider the quantization methods below for inference.
+
+| quantization method | use case |
+|---|---|
+| bitsandbytes | ease of use and QLoRA fine-tuning on NVIDIA GPUs |
+| compressed-tensors | loading specific quantized formats (FP8, Sparse) |
+| GPTQModel or AWQ | good 4-bit accuracy with upfront calibration |
+| HQQ | fast on the fly quantization without calibration |
+| torchao | flexibility and fast inference with torch.compile |
+
+### No Calibration Required (On-the-fly Quantization)
+
+These methods are generally easier to use as they don't need a separate calibration dataset or step.
+
+#### bitsandbytes
+
+| Pros                                                         | Cons                                                    |
+|--------------------------------------------------------------|---------------------------------------------------------|
+| Very simple, no calibration dataset required for inference.  | Primarily optimized for NVIDIA GPUs (CUDA).             |
+| Good community support and widely adopted.                   | Inference speedup isn't guaranteed.                     |
+
+See the [bitsandbytes documentation](./bitsandbytes) for more details.
+
+#### HQQ (Half-Quadratic Quantization)
+
+| Pros                                                                 | Cons                                                                       |
+|----------------------------------------------------------------------|----------------------------------------------------------------------------|
+| Fast quantization process, no calibration data needed.              | Accuracy can degrade significantly at bit depths <4-bit.                     |
+| Multiple backends for fast inference.                                | Inference speed may not match others unless using `torch.compile` or backends. |
+| Compatible with `torch.compile`.                                     |                                                                            |
+| Supports wide range of bit depths (8, 4, 3, 2, 1-bit).              |                                                                            |
+
+See the [HQQ documentation](./hqq) for more details.
+
+#### torchao
+
+| Pros                                                                 | Cons                                                                 |
+|----------------------------------------------------------------------|----------------------------------------------------------------------|
+| Strong integration with `torch.compile` for potential speedups.     | Newer library, ecosystem still evolving.                             |
+| Offers decent CPU quantization support.                              | Performance depends on `torch.compile` working well.                 |
+| Flexibility in quantization schemes (int8, int4, fp8).           | 4-bit quantization (int4wo) may not match GPTQ/AWQ in accuracy.              |
+
+See the [torchao documentation](./torchao) for more details.
+
+### Calibration-based Quantization
+
+These methods require an upfront calibration step using a dataset to potentially achieve higher accuracy.
+
+#### GPTQ/GPTQModel
+
+Calibration for 8B model takes ~20 minutes on one A100 gpu.
+
+| Pros                                                                 | Cons                                                                 |
+|----------------------------------------------------------------------|----------------------------------------------------------------------|
+| Often achieves high accuracy.                                        | Requires a calibration dataset and a separate calibration step.      |
+| Can lead to inference speedups.                                      | Possible to overfit on calibration data.                             |
+| Many pre-quantized GPTQ models on [Hugging Face Hub](https://huggingface.co/models?other=gptq). |                                           |
+
+See the [GPTQ documentation](./gptq) for more details.
+
+#### AWQ (Activation-aware Weight Quantization)
+
+Calibration for 8B model takes ~10 minutes on one A100 gpu.
+
+| Pros                                                                 | Cons                                                |
+|----------------------------------------------------------------------|-----------------------------------------------------|
+| Often achieves high accuracy at 4-bit. (Sometimes surpasses GPTQ on specific tasks.) | Requires calibration if quantizing yourself.        |
+| Can lead to inference speedups.                                      |                                                     |
+| Shorter calibration time than GPTQ.                                  |                                                     |
+| Many pre-quantized AWQ models on [Hugging Face Hub](https://huggingface.co/models?other=awq). |                                                     |
+
+See the [AWQ documentation](./awq) for more details.
+
+### Loading Specific Formats
+
+#### compressed-tensors
+
+| Pros                                                         | Cons                                                        |
+|--------------------------------------------------------------|-------------------------------------------------------------|
+| Supports flexible formats including FP8 and sparsity.        | Primarily for loading pre-quantized models.                 |
+|                                                              | Doesn't perform quantization within Transformers directly.  |
+
+See the [compressed-tensors documentation](./compressed_tensors) for more details.
+
+## Fine-tuning
+
+Consider the quantization method below during fine-tuning to save memory.
+
+### bitsandbytes[[training]]
+
+*   **Description:** The standard method for QLoRA fine-tuning via PEFT.
+*   **Pros:** Enables fine-tuning large models on consumer GPUs; widely supported and documented for PEFT.
+*   **Cons:** Primarily for NVIDIA GPUs.
+
+Other methods offer PEFT compatibility, though bitsandbytes is the most established and straightforward path for QLoRA.
+
+See the [bitsandbytes documentation](./bitsandbytes#qlora) and [PEFT Docs](https://huggingface.co/docs/peft/developer_guides/quantization#aqlm-quantization) for more details. 
+
+## Research
+
+Methods like [AQLM](./aqlm), [SpQR](./spqr), [VPTQ](./vptq), [HIGGS](./higgs), etc., push the boundaries of compression (< 2-bit) or explore novel techniques.
+
+*   Consider these if:
+    *   You need extreme compression (sub-4-bit).
+    *   You are conducting research or require state-of-the-art results from their respective papers.
+    *   You have significant compute resources available for potentially complex quantization procedures.
+We recommend consulting each methods documentation and associated papers carefully before choosing one for use in production.
+
+## Benchmark Comparison
+
+To provide a quantitative comparison of different quantization methods, we benchmarked several popular techniques on the Llama 3.1 8B and 70B models. The following tables show results for accuracy (higher is better), inference throughput measured in tokens/second (higher is better), peak VRAM usage measured in GB (lower is better), and quantization time.
+
+Performance metrics were measured on 2 NVIDIA A100 80GB GPU for Llama 3.1 70B (bfloat16), 1 NVIDIA H100 80GB GPU for FP8 methods, and 1 NVIDIA A100 80GB GPU for all other methods. Throughput was measured with a batch size of 1 and generating 64 tokens.
+Results for `torch.compile` and Marlin kernels are included where applicable and supported.
+
+<iframe
+  src="https://huggingface.co/datasets/derekl35/quantization-benchmarks/embed/viewer/default/train"
+  frameborder="0"
+  width="100%"
+  height="560px"
+  title="benchmarking results dataset"
+></iframe>
+
+The key takeaways are:
+
+| Quantization & Methods                      | Memory Savings (vs bf16) | Accuracy             | Other Notes                                                        |
+|-------------------------------------------- |------------------------- |--------------------- |------------------------------------------------------------------- |
+| **8-bit** (bnb-int8, HQQ, Quanto, torchao, fp8) | ~2x             | Very close to baseline bf16 model   |                                                                    |
+| **4-bit** (AWQ, GPTQ, HQQ, bnb-nf4)    | ~4x                      | Relatively high accuracy            | AWQ/GPTQ often lead in accuracy but need calibration. HQQ/bnb-nf4 are easy on-the-fly. |
+| **Sub-4-bit** (VPTQ, AQLM, 2-bit GPTQ) | Extreme (>4x)            | Noticeable drop, especially at 2-bit | Quantization times can be very long (AQLM, VPTQ). Performance varies. |
+
+> [!TIP]
+> Always benchmark the performance (accuracy and speed) of the quantized model on your specific task and hardware to ensure it meets your requirements. Refer to the individual documentation pages linked above for detailed usage instructions.

docs/source/en/quantization/torchao.md

@@ -11,50 +11,95 @@ rendered properly in your Markdown viewer.
 
 # torchao
 
+[![Open In Colab: Torchao Demo](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/transformers_doc/en/quantization/torchao.ipynb)
+
 [torchao](https://github.com/pytorch/ao) is a PyTorch architecture optimization library with support for custom high performance data types, quantization, and sparsity. It is composable with native PyTorch features such as [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) for even faster inference and training.
 
-Install torchao with the following command.
+See the table below for additional torchao features.
+
+| Feature | Description |
+|--------|-------------|
+| **Quantization Aware Training (QAT)** | Train quantized models with minimal accuracy loss (see [QAT README](https://github.com/pytorch/ao/blob/main/torchao/quantization/qat/README.md)) |
+| **Float8 Training** | High-throughput training with float8 formats (see [torchtitan](https://github.com/pytorch/torchtitan/blob/main/docs/float8.md) and [Accelerate](https://huggingface.co/docs/accelerate/usage_guides/low_precision_training#configuring-torchao) docs) |
+| **Sparsity Support** | Semi-structured (2:4) sparsity for faster inference (see [Accelerating Neural Network Training with Semi-Structured (2:4) Sparsity](https://pytorch.org/blog/accelerating-neural-network-training/) blog post) |
+| **Optimizer Quantization** | Reduce optimizer state memory with 4 and 8-bit variants of Adam |
+| **KV Cache Quantization** | Enables long context inference with lower memory (see [KV Cache Quantization](https://github.com/pytorch/ao/blob/main/torchao/_models/llama/README.md)) |
+| **Custom Kernels Support** | use your own `torch.compile` compatible ops |
+| **FSDP2** | Composable with FSDP2 for training|
+
+> [!TIP]
+> Refer to the torchao [README.md](https://github.com/pytorch/ao#torchao-pytorch-architecture-optimization) for more details about the library.
+
+
+torchao supports the [quantization techniques](https://github.com/pytorch/ao/blob/main/torchao/quantization/README.md) below.
+
+- A16W8 Int8 WeightOnly Quantization
+- A16W4 WeightOnly Quantization
+- A8W8 Int8 Dynamic Quantization
+- A16W8 Float8 WeightOnly Quantization
+- Autoquantization
+
+
+Check the table below to see if your hardware is compatible.
+
+| Component | Compatibility |
+|----------|----------------|
+| CUDA Versions | ✅ cu118, cu124, cu126, cu128 |
+| CPU | ✅ change `device_map="cpu"` (see examples below) |
+
+
+
+Install torchao from PyPi or the PyTorch index with the following commands.
+
+<hfoptions id="install torchao">
+<hfoption id="PyPi">
 
 ```bash
 # Updating 🤗 Transformers to the latest version, as the example script below uses the new auto compilation
-pip install --upgrade torch torchao transformers
+# Stable release from Pypi which will default to CUDA 12.4
+pip install --upgrade torchao transformers
 ```
+</hfoption> 
+<hfoption id="PyTorch Index">
+Stable Release from the PyTorch index
+```bash
+pip install torchao --extra-index-url https://download.pytorch.org/whl/cu124 # options are cpu/cu118/cu124/cu126
+```
+</hfoption>
+</hfoptions>
 
-torchao supports many quantization types for different data types (int4, float8, weight only, etc.).
-Starting with version 0.10.0, torchao provides enhanced flexibility through the `AOBaseConfig` API, allowing for more customized quantization configurations.
-And full access to the techniques offered in the torchao library.
+If your torcha version is below 0.10.0, you need to upgrade it, please refer to the [deprecation notice](#deprecation-notice) for more details.
+
+## Quantization examples
+
+TorchAO provides a variety of quantization configurations. Each configuration can be further customized with parameters such as `group_size`, `scheme`, and `layout` to optimize for specific hardware and model architectures.
+
+For a complete list of available configurations, see the [quantization API documentation](https://github.com/pytorch/ao/blob/main/torchao/quantization/quant_api.py).
 
 You can manually choose the quantization types and settings or automatically select the quantization types.
 
-<hfoptions id="torchao">
-<hfoption id="manual">
+Create a [`TorchAoConfig`] and specify the quantization type and `group_size` of the weights to quantize (for int8 weight only and int4 weight only). Set the `cache_implementation` to `"static"` to automatically [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) the forward method.
 
-
-Create a [`TorchAoConfig`] and specify the quantization type and `group_size` of the weights to quantize. Set the `cache_implementation` to `"static"` to automatically [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) the forward method.
-
-> [!TIP]
-> Run the quantized model on a CPU by changing `device_map` to `"cpu"` and `layout` to `Int4CPULayout()`. This is only available in torchao 0.8.0+.
-
-In torchao 0.10.0+, you can use the more flexible `AOBaseConfig` approach instead of string identifiers:
+<hfoptions id="examples">
+<hfoption id="int8-weight-only cuda">
 
 ```py
 import torch
 from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
-from torchao.quantization import Int4WeightOnlyConfig
+from torchao.quantization import Int8WeightOnlyConfig
 
-# Using AOBaseConfig instance (torchao >= 0.10.0)
-quant_config = Int4WeightOnlyConfig(group_size=128)
+quant_config = Int8WeightOnlyConfig(group_size=128)
 quantization_config = TorchAoConfig(quant_type=quant_config)
 
 # Load and quantize the model
 quantized_model = AutoModelForCausalLM.from_pretrained(
-    "meta-llama/Meta-Llama-3-8B",
+    "meta-llama/Llama-3.1-8B-Instruct",
     torch_dtype="auto",
     device_map="auto",
     quantization_config=quantization_config
 )
 
-tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
 input_text = "What are we having for dinner?"
 input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
 
@@ -62,22 +107,326 @@ input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
 output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
 print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
+</hfoption>
 
-## Available Quantization Schemes
+<hfoption id="int8-weight-only cpu">
 
-TorchAO provides a variety of quantization configurations:
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int8WeightOnlyConfig
 
-- `Int4WeightOnlyConfig`
-- `Int8WeightOnlyConfig`
-- `Int8DynamicActivationInt8WeightConfig`
-- `Float8WeightOnlyConfig`
+quant_config = Int8WeightOnlyConfig(group_size=128)
+quantization_config = TorchAoConfig(quant_type=quant_config)
 
-Each configuration can be further customized with parameters such as `group_size`, `scheme`, and `layout` to optimize for specific hardware and model architectures.
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="cpu",
+    quantization_config=quantization_config
+)
 
-For a complete list of available configurations, see our [quantization API documentation](https://github.com/pytorch/ao/blob/main/torchao/quantization/quant_api.py).
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+<hfoption id="int4-weight-only cuda">
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int4WeightOnlyConfig
+
+quant_config = Int4WeightOnlyConfig(group_size=128)
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="auto",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+
+<hfoption id="int4-weight-only cpu">
+
+> [!TIP]
+> Run the quantized model on a CPU by changing `device_map` to `"cpu"` and `layout` to `Int4CPULayout()`.
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int4WeightOnlyConfig
+from torchao.dtypes import Int4CPULayout
+
+quant_config = Int4WeightOnlyConfig(group_size=128, layout=Int4CPULayout())
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="cpu",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+<hfoption id="int8-dynamic-quantization cuda">
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int8DynamicActivationInt8WeightConfig
+
+quant_config = Int8DynamicActivationInt8WeightConfig()
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="auto",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+<hfoption id="int8-dynamic-quantization cpu">
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int8DynamicActivationInt8WeightConfig
+
+quant_config = Int8DynamicActivationInt8WeightConfig()
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="cpu",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+<hfoption id="float8-weight-only cuda">
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Float8WeightOnlyConfig
+
+quant_config = Float8WeightOnlyConfig()
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="auto",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+
+```
+</hfoption>
+<hfoption id="float8-weight-only cpu">
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Float8WeightOnlyConfig
+
+quant_config = Float8WeightOnlyConfig()
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="cpu",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+</hfoption>
+
+</hfoptions>
+
+### Autoquant
+
+If you want to automatically choose a quantization type for quantizable layers (`nn.Linear`) you can use the [autoquant](https://pytorch.org/ao/stable/generated/torchao.quantization.autoquant.html#torchao.quantization.autoquant) API.
+
+The `autoquant` API automatically chooses a quantization type by micro-benchmarking on input type and shape and compiling a single linear layer.
+
+Create a [`TorchAoConfig`] and set to `"autoquant"`. Set the `cache_implementation` to `"static"` to automatically [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) the forward method. Finally, call `finalize_autoquant` on the quantized model to finalize the quantization and log the input shapes.
+
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+
+quantization_config = TorchAoConfig("autoquant", min_sqnr=None)
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="auto",
+    quantization_config=quantization_config
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+# auto-compile the quantized model with `cache_implementation="static"` to get speed up
+output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
+# explicitly call `finalize_autoquant` (may be refactored and removed in the future)
+quantized_model.finalize_autoquant()
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+
+
+## Serialization
+
+torchao implements [torch.Tensor subclasses](https://pytorch.org/docs/stable/notes/extending.html#subclassing-torch-tensor) for maximum flexibility in supporting new quantized torch.Tensor formats. [Safetensors](https://huggingface.co/docs/safetensors/en/index) serialization and deserialization does not work with torchao.
+
+To avoid arbitrary user code execution, torchao sets `weights_only=True` in [torch.load](https://pytorch.org/docs/stable/generated/torch.load.html) to ensure only tensors are loaded. Any known user functions can be whitelisted with [add_safe_globals](https://pytorch.org/docs/stable/notes/serialization.html#torch.serialization.add_safe_globals).
+
+```py
+# don't serialize model with Safetensors
+output_dir = "llama3-8b-int4wo-128"
+quantized_model.save_pretrained("llama3-8b-int4wo-128", safe_serialization=False)
+```
+
+## Loading quantized models
+
+Loading a quantized model depends on the quantization scheme. For quantization schemes, like int8 and float8, you can quantize the model on any device and also load it on any device. The example below demonstrates quantizing a model on the CPU and then loading it on CUDA.
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int8WeightOnlyConfig
+
+quant_config = Int8WeightOnlyConfig(group_size=128)
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="cpu",
+    quantization_config=quantization_config
+)
+# save the quantized model
+output_dir = "llama-3.1-8b-torchao-int8-cuda"
+quantized_model.save_pretrained(output_dir, safe_serialization=False)
+
+# reload the quantized model
+reloaded_model = AutoModelForCausalLM.from_pretrained(
+    output_dir, 
+    device_map="auto", 
+    torch_dtype=torch.bfloat16
+)
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+output = reloaded_model.generate(**input_ids, max_new_tokens=10)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+
+```
+For int4, the model can only be loaded on the same device it was quantized on because the layout is specific to the device. The example below demonstrates quantizing and loading a model on the CPU.
+
+```py
+import torch
+from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+from torchao.quantization import Int4WeightOnlyConfig
+from torchao.dtypes import Int4CPULayout
+
+quant_config = Int4WeightOnlyConfig(group_size=128, layout=Int4CPULayout())
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+# Load and quantize the model
+quantized_model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/Llama-3.1-8B-Instruct",
+    torch_dtype="auto",
+    device_map="cpu",
+    quantization_config=quantization_config
+)
+# save the quantized model
+output_dir = "llama-3.1-8b-torchao-int4-cpu"
+quantized_model.save_pretrained(output_dir, safe_serialization=False)
+
+# reload the quantized model
+reloaded_model = AutoModelForCausalLM.from_pretrained(
+    output_dir, 
+    device_map="cpu", 
+    torch_dtype=torch.bfloat16
+)
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+input_text = "What are we having for dinner?"
+input_ids = tokenizer(input_text, return_tensors="pt")
+
+output = reloaded_model.generate(**input_ids, max_new_tokens=10)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+
+```
+
+## ⚠️ Deprecation Notice
 
-> **⚠️ DEPRECATION WARNING**
->
 > Starting with version 0.10.0, the string-based API for quantization configuration (e.g., `TorchAoConfig("int4_weight_only", group_size=128)`) is **deprecated** and will be removed in a future release.
 >
 > Please use the new `AOBaseConfig`-based approach instead:
@@ -94,7 +443,7 @@ For a complete list of available configurations, see our [quantization API docum
 >
 > The new API offers greater flexibility, better type safety, and access to the full range of features available in torchao.
 >
-> ## Migration Guide
+> [Migration Guide](#migration-guide)
 >
 > Here's how to migrate from common string identifiers to their `AOBaseConfig` equivalents:
 >
@@ -107,30 +456,10 @@ For a complete list of available configurations, see our [quantization API docum
 > All configuration objects accept parameters for customization (e.g., `group_size`, `scheme`, `layout`).
 
 
-Below is the API for for torchao < `0.9.0`
 
-```py
-import torch
-from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
+## Resources
 
-quantization_config = TorchAoConfig("int4_weight_only", group_size=128)
-quantized_model = AutoModelForCausalLM.from_pretrained(
-    "meta-llama/Meta-Llama-3-8B",
-    torch_dtype="auto",
-    device_map="auto",
-    quantization_config=quantization_config
-)
-
-tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")
-input_text = "What are we having for dinner?"
-input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
-
-# auto-compile the quantized model with `cache_implementation="static"` to get speed up
-output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
-print(tokenizer.decode(output[0], skip_special_tokens=True))
-```
-
-Run the code below to benchmark the quantized models performance.
+For a better sense of expected performance, view the [benchmarks](https://github.com/pytorch/ao/tree/main/torchao/quantization#benchmarks) for various models with CUDA and XPU backends. You can also run the code below to benchmark a model yourself.
 
 ```py
 from torch._inductor.utils import do_bench_using_profiling
@@ -153,76 +482,8 @@ print("bf16 model:", benchmark_fn(bf16_model.generate, **input_ids, max_new_toke
 > [!TIP]
 > For best performance, you can use recommended settings by calling `torchao.quantization.utils.recommended_inductor_config_setter()`
 
-</hfoption>
-<hfoption id="automatic">
-
-The [autoquant](https://pytorch.org/ao/stable/generated/torchao.quantization.autoquant.html#torchao.quantization.autoquant) API automatically chooses a quantization type for quantizable layers (`nn.Linear`) by micro-benchmarking on input type and shape and compiling a single linear layer.
-
-Create a [`TorchAoConfig`] and set to `"autoquant"`. Set the `cache_implementation` to `"static"` to automatically [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) the forward method. Finally, call `finalize_autoquant` on the quantized model to finalize the quantization and log the input shapes.
-
-> [!TIP]
-> Run the quantized model on a CPU by changing `device_map` to `"cpu"` and `layout` to `Int4CPULayout()`. This is only available in torchao 0.8.0+.
-
-```py
-import torch
-from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
-
-quantization_config = TorchAoConfig("autoquant", min_sqnr=None)
-quantized_model = AutoModelForCausalLM.from_pretrained(
-    "meta-llama/Meta-Llama-3-8B",
-    torch_dtype="auto",
-    device_map="auto",
-    quantization_config=quantization_config
-)
-
-tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")
-input_text = "What are we having for dinner?"
-input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
-
-# auto-compile the quantized model with `cache_implementation="static"` to get speed up
-output = quantized_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static")
-# explicitly call `finalize_autoquant` (may be refactored and removed in the future)
-quantized_model.finalize_autoquant()
-print(tokenizer.decode(output[0], skip_special_tokens=True))
-```
-
-Run the code below to benchmark the quantized models performance.
-
-```py
-from torch._inductor.utils import do_bench_using_profiling
-from typing import Callable
-
-def benchmark_fn(func: Callable, *args, **kwargs) -> float:
-    """Thin wrapper around do_bench_using_profiling"""
-    no_args = lambda: func(*args, **kwargs)
-    time = do_bench_using_profiling(no_args)
-    return time * 1e3
-
-MAX_NEW_TOKENS = 1000
-print("autoquantized model:", benchmark_fn(quantized_model.generate, **input_ids, max_new_tokens=MAX_NEW_TOKENS, cache_implementation="static"))
-
-bf16_model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.bfloat16)
-output = bf16_model.generate(**input_ids, max_new_tokens=10, cache_implementation="static") # auto-compile
-print("bf16 model:", benchmark_fn(bf16_model.generate, **input_ids, max_new_tokens=MAX_NEW_TOKENS, cache_implementation="static"))
-```
-
-</hfoption>
-</hfoptions>
-
-## Serialization
-
-torchao implements [torch.Tensor subclasses](https://pytorch.org/docs/stable/notes/extending.html#subclassing-torch-tensor) for maximum flexibility in supporting new quantized torch.Tensor formats. [Safetensors](https://huggingface.co/docs/safetensors/en/index) serialization and deserialization does not work with torchao.
-
-To avoid arbitrary user code execution, torchao sets `weights_only=True` in [torch.load](https://pytorch.org/docs/stable/generated/torch.load.html) to ensure only tensors are loaded. Any known user functions can be whitelisted with [add_safe_globals](https://pytorch.org/docs/stable/notes/serialization.html#torch.serialization.add_safe_globals).
-
-```py
-# don't serialize model with Safetensors
-output_dir = "llama3-8b-int4wo-128"
-quantized_model.save_pretrained("llama3-8b-int4wo-128", safe_serialization=False)
-```
-
-## Resources
-
-For a better sense of expected performance, view the [benchmarks](https://github.com/pytorch/ao/tree/main/torchao/quantization#benchmarks) for various models with CUDA and XPU backends.
-
 Refer to [Other Available Quantization Techniques](https://github.com/pytorch/ao/tree/main/torchao/quantization#other-available-quantization-techniques) for more examples and documentation.
+
+## Issues
+
+If you encounter any issues with the Transformers integration, please open an issue on the [Transformers](https://github.com/huggingface/transformers/issues) repository. For issues directly related to torchao, please open an issue on the [torchao](https://github.com/pytorch/ao/issues) repository.

docs/source/en/tasks/visual_document_retrieval.md

@@ -0,0 +1,144 @@
+<!--Copyright 2025 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.
+
+-->
+# Visual document retrieval
+
+Documents can contain multimodal data if they include charts, tables, and visuals in addition to text. Retrieving information from these documents is challenging because text retrieval models alone can't handle visual data and image retrieval models lack the granularity and document processing capabilities.
+
+Visual document retrieval can help retrieve information from all types of documents, including multimodal retrieval augmented generation (RAG). These models accept documents (as images) and texts and calculates the similarity scores between them.
+
+This guide demonstrates how to index and retrieve documents with [ColPali](../model_doc/colpali).  
+
+> [!TIP]
+> For large scale use cases, you may want to index and retrieve documents with a vector database.
+
+Make sure Transformers and Datasets is installed.
+
+```bash
+pip install -q datasets transformers
+```
+
+We will index a dataset of documents related to UFO sightings. We filter the examples where our column of interest is missing. It contains several columns, we are interested in the column `specific_detail_query` where it contains short summary of the document, and `image` column that contains our documents.
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("davanstrien/ufo-ColPali")
+dataset = dataset["train"]
+dataset = dataset.filter(lambda example: example["specific_detail_query"] is not None)
+dataset
+```
+```
+Dataset({
+    features: ['image', 'raw_queries', 'broad_topical_query', 'broad_topical_explanation', 'specific_detail_query', 'specific_detail_explanation', 'visual_element_query', 'visual_element_explanation', 'parsed_into_json'],
+    num_rows: 2172
+})
+```
+
+Let's load the model and the tokenizer.
+
+```python
+import torch
+from transformers import ColPaliForRetrieval, ColPaliProcessor
+
+model_name = "vidore/colpali-v1.2-hf"
+
+processor = ColPaliProcessor.from_pretrained(model_name)
+
+model = ColPaliForRetrieval.from_pretrained(
+    model_name,
+    torch_dtype=torch.bfloat16,
+    device_map="cuda",
+).eval()
+```
+
+Pass the text query to the processor and return the indexed text embeddings from the model. For image-to-text search, replace the `text` parameter in [`ColPaliProcessor`] with the `images` parameter to pass images.
+
+```python
+inputs = processor(text="a document about Mars expedition").to("cuda")
+with torch.no_grad():
+  text_embeds = model(**inputs, return_tensors="pt").embeddings
+```
+
+Index the images offline, and during inference, return the query text embeddings to get its closest image embeddings.
+
+Store the image and image embeddings by writing them to the dataset with [`~datasets.Dataset.map`] as shown below. Add an `embeddings` column that contains the indexed embeddings. ColPali embeddings take up a lot of storage, so remove them from the GPU and store them in the CPU as NumPy vectors.
+
+```python
+ds_with_embeddings = dataset.map(lambda example: {'embeddings': model(**processor(images=example["image"]).to("cuda"), return_tensors="pt").embeddings.to(torch.float32).detach().cpu().numpy()})
+```
+
+For online inference, create a function to search the image embeddings in batches and retrieve the k-most relevant images. The function below returns the indices in the dataset and their scores for a given indexed dataset, text embeddings, number of top results, and the batch size.
+
+```python
+def find_top_k_indices_batched(dataset, text_embedding, processor, k=10, batch_size=4):
+    scores_and_indices = []
+
+    for start_idx in range(0, len(dataset), batch_size):
+
+        end_idx = min(start_idx + batch_size, len(dataset))
+        batch = dataset[start_idx:end_idx]        
+        batch_embeddings = [torch.tensor(emb[0], dtype=torch.float32) for emb in batch["embeddings"]]
+        scores = processor.score_retrieval(text_embedding.to("cpu").to(torch.float32), batch_embeddings)
+
+        if hasattr(scores, "tolist"):
+            scores = scores.tolist()[0]
+
+        for i, score in enumerate(scores):
+            scores_and_indices.append((score, start_idx + i))
+
+    sorted_results = sorted(scores_and_indices, key=lambda x: -x[0])
+
+    topk = sorted_results[:k]
+    indices = [idx for _, idx in topk]
+    scores = [score for score, _ in topk]
+
+    return indices, scores
+```
+
+Generate the text embeddings and pass them to the function above to return the dataset indices and scores.
+
+```python
+with torch.no_grad():
+  text_embeds = model(**processor(text="a document about Mars expedition").to("cuda"), return_tensors="pt").embeddings
+indices, scores = find_top_k_indices_batched(ds_with_embeddings, text_embeds, processor, k=3, batch_size=4)
+print(indices, scores)
+```
+
+```
+([440, 442, 443],
+ [14.370786666870117,
+  13.675487518310547,
+  12.9899320602417])
+```
+
+Display the images to view the Mars related documents.
+
+```python
+for i in indices:
+  display(dataset[i]["image"])
+```
+
+<div style="display: flex; align-items: center;">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/doc_1.png" 
+         alt="Document 1" 
+         style="height: 200px; object-fit: contain; margin-right: 10px;">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/doc_2.png" 
+         alt="Document 2" 
+         style="height: 200px; object-fit: contain;">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/doc_3.png" 
+         alt="Document 3" 
+         style="height: 200px; object-fit: contain;">
+</div>

docs/source/en/tools.md

@@ -15,238 +15,4 @@ rendered properly in your Markdown viewer.
 -->
 
 > [!WARNING]
-> Agents and tools are being spun out into the standalone [smolagents](https://huggingface.co/docs/smolagents/index) library. These docs will be deprecated in the future!
-
-# Tools
-
-A tool is a function an agent can use to complete a task. Depending on your task, a tool can perform a web search, answer questions about a document, transcribe speech to text, and much more.
-
-Transformers provides a default set of tools for agents. These include the tools mentioned above as well as image question answering, text-to-speech, translation, and a Python code interpreter that executes the Python code generated by a LLM in a secure environment.
-
-Set `add_base_tools=True` to enable this default set of tools. The `tools` parameter is for adding additional tools. Leave the list empty if you aren't planning on adding any other tools to the toolbox.
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], add_base_tools=True)
-```
-
-You could also manually load a tool with [`load_tool`].
-
-```py
-from transformers import load_tool, ReactCodeAgent
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text-to-speech tool")
-agent = ReactCodeAgent(tools=[audio])
-```
-
-This guide will help you learn how to create your own tools and manage an agents toolbox.
-
-## Create a new tool
-
-You can create any tool you can dream of to empower an agent. The example in this section creates a tool that returns the most downloaded model for a task from the Hub, and the code for it is shown below.
-
-```py
-from huggingface_hub import list_models
-
-task = "text-classification"
-model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-print(model.id)
-```
-
-There are two ways you can create a tool, using a decorator or a superclass.
-
-### Tool decorator
-
-A fast and simple way to create a tool is to add the `@tool` decorator.
-
-Convert the code above into a tool by wrapping it in a function and adding the `@tool` decorator. The function needs:
-
-- A clear name that describes what the tool does, `model_download_counter`.
-- Type hints for the input and output (`str`).
-- A description that describes the tool in more detail and its arguments. This description is incorporated in the agents system prompt. It tells the agent *how* to use the tool, so try to make it as clear as possible!
-
-```py
-from transformers import tool
-
-@tool
-def model_download_counter(task: str) -> str:
-    """
-    This is a tool that returns the checkpoint name of the most downloaded model for a task from the Hugging Face Hub.
-
-    Args:
-        task: The task to retrieve the most downloaded model from.
-    """
-    model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-    return model.id
-```
-
-Pass the `model_download_counter` tool to the agents `tools` parameter to use it.
-
-```py
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[model_download_counter], add_base_tools=True)
-agent.run(
-    "Can you give me the name of the model that has the most downloads on the 'text-to-video' task on the Hugging Face Hub?"
-)
-```
-
-### Tool superclass
-
-Inheritance allows you to customize the [`Tool`] superclass or build a tool much more flexibly and comprehensively. This example will show you how to build the same `model_download_counter` tool as a [`Tool`] class.
-
-The [`Tool`] class needs:
-
-- A clear name that describes what the tool does, `model_download_counter`.
-- A description that describes the tool in more detail and its arguments. This description is incorporated in the agents system prompt. It tells the agent *how* to use the tool, so try to make it as clear as possible!
-- An `inputs` attribute that describes the input type. This is a dictionary with the keys, `type` and `description`.
-- An `outputs` attribute that describes the output type.
-- A `forward` method containing the code to be executed when the tool is called.
-
-Write the following code below to a file named `model_download.py`.
-
-```py
-from transformers import Tool
-from huggingface_hub import list_models
-
-class HFModelDownloadsTool(Tool):
-    name = "model_download_counter"
-    description = """
-    This is a tool that returns the checkpoint name of the most downloaded model for a task from the Hugging Face Hub."""
-
-    inputs = {
-        "task": {
-            "type": "string",
-            "description": "the task category (such as text-classification, depth-estimation, etc)",
-        }
-    }
-    output_type = "string"
-
-    def forward(self, task: str):
-        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-        return model.id
-```
-
-Import the tool from `model_download.py` and use [`load_tool`] to load it into the agent.
-
-```py
-from model_download import HFModelDownloadsTool
-from transformers import load_tool, CodeAgent
-
-tool = HFModelDownloadsTool()
-model_counter = load_tool(tool)
-agent = CodeAgent(tools=[model_counter], add_base_tools=True)
-```
-
-Also consider sharing your tool to the Hub with [`~Tool.push_to_hub`] so that everyone can use it!
-
-```py
-from model_download import HFModelDownloadsTool
-from transformers import load_tool, CodeAgent
-
-tool = HFModelDownloadsTool()
-tool.push_to_hub("{your_username}/hf-model-downloads")
-model_counter = load_tool("m-ric/hf-model-downloads")
-agent = CodeAgent(tools=[model_counter], add_base_tools=True)
-```
-
-## Add and replace tools
-
-Once an agent is initialized, add or replace its available tools without reinitializing the agent from scratch.
-
-Use [`add_tool`] to add a tool to an existing agent.
-
-```py
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], add_base_tools=True)
-agent.toolbox.add_tool(model_download_counter)
-```
-
-Now you can use the default text-to-speech tool to read aloud the most downloaded model for the text-to-video task.
-
-```py
-agent.run(
-    "Can you read out loud the name of the model that has the most downloads on the 'text-to-video' task on the Hugging Face Hub and return the audio?"
-)
-```
-
-> [!WARNING]
-> When adding tools to an agent that already works well, it can bias the agent towards your tool or a tool other than the one currently defined.
-
-Use [`update_tool`] to replace an agents existing tool. This is useful if the new tool is a one-to-one replacement of the existing tool because the agent already knows how to perform the task. The new tool should follow the same API as the tool it replaced or the system prompt template should be adapted to ensure all examples using the replaced tool are updated.
-
-```py
-agent.toolbox.update_tool(new_model_download_counter)
-```
-
-## ToolCollection
-
-A [`ToolCollection`] is a collection of Hugging Face [Spaces](https://hf.co/spaces) that can be quickly loaded and used by an agent.
-
-> [!TIP]
-> Learn more about creating collections on the Hub.
-
-Create a [`ToolCollection`] object and specify the `collection_slug` of the collection you want to use, and then pass it to the agent. To speed up the starting process, tools are only loaded if they're called by the agent.
-
-The example loads a collection of image generation tools.
-
-```py
-from transformers import ToolCollection, ReactCodeAgent
-
-image_tool_collection = ToolCollection(collection_slug="")
-agent = ReactCodeAgent(tools=[*image_tool_collection], add_base_tools=True)
-agent.run(
-    "Please draw me a picture of rivers and lakes."
-)
-```
-
-## Tool integrations
-
-Transformers supports tools from several other libraries, such as [gradio-tools](https://github.com/freddyaboulton/gradio-tools) and [LangChain](https://python.langchain.com/docs/introduction/).
-
-### gradio-tools
-
-gradio-tools is a library that enables [Gradio](https://www.gradio.app/) apps to be used as tools. With the wide variety of Gradio apps available, you can enhance your agent with a range of tools like generating images and videos or transcribing audio and summarizing it.
-
-Import and instantiate a tool from gradio-tools, for example, the [StableDiffusionPromptGeneratorTool](https://github.com/freddyaboulton/gradio-tools/blob/main/gradio_tools/tools/prompt_generator.py). This tool can help improve prompts to generate better images.
-
-> [!WARNING]
-> gradio-tools require text inputs and outputs even when working with different modalities like images and audio, which are currently incompatible.
-
-Use [`~Tool.from_gradio`] to load the prompt generator tool.
-
-```py
-from gradio_tools import StableDiffusionPromptGeneratorTool
-from transformers import Tool, load_tool, CodeAgent
-
-gradio_prompt_generator_tool = StableDiffusionPromptGeneratorTool()
-prompt_generator_tool = Tool.from_gradio(gradio_prompt_generator_tool)
-```
-
-Now pass it to the agent along with a text-to-image tool.
-
-```py
-image_generation_tool = load_tool("huggingface-tools/text-to-image")
-agent = CodeAgent(tools=[prompt_generator_tool, image_generation_tool], llm_engine=llm_engine)
-agent.run(
-    "Improve this prompt, then generate an image of it.", prompt="A rabbit wearing a space suit"
-)
-```
-
-### LangChain
-
-LangChain is a library for working with LLMs which includes agents and tools. Use the [`~Tool.from_langchain`] method to load any LangChain tool into an agent.
-
-The example below demonstrates how to use LangChains web search tool.
-
-```py
-from langchain.agents import load_tools
-from transformers import Tool, ReactCodeAgent
-
-search_tool = Tool.from_langchain(load_tools(["serpapi"])[0])
-agent = ReactCodeAgent(tools=[search_tool])
-agent.run("How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?")
-```
+> Agents and tools were spun out into the standalone [smolagents](https://huggingface.co/docs/smolagents/index) library. They were removed from `transformers` in v4.52.

docs/source/en/trainer.md

@@ -341,29 +341,9 @@ use_cpu: false
 ```
 
 </hfoption>
-<hfoption id="Tensor parallelism with PyTorch 2">
-
-```yaml
-compute_environment: LOCAL_MACHINE
-tp_config:
-  tp_size: 4
-distributed_type: TP
-downcast_bf16: 'no'
-machine_rank: 0
-main_training_function: main
-mixed_precision: 'no'
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
-```
-
 </hfoptions>
 
+
 Run [accelerate_launch](https://hf.co/docs/accelerate/package_reference/cli#accelerate-launch) to start training with the configurations set in `config_file.yaml`. This file is saved to the Accelerate cache folder and automatically loaded when you run `accelerate_launch`.
 
 The example below launches the [run_glue.py](../../../examples/pytorch/text-classification/run_glue) script with the FSDP configuration shown earlier. Parameters from the `config_file.yaml` file can also be directly set in the command line.

docs/source/es/trainer.md

@@ -363,29 +363,6 @@ use_cpu: false
 
 </hfoption>
 
-<hfoption id="Tensor Parallelism with PyTorch 2">
-
-```yml
-compute_environment: LOCAL_MACHINE
-tp_config:
-  tp_size: 4
-distributed_type: TP
-downcast_bf16: 'no'
-machine_rank: 0
-main_training_function: main
-mixed_precision: 'no'
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
-
-```
-
-</hfoption>
 </hfoptions>
 
 El comando [`accelerate_launch`](https://huggingface.co/docs/accelerate/package_reference/cli#accelerate-launch) es la forma recomendada de lanzar tu script de entrenamiento en un sistema distribuido con Accelerate y [`Trainer`] con los parámetros especificados en `config_file.yaml`. Este archivo se guarda en la carpeta de caché de Accelerate y se carga automáticamente cuando ejecutas `accelerate_launch`.

docs/source/fr/_toctree.yml

@@ -23,8 +23,6 @@
     title: Chargement et entraînement des adaptateurs avec 🤗 PEFT
   - local: in_translation
     title: Partager un modèle
-  - local: in_translation
-    title: Agents
   - local: in_translation
     title: Génération avec LLMs
   title: Tutoriels
@@ -33,4 +31,4 @@
     title: Ce que 🤗 Transformers peut faire
   - local: tasks_explained
     title: Comment 🤗 Transformers résout ces tâches
-  title: Guides conceptuels
+  title: Guides conceptuels

docs/source/ja/_toctree.yml

@@ -23,8 +23,6 @@
     title: 🤗 PEFT を使用してアダプターをロードしてトレーニングする
   - local: model_sharing
     title: モデルを共有する
-  - local: transformers_agents
-    title: エージェント
   - local: llm_tutorial
     title: LLM を使用した生成
   title: Tutorials
@@ -119,8 +117,6 @@
     title: トーチスクリプトへのエクスポート
   - local: community
     title: コミュニティリソース
-  - local: custom_tools
-    title: カスタムツールとプロンプト
   - local: troubleshooting
     title: トラブルシューティング
   title: 開発者ガイド
@@ -200,8 +196,6 @@
   title: コンセプチュアルガイド
 - sections:
   - sections:
-    - local: main_classes/agent
-      title: エージェントとツール
     - local: model_doc/auto
       title: Auto Classes
     - local: main_classes/callback

docs/source/ja/custom_tools.md

@@ -1,26 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Custom Tools and Prompts
-
-<Tip warning={true}>
-
-The Agents framework has significantly changed in version v4.41.0.
-This document has been removed as it was referencing an older API.
-
-We eagerly welcome new contributions for the updated API.
-
-</Tip>

docs/source/ja/model_doc/bit.md

@@ -54,6 +54,11 @@ BiT を始めるのに役立つ公式 Hugging Face およびコミュニティ (
 [[autodoc]] BitImageProcessor
     - preprocess
 
+## BitImageProcessorFast
+
+[[autodoc]] BitImageProcessorFast
+    - preprocess
+
 ## BitModel
 
 [[autodoc]] BitModel

docs/source/ja/model_doc/bridgetower.md

@@ -144,6 +144,11 @@ BridgeTower は、ビジュアル エンコーダー、テキスト エンコー
 [[autodoc]] BridgeTowerImageProcessor
     - preprocess
 
+## BridgeTowerImageProcessorFast
+
+[[autodoc]] BridgeTowerImageProcessorFast
+    - preprocess
+
 ## BridgeTowerProcessor
 
 [[autodoc]] BridgeTowerProcessor

docs/source/ja/model_doc/chinese_clip.md

@@ -86,6 +86,11 @@ Chinese-CLIP モデルは、[OFA-Sys](https://huggingface.co/OFA-Sys) によっ
 [[autodoc]] ChineseCLIPImageProcessor
     - preprocess
 
+## ChineseCLIPImageProcessorFast
+
+[[autodoc]] ChineseCLIPImageProcessorFast
+    - preprocess
+
 ## ChineseCLIPFeatureExtractor
 
 [[autodoc]] ChineseCLIPFeatureExtractor

docs/source/ja/model_doc/conditional_detr.md

@@ -43,6 +43,11 @@ alt="描画" width="600"/>
 
 [[autodoc]] ConditionalDetrImageProcessor
     - preprocess
+
+## ConditionalDetrImageProcessorFast
+
+[[autodoc]] ConditionalDetrImageProcessorFast
+    - preprocess
     - post_process_object_detection
     - post_process_instance_segmentation
     - post_process_semantic_segmentation

docs/source/ja/transformers_agents.md

@@ -1,282 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Transformers Agents
-
-<Tip warning={true}>
-
-Transformers Agentsは、いつでも変更される可能性のある実験的なAPIです。エージェントが返す結果は、APIまたは基礎となるモデルが変更される可能性があるため、異なることがあります。
-
-</Tip>
-
-Transformersバージョンv4.29.0は、*ツール*と*エージェント*のコンセプトを基に構築されています。この[colab](https://colab.research.google.com/drive/1c7MHD-T1forUPGcC_jlwsIptOzpG3hSj)で試すことができます。
-
-要するに、これはtransformersの上に自然言語APIを提供するものです：私たちは一連の厳選されたツールを定義し、自然言語を解釈し、これらのツールを使用するエージェントを設計します。これは設計上拡張可能です。私たちはいくつかの関連するツールを厳選しましたが、コミュニティによって開発された任意のツールを使用するためにシステムを簡単に拡張できる方法も示します。
-
-この新しいAPIで何ができるかのいくつかの例から始めましょう。特に多モーダルなタスクに関して強力ですので、画像を生成したりテキストを読み上げたりするのに最適です。
-
-上記のテキストの上に、日本語の翻訳を提供します。
-
-
-```py
-agent.run("Caption the following image", image=image)
-```
-
-| **Input**                                                                                                                   | **Output**                        |
-|-----------------------------------------------------------------------------------------------------------------------------|-----------------------------------|
-| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/beaver.png" width=200> | A beaver is swimming in the water |
-
----
-
-```py
-agent.run("Read the following text out loud", text=text)
-```
-| **Input**                                                                                                               | **Output**                                   |
-|-------------------------------------------------------------------------------------------------------------------------|----------------------------------------------|
-| A beaver is swimming in the water | <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tts_example.wav" type="audio/wav"> your browser does not support the audio element. </audio>
-
----
-
-```py
-agent.run(
-    "In the following `document`, where will the TRRF Scientific Advisory Council Meeting take place?",
-    document=document,
-)
-```
-| **Input**                                                                                                                   | **Output**     |
-|-----------------------------------------------------------------------------------------------------------------------------|----------------|
-| <img src="https://datasets-server.huggingface.co/assets/hf-internal-testing/example-documents/--/hf-internal-testing--example-documents/test/0/image/image.jpg" width=200> | ballroom foyer |
-
-## Quickstart
-
-`agent.run`を使用する前に、エージェントをインスタンス化する必要があります。エージェントは、大規模な言語モデル（LLM）です。
-OpenAIモデルとBigCode、OpenAssistantからのオープンソースの代替モデルをサポートしています。OpenAIモデルはパフォーマンスが優れていますが、OpenAIのAPIキーが必要であり、無料で使用することはできません。一方、Hugging FaceはBigCodeとOpenAssistantモデルのエンドポイントへの無料アクセスを提供しています。
-
-まず、デフォルトの依存関係をすべてインストールするために`agents`のエクストラをインストールしてください。
-
-
-```bash
-pip install transformers[agents]
-```
-
-OpenAIモデルを使用するには、`openai`の依存関係をインストールした後、`OpenAiAgent`をインスタンス化します。
-
-
-```bash
-pip install openai
-```
-
-
-```py
-from transformers import OpenAiAgent
-
-agent = OpenAiAgent(model="text-davinci-003", api_key="<your_api_key>")
-```
-
-BigCodeまたはOpenAssistantを使用するには、まずログインしてInference APIにアクセスしてください。
-
-```py
-from huggingface_hub import login
-
-login("<YOUR_TOKEN>")
-```
-
-次に、エージェントをインスタンス化してください。
-
-```py
-from transformers import HfAgent
-
-# Starcoder
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
-# StarcoderBase
-# agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoderbase")
-# OpenAssistant
-# agent = HfAgent(url_endpoint="https://api-inference.huggingface.co/models/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5")
-```
-
-これは、Hugging Faceが現在無料で提供している推論APIを使用しています。このモデル（または別のモデル）の独自の推論エンドポイントをお持ちの場合は、上記のURLエンドポイントをご自分のURLエンドポイントで置き換えることができます。
-
-<Tip>
-
-StarCoderとOpenAssistantは無料で利用でき、シンプルなタスクには非常に優れた性能を発揮します。ただし、より複雑なプロンプトを処理する際には、チェックポイントが十分でないことがあります。そのような場合には、現時点ではオープンソースではないものの、パフォーマンスが向上する可能性のあるOpenAIモデルを試してみることをお勧めします。
-
-</Tip>
-
-これで準備が整いました！これから、あなたが利用できる2つのAPIについて詳しく説明します。
-
-### Single execution (run)
-
-単一実行メソッドは、エージェントの [`~Agent.run`] メソッドを使用する場合です。
-
-
-```py
-agent.run("Draw me a picture of rivers and lakes.")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200>
-
-
-これは、実行したいタスクに適したツール（またはツール）を自動的に選択し、適切に実行します。1つまたは複数のタスクを同じ命令で実行することができます（ただし、命令が複雑であるほど、エージェントが失敗する可能性が高くなります）。
-
-
-```py
-agent.run("Draw me a picture of the sea then transform the picture to add an island")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/sea_and_island.png" width=200>
-
-<br/>
-
-[`~Agent.run`] 操作は独立して実行できますので、異なるタスクで何度も実行することができます。
-
-注意点として、あなたの `agent` は単なる大規模な言語モデルであるため、プロンプトのわずかな変更でも完全に異なる結果が得られる可能性があります。したがって、実行したいタスクをできるだけ明確に説明することが重要です。良いプロンプトの書き方については、[こちら](custom_tools#writing-good-user-inputs) で詳しく説明しています。
-
-実行ごとに状態を保持したり、テキスト以外のオブジェクトをエージェントに渡したりする場合は、エージェントが使用する変数を指定することができます。例えば、最初の川や湖の画像を生成し、その画像に島を追加するようにモデルに指示するには、次のように行うことができます：
-
-```python
-picture = agent.run("Generate a picture of rivers and lakes.")
-updated_picture = agent.run("Transform the image in `picture` to add an island to it.", picture=picture)
-```
-
-<Tip>
-
-これは、モデルがあなたのリクエストを理解できない場合や、ツールを混同する場合に役立つことがあります。例えば：
-
-```py
-agent.run("Draw me the picture of a capybara swimming in the sea")
-```
-
-ここでは、モデルは2つの方法で解釈できます：
-- `text-to-image`に海で泳ぐカピバラを生成させる
-- または、`text-to-image`でカピバラを生成し、それを海で泳がせるために`image-transformation`ツールを使用する
-
-最初のシナリオを強制したい場合は、プロンプトを引数として渡すことができます：
-
-
-```py
-agent.run("Draw me a picture of the `prompt`", prompt="a capybara swimming in the sea")
-```
-
-</Tip>
-
-
-### Chat-based execution (チャット)
-
-エージェントは、[`~Agent.chat`] メソッドを使用することで、チャットベースのアプローチも可能です。
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200> 
-
-```py
-agent.chat("Transform the picture so that there is a rock in there")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_and_beaver.png" width=200>
-
-<br/>
-
-これは、指示をまたいで状態を保持したい場合に便利なアプローチで、単一の指示に比べて複雑な指示を処理するのは難しいかもしれません（その場合は [`~Agent.run`] メソッドの方が適しています）。
-
-このメソッドは、非テキスト型の引数や特定のプロンプトを渡したい場合にも使用できます。
-
-### ⚠️ Remote execution
-
-デモンストレーションの目的やすべてのセットアップで使用できるように、リリースのためにいくつかのデフォルトツール用のリモート実行ツールも作成しました。これらは [推論エンドポイント](https://huggingface.co/inference-endpoints) を使用して作成されます。
-
-これらは現在オフになっていますが、リモート実行ツールを自分で設定する方法については、[カスタムツールガイド](./custom_tools) を読むことをお勧めします。
-
-### What's happening here? What are tools, and what are agents?
-
-![エージェントとツールのダイアグラム](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/diagram.png)
-
-#### Agents
-
-ここでの「エージェント」とは、大規模な言語モデルのことであり、特定の一連のツールにアクセスできるようにプロンプトを設定しています。
-
-LLM（大規模言語モデル）は、コードの小さなサンプルを生成するのにかなり優れており、このAPIは、エージェントに特定のツールセットを使用してタスクを実行するコードの小さなサンプルを生成させることに利用しています。このプロンプトは、エージェントにタスクとツールの説明を提供することで、エージェントが使用しているツールのドキュメントにアクセスし、関連するコードを生成できるようになります。
-
-#### Tools
-
-ツールは非常に単純で、名前と説明からなる単一の関数です。それから、これらのツールの説明を使用してエージェントをプロンプトします。プロンプトを通じて、エージェントに、ツールを使用してクエリで要求されたタスクをどのように実行するかを示します。特に、ツールの期待される入力と出力を示します。
-
-これは新しいツールを使用しており、パイプラインではなくツールを使用しています。なぜなら、エージェントは非常に原子的なツールでより良いコードを生成するからです。パイプラインはよりリファクタリングされ、しばしば複数のタスクを組み合わせています。ツールは非常に単純なタスクに焦点を当てることを意図しています。
-
-#### Code-execution?!
-
-このコードは、ツールとツールと一緒に渡される入力のセットで、当社の小規模なPythonインタープリタで実行されます。すでに提供されたツールとprint関数しか呼び出すことができないため、実行できることはすでに制限されています。Hugging Faceのツールに制限されているため、安全だと考えても問題ありません。
-
-さらに、属性の検索やインポートは許可しておらず（それらは渡された入力/出力を処理するためには必要ないはずです）、最も明らかな攻撃は問題ありません（エージェントにそれらを出力するようにプロンプトする必要があります）。超安全な側に立ちたい場合は、追加の引数 return_code=True を指定して run() メソッドを実行できます。その場合、エージェントは実行するコードを返すだけで、実行するかどうかはあなた次第です。
-
-実行は、違法な操作を試みる行またはエージェントが生成したコードに通常のPythonエラーがある場合に停止します。
-
-### A curated set of tools
-
-私たちは、このようなエージェントを強化できるツールのセットを特定します。以下は、`transformers`に統合されたツールの更新されたリストです：
-
-- **ドキュメント質問応答**: 画像形式のドキュメント（PDFなど）が与えられた場合、このドキュメントに関する質問に回答します（[Donut](./model_doc/donut)）
-- **テキスト質問応答**: 長いテキストと質問が与えられた場合、テキスト内の質問に回答します（[Flan-T5](./model_doc/flan-t5)）
-- **無条件の画像キャプション**: 画像にキャプションを付けます！（[BLIP](./model_doc/blip)）
-- **画像質問応答**: 画像が与えられた場合、その画像に関する質問に回答します（[VILT](./model_doc/vilt)）
-- **画像セグメンテーション**: 画像とプロンプトが与えられた場合、そのプロンプトのセグメンテーションマスクを出力します（[CLIPSeg](./model_doc/clipseg)）
-- **音声からテキストへの変換**: 人の話し声のオーディオ録音が与えられた場合、その音声をテキストに転記します（[Whisper](./model_doc/whisper)）
-- **テキストから音声への変換**: テキストを音声に変換します（[SpeechT5](./model_doc/speecht5)）
-- **ゼロショットテキスト分類**: テキストとラベルのリストが与えられた場合、テキストが最も対応するラベルを識別します（[BART](./model_doc/bart)）
-- **テキスト要約**: 長いテキストを1つまたは数文に要約します（[BART](./model_doc/bart)）
-- **翻訳**: テキストを指定された言語に翻訳します（[NLLB](./model_doc/nllb)）
-
-これらのツールはtransformersに統合されており、手動でも使用できます。たとえば、次のように使用できます：
-
-```py
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-### Custom tools
-
-私たちは、厳選されたツールのセットを特定する一方、この実装が提供する主要な価値は、カスタムツールを迅速に作成して共有できる能力だと強く信じています。
-
-ツールのコードをHugging Face Spaceまたはモデルリポジトリにプッシュすることで、エージェントと直接連携してツールを活用できます。[`huggingface-tools` organization](https://huggingface.co/huggingface-tools)には、**transformers非依存**のいくつかのツールが追加されました：
-
-- **テキストダウンローダー**: ウェブURLからテキストをダウンロードするためのツール
-- **テキストから画像へ**: プロンプトに従って画像を生成するためのツール。安定した拡散を活用します
-- **画像変換**: 初期画像とプロンプトを指定して画像を変更するためのツール。instruct pix2pixの安定した拡散を活用します
-- **テキストからビデオへ**: プロンプトに従って小さなビデオを生成するためのツール。damo-vilabを活用します
-
-最初から使用しているテキストから画像へのツールは、[*huggingface-tools/text-to-image*](https://huggingface.co/spaces/huggingface-tools/text-to-image)にあるリモートツールです！今後も、この組織および他の組織にさらにこのようなツールをリリースし、この実装をさらに強化していきます。
-
-エージェントはデフォルトで[`huggingface-tools`](https://huggingface.co/huggingface-tools)にあるツールにアクセスできます。
-ツールの作成と共有方法、またHubに存在するカスタムツールを活用する方法についての詳細は、[次のガイド](custom_tools)で説明しています。
-
-### Code generation
-
-これまで、エージェントを使用してあなたのためにアクションを実行する方法を示しました。ただし、エージェントはコードを生成するだけで、非常に制限されたPythonインタープリタを使用して実行します。生成されたコードを異なる環境で使用したい場合、エージェントにコードを返すように指示できます。ツールの定義と正確なインポートも含めて。
-
-例えば、以下の命令：
-```python
-agent.run("Draw me a picture of rivers and lakes", return_code=True)
-```
-
-次のコードを返します
-```python
-from transformers import load_tool
-
-image_generator = load_tool("huggingface-tools/text-to-image")
-
-image = image_generator(prompt="rivers and lakes")
-```
-
-その後、自分で変更して実行できます。

docs/source/ko/_toctree.yml

@@ -23,8 +23,6 @@
     title: 🤗 PEFT로 어댑터 로드 및 학습하기
   - local: model_sharing
     title: 만든 모델 공유하기
-  - local: transformers_agents
-    title: 에이전트
   - local: llm_tutorial
     title: 대규모 언어 모델로 생성하기
   - local: conversations
@@ -248,8 +246,6 @@
   title: (번역중) 개념 가이드
 - sections:
   - sections:
-    - local: main_classes/agent
-      title: 에이전트와 도구
     - local: model_doc/auto
       title: 자동 클래스
     - local: in_translation

docs/source/ko/gguf.md

@@ -88,7 +88,7 @@ model = AutoModelForCausalLM.from_pretrained(model_id, gguf_file=filename)
 
 이제 PyTorch 생태계에서 모델의 양자화되지 않은 전체 버전에 접근할 수 있으며, 다른 여러 도구들과 결합하여 사용할 수 있습니다.
 
-`gguf` 파일로 다시 변환하려면 llama.cpp의 [`convert-hf-to-gguf.py`](https://github.com/ggerganov/llama.cpp/blob/master/convert-hf-to-gguf.py)를 사용하는 것을 권장합니다.
+`gguf` 파일로 다시 변환하려면 llama.cpp의 [`convert-hf-to-gguf.py`](https://github.com/ggerganov/llama.cpp/blob/master/convert_hf_to_gguf.py)를 사용하는 것을 권장합니다.
 
 위의 스크립트를 완료하여 모델을 저장하고 다시 `gguf`로 내보내는 방법은 다음과 같습니다:
 

docs/source/ko/main_classes/agent.md

@@ -1,134 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# 에이전트 & 도구 [[agents-tools]]
-
-<Tip warning={true}>
-
-Transformers Agent는 실험 중인 API이므로 언제든지 변경될 수 있습니다. 
-API나 기반 모델이 자주 업데이트되므로, 에이전트가 제공하는 결과물은 달라질 수 있습니다.
-
-</Tip>
-
-에이전트와 도구에 대해 더 알아보려면 [소개 가이드](../transformers_agents)를 꼭 읽어보세요. 
-이 페이지에는 기본 클래스에 대한 API 문서가 포함되어 있습니다.
-
-## 에이전트 [[agents]]
-
-우리는 기본 [`Agent`] 클래스를 기반으로 두 가지 유형의 에이전트를 제공합니다:
-- [`CodeAgent`]는 한 번에 동작합니다. 작업을 해결하기 위해 코드를 생성한 다음, 바로 실행합니다.
-- [`ReactAgent`]는 단계별로 동작하며, 각 단계는 하나의 생각, 하나의 도구 호출 및 실행으로 구성됩니다. 이 에이전트에는 두 가지 클래스가 있습니다:
-  - [`ReactJsonAgent`]는 도구 호출을 JSON으로 작성합니다.
-  - [`ReactCodeAgent`]는 도구 호출을 Python 코드로 작성합니다.
-
-### Agent [[agent]]
-
-[[autodoc]] Agent
-
-### CodeAgent [[codeagent]]
-
-[[autodoc]] CodeAgent
-
-### React agents [[react-agents]]
-
-[[autodoc]] ReactAgent
-
-[[autodoc]] ReactJsonAgent
-
-[[autodoc]] ReactCodeAgent
-
-## Tools [[tools]]
-
-### load_tool [[loadtool]]
-
-[[autodoc]] load_tool
-
-### Tool [[tool]]
-
-[[autodoc]] Tool
-
-### Toolbox [[toolbox]]
-
-[[autodoc]] Toolbox
-
-### PipelineTool [[pipelinetool]]
-
-[[autodoc]] PipelineTool
-
-### launch_gradio_demo [[launchgradiodemo]]
-
-[[autodoc]] launch_gradio_demo
-
-### ToolCollection [[toolcollection]]
-
-[[autodoc]] ToolCollection
-
-## 엔진 [[engines]]
-
-에이전트 프레임워크에서 사용할 수 있는 엔진을 자유롭게 만들고 사용할 수 있습니다.
-이 엔진들은 다음과 같은 사양을 가지고 있습니다:
-1. 입력(`List[Dict[str, str]]`)에 대한 [메시지 형식](../chat_templating.md)을 따르고 문자열을 반환해야 합니다.
-2. 인수 `stop_sequences`에 시퀀스가 전달되기 *전에* 출력을 생성하는 것을 중지해야 합니다.
-
-### HfApiEngine [[HfApiEngine]]
-
-편의를 위해, 위의 사항을 구현하고 대규모 언어 모델 실행을 위해 추론 엔드포인트를 사용하는 `HfApiEngine`을 추가했습니다.
-
-```python
->>> from transformers import HfApiEngine
-
->>> messages = [
-...   {"role": "user", "content": "Hello, how are you?"},
-...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
-...   {"role": "user", "content": "No need to help, take it easy."},
-... ]
-
->>> HfApiEngine()(messages, stop_sequences=["conversation"])
-
-"That's very kind of you to say! It's always nice to have a relaxed "
-```
-
-[[autodoc]] HfApiEngine
-
-
-## 에이전트 유형 [[agent-types]]
-
-에이전트는 도구 간의 모든 유형의 객체를 처리할 수 있습니다; 도구는 완전히 멀티모달이므로 텍스트, 이미지, 오디오, 비디오 등 다양한 유형을 수락하고 반환할 수 있습니다. 
-도구 간의 호환성을 높이고 ipython (jupyter, colab, ipython 노트북, ...)에서 이러한 
-반환 값을 올바르게 렌더링하기 위해 이러한 유형을 중심으로 래퍼 클래스를 
-구현합니다.
-
-래핑된 객체는 처음과 동일하게 작동해야 합니다; 텍스트 객체는 여전히 문자열로 작동해야 하며, 
-이미지 객체는 여전히 `PIL.Image`로 작동해야 합니다.
-
-이러한 유형에는 세 가지 특정 목적이 있습니다:
-
-- `to_raw`를 호출하면 기본 객체가 반환되어야 합니다.
-- `to_string`을 호출하면 객체가 문자열로 반환되어야 합니다: 
-`AgentText`의 경우 문자열이 될 수 있지만, 다른 경우에는 객체의 직렬화된 버전의 경로일 수 있습니다.
-- ipython 커널에서 표시할 때 객체가 올바르게 표시되어야 합니다.
-
-### AgentText [[agenttext]]
-
-[[autodoc]] transformers.agents.agent_types.AgentText
-
-### AgentImage [[agentimage]]
-
-[[autodoc]] transformers.agents.agent_types.AgentImage
-
-### AgentAudio [[agentaudio]]
-
-[[autodoc]] transformers.agents.agent_types.AgentAudio

docs/source/ko/trainer.md

@@ -549,29 +549,7 @@ use_cpu: false
 ```
 
 </hfoption>
-<hfoption id="Tensor Parallelism with PyTorch 2">
 
-```yml
-compute_environment: LOCAL_MACHINE
-tp_config:
-  tp_size: 4
-distributed_type: TP
-downcast_bf16: 'no'
-machine_rank: 0
-main_training_function: main
-mixed_precision: 'no'
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
-
-```
-
-</hfoption>
 </hfoptions>
 
 [`accelerate_launch`](https://huggingface.co/docs/accelerate/package_reference/cli#accelerate-launch) 명령은 Accelerate와 [`Trainer`]를 사용하여 분산 시스템에서 훈련 스크립트를 실행하는 권장 방법이며, `config_file.yaml`에 지정된 매개변수를 사용합니다. 이 파일은 Accelerate 캐시 폴더에 저장되며 `accelerate_launch`를 실행할 때 자동으로 로드됩니다.

docs/source/ko/transformers_agents.md

@@ -1,328 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Transformers Agent [[transformers-agent]]
-
-<Tip warning={true}>
-
-Transformers Agent는 실험 중인 API로 언제든지 변경될 수 있습니다. 
-API 또는 기반 모델이 변경되기 쉽기 때문에 에이전트가 반환하는 결과도 달라질 수 있습니다.
-
-</Tip>
-
-Transformers 버전 4.29.0.에서 *도구*와 *에이전트*라는 컨셉을 도입했습니다. [이 colab](https://colab.research.google.com/drive/1c7MHD-T1forUPGcC_jlwsIptOzpG3hSj)에서 사용해볼 수 있습니다.
-
-간단히 말하면, Agent는 트랜스포머 위에 자연어 API를 제공합니다. 
-엄선된 도구 세트를 정의하고, 자연어를 해석하여 이러한 도구를 사용할 수 있는 에이전트를 설계했습니다. 
-이 API는 확장이 가능하도록 설계 되었습니다. 
-주요 도구를 선별해두었지만, 커뮤니티에서 개발한 모든 도구를 사용할 수 있도록 시스템을 쉽게 확장할 수 있는 방법도 보여드리겠습니다.
-
-몇 가지 예를 통해 새로운 API로 무엇을 할 수 있는지 살펴보겠습니다. 
-이 API는 특히 멀티모달 작업에서 강력하므로 이미지를 생성하고 텍스트를 소리내어 읽어보겠습니다.
-
-```py
-agent.run("Caption the following image", image=image)
-```
-
-| **Input**                                                                                                                   | **Output**                        |
-|-----------------------------------------------------------------------------------------------------------------------------|-----------------------------------|
-| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/beaver.png" width=200> | A beaver is swimming in the water |
-
----
-
-```py
-agent.run("Read the following text out loud", text=text)
-```
-| **Input**                                                                                                               | **Output**                                   |
-|-------------------------------------------------------------------------------------------------------------------------|----------------------------------------------|
-| A beaver is swimming in the water | <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tts_example.wav" type="audio/wav"> your browser does not support the audio element. </audio>
-
----
-
-```py
-agent.run(
-    "In the following `document`, where will the TRRF Scientific Advisory Council Meeting take place?",
-    document=document,
-)
-```
-| **Input**                                                                                                                   | **Output**     |
-|-----------------------------------------------------------------------------------------------------------------------------|----------------|
-| <img src="https://datasets-server.huggingface.co/assets/hf-internal-testing/example-documents/--/hf-internal-testing--example-documents/test/0/image/image.jpg" width=200> | ballroom foyer |
-
-## 바로 시작하기 [[quickstart]]
-
-`agent.run`을 사용하려면 먼저 대규모 언어 모델(LLM)인 에이전트를 인스턴스화해야 합니다. 
-저희는 openAI 모델뿐만 아니라 BigCode 및 OpenAssistant의 오픈소스 대체 모델도 지원합니다. 
-openAI 모델의 성능이 더 우수하지만(단, openAI API 키가 필요하므로 무료로 사용할 수 없음), 
-Hugging Face는 BigCode와 OpenAssistant 모델의 엔드포인트에 대한 무료 액세스를 제공하고 있습니다.
-
-우선 모든 기본 종속성을 설치하려면 `agents`를 추가로 설치하세요.
-```bash
-pip install transformers[agents]
-```
-
-openAI 모델을 사용하려면 `openai` 종속성을 설치한 후 [`OpenAiAgent`]를 인스턴스화합니다:
-
-```bash
-pip install openai
-```
-
-
-```py
-from transformers import OpenAiAgent
-
-agent = OpenAiAgent(model="text-davinci-003", api_key="<your_api_key>")
-```
-
-BigCode 또는 OpenAssistant를 사용하려면 먼저 로그인하여 Inference API에 액세스하세요:
-
-```py
-from huggingface_hub import login
-
-login("<YOUR_TOKEN>")
-```
-
-그런 다음 에이전트를 인스턴스화합니다.
-
-```py
-from transformers import HfAgent
-
-# Starcoder
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
-# StarcoderBase
-# agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoderbase")
-# OpenAssistant
-# agent = HfAgent(url_endpoint="https://api-inference.huggingface.co/models/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5")
-```
-
-현재 Hugging Face에서 무료로 제공하는 추론 API를 사용하고 있습니다. 
-이 모델에 대한 자체 추론 엔드포인트가 있는 경우(또는 다른 엔드포인트가 있는 경우) 위의 URL을 해당 URL 엔드포인트로 바꿀 수 있습니다.
-
-<Tip>
-
-StarCoder와 OpenAssistant는 무료로 사용할 수 있으며 간단한 작업에서 놀라울 정도로 잘 작동합니다. 
-그러나 더 복잡한 프롬프트를 처리할 때는 체크포인트가 잘 작동하지 않습니다. 
-이러한 문제가 발생하면 OpenAI 모델을 사용해 보시기 바랍니다. 아쉽게도 오픈소스는 아니지만 현재로서는 더 나은 성능을 제공합니다.
-
-</Tip>
-
-이제 준비가 완료되었습니다! 이제 자유롭게 사용할 수 있는 두 가지 API에 대해 자세히 알아보겠습니다.
-
-### 단일 실행 (run) [[single-execution-(run)]] 
-
-단일 실행 방법은 에이전트의 [`~Agent.run`] 메소드를 사용하는 경우입니다:
-
-```py
-agent.run("Draw me a picture of rivers and lakes.")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200>
-
-수행하려는 작업에 적합한 도구를 자동으로 선택하여 적절하게 실행합니다. 
-동일한 명령어에서 하나 또는 여러 개의 작업을 수행할 수 있습니다
-(다만, 명령어가 복잡할수록 에이전트가 실패할 가능성이 높아집니다).
-
-```py
-agent.run("Draw me a picture of the sea then transform the picture to add an island")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/sea_and_island.png" width=200>
-
-<br/>
-
-
-모든 [`~Agent.run`] 작업은 독립적이므로 다른 작업으로 여러 번 연속해서 실행할 수 있습니다.
-
-`agent`는 큰 언어 모델일 뿐이므로 프롬프트에 약간의 변화를 주면 완전히 다른 결과가 나올 수 있다는 점에 유의하세요. 
-수행하려는 작업을 최대한 명확하게 설명하는 것이 중요합니다. 
-좋은 프롬프트를 작성하는 방법은 [여기](custom_tools#writing-good-user-inputs)에서 자세히 확인할 수 있습니다.
-
-여러 실행에 걸쳐 상태를 유지하거나 텍스트가 아닌 개체를 에이전트에게 전달하려는 경우에는 에이전트가 사용할 변수를 지정할 수 있습니다. 
-예를 들어 강과 호수의 첫 번째 이미지를 생성한 뒤, 
-모델이 해당 그림에 섬을 추가하도록 다음과 같이 요청할 수 있습니다:
-
-```python
-picture = agent.run("Generate a picture of rivers and lakes.")
-updated_picture = agent.run("Transform the image in `picture` to add an island to it.", picture=picture)
-```
-
-<Tip>
-
-이 방법은 모델이 요청을 이해하지 못하고 도구를 혼합할 때 유용할 수 있습니다. 예를 들면 다음과 같습니다:
-
-```py
-agent.run("Draw me the picture of a capybara swimming in the sea")
-```
-
-여기서 모델은 두 가지 방식으로 해석할 수 있습니다:
-- `text-to-image`이 바다에서 헤엄치는 카피바라를 생성하도록 합니다.
-- 또는 `text-to-image`이 카피바라를 생성한 다음 `image-transformation` 도구를 사용하여 바다에서 헤엄치도록 합니다.
-
-첫 번째 시나리오를 강제로 실행하려면 프롬프트를 인수로 전달하여 실행할 수 있습니다:
-
-```py
-agent.run("Draw me a picture of the `prompt`", prompt="a capybara swimming in the sea")
-```
-
-</Tip>
-
-
-### 대화 기반 실행 (chat) [[chat-based-execution-(chat)]]
-
-에이전트는 [`~Agent.chat`] 메소드를 사용하는 대화 기반 접근 방식도 있습니다:
-
-```py
-agent.chat("Generate a picture of rivers and lakes")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200> 
-
-```py
-agent.chat("Transform the picture so that there is a rock in there")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_and_beaver.png" width=200>
-
-<br/>
-
-이 방식은 여러 명령어에 걸쳐 상태를 유지하고자 할 때 흥미로운 접근 방식입니다. 
-실험용으로 더 좋지만 복잡한 명령어보다는 
-단일 명령어([`~Agent.run`] 메소드가 더 잘 처리하는 명령어)에 훨씬 더 잘 작동하는 경향이 있습니다.
-
-이 메소드는 텍스트가 아닌 유형이나 특정 프롬프트를 전달하려는 경우 인수를 받을 수도 있습니다.
-
-### ⚠️ 원격 실행 [[remote-execution]]
-
-데모 목적과 모든 설정에서 사용할 수 있도록 
-에이전트가 접근할 수 있는 몇 가지 기본 도구에 대한 원격 실행기를 만들었습니다. 
-이러한 도구는 [inference endpoints](https://huggingface.co/inference-endpoints)를 사용하여 만들어졌습니다. 
-원격 실행기 도구를 직접 설정하는 방법을 보려면 [사용자 정의 도구 가이드](./custom_tools)를 읽어보시기 바랍니다.
-
-원격 도구로 실행하려면 [`~Agent.run`] 또는 [`~Agent.chat`] 중 하나에 `remote=True`를 지정하기만 하면 됩니다.
-
-예를 들어 다음 명령은 많은 RAM이나 GPU 없이도 모든 장치에서 효율적으로 실행할 수 있습니다:
-
-```py
-agent.run("Draw me a picture of rivers and lakes", remote=True)
-```
-
-[`~Agent.chat`]도 마찬가지입니다:
-
-```py
-agent.chat("Draw me a picture of rivers and lakes", remote=True)
-```
-
-### 여기서 무슨 일이 일어나는 거죠? 도구란 무엇이고, 에이전트란 무엇인가요? [[whats-happening-here-what-are-tools-and-what-are-agents]]
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/diagram.png">
-
-#### 에이전트 [[agents]]
-
-여기서 "에이전트"는 대규모 언어 모델이며, 특정 도구 모음에 접근할 수 있도록 프롬프트하고 있습니다.
-
-LLM은 작은 코드 샘플을 생성하는 데 상당히 능숙하므로,
-이 장점을 활용해 도구 모음을 사용하여 작업을 수행하는 작은 코드 샘플을 제공하라는 메시지를 표시합니다. 
-그런 다음 에이전트에게 제공하는 작업과 제공하는 도구에 대한 설명으로 이 프롬프트가 완료됩니다. 
-이렇게 하면 사용 중인 도구들의 문서에 접근할 수 있으며, 해당 도구들의 입력과 출력을 예상하고, 관련된 코드를 생성할 수 있습니다.
-
-#### 도구 [[tools]]
-
-도구는 매우 간단합니다. 이름과 설명이 있는 단일 기능으로 구성되어 있습니다. 
-그런 다음 이러한 도구의 설명을 사용하여 상담원에게 프롬프트를 표시합니다. 
-이 프롬프트를 통해 상담원에게 쿼리에서 요청된 작업을 수행하기 위해 도구를 활용하는 방법을 보여줍니다.
-
-에이전트가 매우 원자적인 도구를 사용하여 더 나은 코드를 작성하기 때문에 파이프라인이 아닌 완전히 새로운 도구를 사용합니다. 
-파이프라인은 더 많이 리팩터링되며 종종 여러 작업을 하나로 결합합니다. 
-도구는 하나의 매우 간단한 작업에만 집중하도록 되어 있습니다.
-
-#### 코드 실행?! [[code-execution]]
-
-그런 다음 이 코드는 도구와 함께 전달된 입력 세트에 대해 작은 Python 인터프리터를 사용하여 실행됩니다. 
-"임의 코드 실행이라니!"이라고 비명을 지르는 소리가 들리겠지만, 그렇지 않은 이유를 설명하겠습니다.
-
-호출할 수 있는 함수는 제공한 도구와 인쇄 기능뿐이므로 이미 실행할 수 있는 기능이 제한되어 있습니다. 
-Hugging Face 도구로 제한되어 있다면 안전할 것입니다. 
-
-그리고 어트리뷰트 조회나 가져오기를 허용하지 않으므로
-(어차피 작은 함수 집합에 입/출력을 전달할 때는 필요하지 않아야 합니다) 
-가장 명백한 공격(어차피 LLM에 출력하라는 메시지를 표시해야 합니다)은 문제가 되지 않습니다. 
-매우 안전하게 하고 싶다면 추가 인수 return_code=True를 사용하여 run() 메소드를 실행하면 됩니다.
-이 경우 에이전트가 실행할 코드를 반환하고 실행할지 여부를 결정할 수 있습니다.
-
-불법적인 연산을 수행하려고 하거나 에이전트가 생성한 코드에 일반적인 파이썬 오류가 있는 경우 
-실행이 중지됩니다.
-
-### 엄선된 도구 모음 [[a-curated-set-of-tools]]
-
-저희는 이러한 에이전트들의 역량을 강화할 수 있는 일련의 도구를 확인하고 있습니다. 
-다음은 연동된 도구의 최신 목록입니다:
-
-- **문서 질문 답변**: 이미지 형식의 문서(예: PDF)가 주어지면 이 문서에 대한 질문에 답변합니다. ([Donut](./model_doc/donut))
-- **텍스트 질문 답변**: 긴 텍스트와 질문이 주어지면 텍스트에서 질문에 답변합니다. ([Flan-T5](./model_doc/flan-t5))
-- **무조건 이미지 캡셔닝**: 이미지에 캡션을 답니다! ([BLIP](./model_doc/blip))
-- **이미지 질문 답변**: 이미지가 주어지면 이 이미지에 대한 질문에 답변하기. ([VILT](./model_doc/vilt))
-- **이미지 분할**: 이미지와 프롬프트가 주어지면 해당 프롬프트의 분할 마스크를 출력합니다. ([CLIPSeg](./model_doc/clipseg))
-- **음성을 텍스트로 변환**: 사람이 말하는 오디오 녹음이 주어지면 음성을 텍스트로 변환합니다. ([Whisper](./model_doc/whisper))
-- **텍스트 음성 변환**: 텍스트를 음성으로 변환합니다. ([SpeechT5](./model_doc/speecht5))
-- **제로 샷(zero-shot) 텍스트 분류**: 텍스트와 레이블 목록이 주어지면 텍스트와 가장 관련 있는 레이블을 식별합니다. ([BART](./model_doc/bart))
-- **텍스트 요약**: 긴 텍스트를 한 문장 또는 몇 문장으로 요약합니다. ([BART](./model_doc/bart))
-- **번역**: 텍스트를 지정된 언어로 번역합니다. ([NLLB](./model_doc/nllb))
-
-이러한 도구는 트랜스포머에 통합되어 있으며, 예를 들어 수동으로도 사용할 수 있습니다:
-
-```py
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-### 사용자 정의 도구 [[custom-tools]]
-
-엄선된 도구 세트도 있지만, 이 구현이 제공하는 가장 큰 가치는 사용자 지정 도구를 빠르게 만들고 공유할 수 있다는 점입니다.
-
-도구의 코드를 Hugging Face Space나 모델 저장소에 푸시하면 에이전트에게 직접 도구를 활용할 수 있습니다.  [`huggingface-tools` organization](https://huggingface.co/huggingface-tools)에 몇 가지 **트랜스포머에 구애받지 않는** 툴을 추가했습니다:
-
-- **텍스트 다운로더**: 웹 URL에서 텍스트를 다운로드합니다.
-- **텍스트 이미지 변환**: 프롬프트에 따라 이미지를 생성하여 안정적인 확산을 활용합니다.
-- **이미지 변환**: 초기 이미지와 프롬프트가 주어진 이미지를 수정하고, 안정적인 확산을 활용하는 지시 픽셀 2 픽셀을 활용합니다.
-- **텍스트 비디오 변환**: 프롬프트에 따라 작은 비디오를 생성하며, damo-vilab을 활용합니다.
-
-저희가 처음부터 사용하고 있는 텍스트-이미지 변환 도구는 [*huggingface-tools/text-to-image*](https://huggingface.co/spaces/huggingface-tools/text-to-image)에 있는 원격 도구입니다! 저희는 이 도구와 다른 조직에 이러한 도구를 계속 출시하여 이 구현을 더욱 강화할 것입니다.
-
-에이전트는 기본적으로 [`huggingface-tools`](https://huggingface.co/huggingface-tools)에 있는 도구에 접근할 수 있습니다.
-[다음 가이드](custom_tools)에서 도구를 작성하고 공유하는 방법과 Hub에 있는 사용자 지정 도구를 활용하는 방법에 대해 설명합니다.
-
-### 코드 생성[[code-generation]]
-
-지금까지 에이전트를 사용하여 작업을 수행하는 방법을 보여드렸습니다. 하지만 에이전트는 매우 제한된 Python 인터프리터를 사용하여 실행할 코드만 생성하고 있습니다. 다른 설정에서 생성된 코드를 사용하려는 경우 에이전트에게 도구 정의 및 정확한 가져오기와 함께 코드를 반환하라는 메시지를 표시할 수 있습니다.
-
-예를 들어 다음 명령어는 
-```python
-agent.run("Draw me a picture of rivers and lakes", return_code=True)
-```
-
-다음 코드를 반환합니다.
-
-```python
-from transformers import load_tool
-
-image_generator = load_tool("huggingface-tools/text-to-image")
-
-image = image_generator(prompt="rivers and lakes")
-```
-
-이 코드는 직접 수정하고 실행할 수 있습니다.

docs/source/ms/_toctree.yml

@@ -21,8 +21,6 @@
       title: Sediakan latihan yang diedarkan dengan 🤗 Accelerate
     - local: model_sharing
       title: Kongsi model anda
-    - local: transformers_agents
-      title: Ejen
   title: Tutorials
 - sections:
     - sections:
@@ -179,8 +177,6 @@
   title: Panduan konsep
 - sections:
     - sections:
-        - local: main_classes/agent
-          title: Ejen dan Alat
         - local: model_doc/auto
           title: Kelas Auto
         - local: main_classes/callback

docs/source/zh/_toctree.yml

@@ -23,10 +23,6 @@
     title: 使用🤗 PEFT加载和训练adapters
   - local: model_sharing
     title: 分享您的模型
-  - local: agents
-    title: 智能体和工具
-  - local: agents_advanced
-    title: 智能体，超强版 - 多智能体、外部工具等
   - local: llm_tutorial
     title: 使用LLMs进行生成
   title: 教程
@@ -105,8 +101,6 @@
   title: 概念指南
 - sections:
   - sections:
-    - local: main_classes/agent
-      title: 智能体和工具
     - local: main_classes/callback
       title: Callbacks
     - local: main_classes/configuration

docs/source/zh/agents.md

@@ -1,427 +0,0 @@
-<!--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.
-
--->
-# 智能体和工具
-
-[[在colab里打开]]
-
-### 什么是智能体 (Agent)？
-
-大型语言模型（LLM）经过 [因果语言建模训练](./tasks/language_modeling) 可以应对各种任务，但在一些基本任务（如逻辑推理、计算和搜索）上常常表现不佳。当它们被用在自己不擅长的领域时，往往无法生成我们期望的答案。
-
-为了解决这个问题，可以创建**智能体**.
-
-智能体是一个系统，它使用 LLM 作为引擎，并且能够访问称为**工具**的功能。
-
-这些**工具**是执行任务的函数，包含所有必要的描述信息，帮助智能体正确使用它们。
-
-智能体可以被编程为：
-- 一次性设计一系列工具并同时执行它们，像  [`CodeAgent`]
-- 一次执行一个工具，并等待每个工具的结果后再启动下一个，像 [`ReactJsonAgent`]
-
-### 智能体类型
-
-#### 代码智能体
-
-此智能体包含一个规划步骤，然后生成 Python 代码一次性执行所有任务。它原生支持处理不同输入和输出类型，因此推荐用于多模态任务。
-
-#### 推理智能体
-
-这是解决推理任务的首选代理，因为 ReAct 框架 ([Yao et al., 2022](https://huggingface.co/papers/2210.03629)) 使其在基于之前观察进行推理时非常高效。
-
-我们实现了两种版本的 ReactJsonAgent：
-- [`ReactJsonAgent`] 将工具调用作为 JSON 格式输出。
-- [`ReactCodeAgent`] 是 ReactJsonAgent 的一种新型，生成工具调用的代码块，对于具备强大编程能力的 LLM 非常适用。
-
-> [TIP]
-> 阅读 [Open-source LLMs as LangChain Agents](https://huggingface.co/blog/open-source-llms-as-agents) 博文，了解更多关于推理智能体的信息。
-
-<div class="flex justify-center">
-    <img
-        class="block dark:hidden"
-        src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/Agent_ManimCE.gif"
-    />
-    <img
-        class="hidden dark:block"
-        src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/Agent_ManimCE.gif"
-    />
-</div>
-
-![推理智能体的框架](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/open-source-llms-as-agents/ReAct.png)
-
-以下是一个推理代码智能体如何处理以下问题的示例：
-
-```py3
->>> agent.run(
-...     "How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?",
-... )
-=====New task=====
-How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?
-====Agent is executing the code below:
-bert_blocks = search(query="number of blocks in BERT base encoder")
-print("BERT blocks:", bert_blocks)
-====
-Print outputs:
-BERT blocks: twelve encoder blocks
-
-====Agent is executing the code below:
-attention_layer = search(query="number of layers in Attention is All You Need")
-print("Attention layers:", attention_layer)
-====
-Print outputs:
-Attention layers: Encoder: The encoder is composed of a stack of N = 6 identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position- 2 Page 3 Figure 1: The Transformer - model architecture.
-
-====Agent is executing the code below:
-bert_blocks = 12
-attention_layers = 6
-diff = bert_blocks - attention_layers
-print("Difference in blocks:", diff)
-final_answer(diff)
-====
-
-Print outputs:
-Difference in blocks: 6
-
-Final answer: 6
-```
-
-### 如何构建智能体？
-
-要初始化一个智能体，您需要以下参数：
-
-- **一个 LLM** 来驱动智能体——智能体本身并不是 LLM，而是一个使用 LLM 作为引擎的程序。
-- **一个系统提示**：告诉 LLM 引擎应该如何生成输出。
-- **一个工具箱**，智能体可以从中选择工具执行。
-- **一个解析器**，从 LLM 输出中提取出哪些工具需要调用，以及使用哪些参数。
-
-在智能体系统初始化时，工具属性将生成工具描述，并嵌入到智能体的系统提示中，告知智能体可以使用哪些工具，并且为什么使用它们。
-
-**安装依赖**
-
-首先，您需要安装**智能体**所需的额外依赖：
-
-```bash
-pip install transformers[agents]
-```
-**创建LLM引擎**
-
-定义一个 `llm_engine` 方法，该方法接受一系列[消息](./chat_templating)并返回文本。该 `callable` 还需要接受一个 `stop` 参数，用于指示何时停止生成输出。 
-
-```python
-from huggingface_hub import login, InferenceClient
-
-login("<YOUR_HUGGINGFACEHUB_API_TOKEN>")
-
-client = InferenceClient(model="meta-llama/Meta-Llama-3-70B-Instruct")
-
-def llm_engine(messages, stop_sequences=["Task"]) -> str:
-    response = client.chat_completion(messages, stop=stop_sequences, max_tokens=1000)
-    answer = response.choices[0].message.content
-    return answer
-```
-
-您可以使用任何符合以下要求的 `llm_engine` 方法：
-1. [输入格式](./chat_templating)为 (`List[Dict[str, str]]`)，并且返回一个字符串。
-2. 它在 `stop_sequences` 参数传递的序列处停止生成输出。
-
-此外，`llm_engine` 还可以接受一个 `grammar` 参数。如果在智能体初始化时指定了 `grammar`，则该参数将传递给 `llm_engine` 的调用，以允许[受限生成](https://huggingface.co/docs/text-generation-inference/conceptual/guidance)，以强制生成格式正确的智能体输出。
-
-您还需要一个 `tools` 参数，它接受一个 `Tools` 列表 —— 可以是空列表。您也可以通过定义可选参数 `add_base_tools=True` 来将默认工具箱添加到工具列表中。
-
-现在，您可以创建一个智能体，例如 [`CodeAgent`]，并运行它。您还可以创建一个 [`TransformersEngine`]，使用 `transformers` 在本地机器上运行预初始化的推理管道。 为了方便起见，由于智能体行为通常需要更强大的模型，例如 `Llama-3.1-70B-Instruct`，它们目前较难在本地运行，我们还提供了 [`HfApiEngine`] 类，它在底层初始化了一个 `huggingface_hub.InferenceClient`。
-
-```python
-from transformers import CodeAgent, HfApiEngine
-
-llm_engine = HfApiEngine(model="meta-llama/Meta-Llama-3-70B-Instruct")
-agent = CodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-
-agent.run(
-    "Could you translate this sentence from French, say it out loud and return the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-当你急需某个东西时这将会很有用!
-您甚至可以将 `llm_engine` 参数留空，默认情况下会创建一个 [`HfApiEngine`]。
-
-```python
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], add_base_tools=True)
-
-agent.run(
-    "Could you translate this sentence from French, say it out loud and give me the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-请注意，我们使用了额外的 `sentence` 参数：您可以将文本作为附加参数传递给模型。
-
-您还可以使用这个来指定本地或远程文件的路径供模型使用：
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-
-agent.run("Why does Mike not know many people in New York?", audio="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/recording.mp3")
-```
-
-系统提示和输出解析器会自动定义，但您可以通过调用智能体的 `system_prompt_template` 来轻松查看它们。
-
-```python
-print(agent.system_prompt_template)
-```
-
-尽可能清楚地解释您要执行的任务非常重要。 每次 [`~Agent.run`] 操作都是独立的，并且由于智能体是由 LLM 驱动的，提示中的细微变化可能会导致完全不同的结果。 
-您还可以连续运行多个任务，每次都会重新初始化智能体的 `agent.task` 和 `agent.logs` 属性。
-
-
-#### 代码执行
-
-Python 解释器在一组输入和工具上执行代码。 这应该是安全的，因为只能调用您提供的工具（特别是 Hugging Face 的工具）和 print 函数，因此您已经限制了可以执行的操作。
-
-Python 解释器默认不允许导入不在安全列表中的模块，因此大多数明显的攻击问题应该不成问题。 您仍然可以通过在 [`ReactCodeAgent`] 或 [`CodeAgent`] 初始化时通过 `additional_authorized_imports` 参数传递一个授权的模块列表来授权额外的导入：
-
-```py
->>> 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'?")
-
-(...)
-'Hugging Face – Blog'
-```
-
-如果有任何代码尝试执行非法操作，或者生成的代码出现常规 Python 错误，执行将停止。
-
-> [!WARNING]
-> 在使用大语言模型（LLM）生成代码时，生成的代码会被执行，避免导入或使用任何不安全的库或模块。
-
-### 系统提示
-
-智能体，或者说驱动智能体的 LLM，根据系统提示生成输出。系统提示可以定制并根据目标任务进行调整。例如，检查 [`ReactCodeAgent`] 的系统提示（以下版本经过简化）。
-
-```text
-You will be given a task to solve as best you can.
-You have access to the following tools:
-<<tool_descriptions>>
-
-To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
-
-At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task, then the tools that you want to use.
-Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '/End code' sequence.
-During each intermediate step, you can use 'print()' to save whatever important information you will then need.
-These print outputs will then be available in the 'Observation:' field, for using this information as input for the next step.
-
-In the end you have to return a final answer using the `final_answer` tool.
-
-Here are a few examples using notional tools:
----
-{examples}
-
-Above example were using notional tools that might not exist for you. You only have access to those tools:
-<<tool_names>>
-You also can perform computations in the python code you generate.
-
-Always provide a 'Thought:' and a 'Code:\n```py' sequence ending with '```<end_code>' sequence. You MUST provide at least the 'Code:' sequence to move forward.
-
-Remember to not perform too many operations in a single code block! You should split the task into intermediate code blocks.
-Print results at the end of each step to save the intermediate results. Then use final_answer() to return the final result.
-
-Remember to make sure that variables you use are all defined.
-
-Now Begin!
-```
-
-系统提示包括：
-- 解释智能体应该如何工作以及工具的**介绍**。
-- 所有工具的描述由 `<<tool_descriptions>>` 标记在运行时动态替换，这样智能体就知道可以使用哪些工具及其用途。
-    - 工具的描述来自工具的属性,`name`、`description`、`inputs` 和 `output_type`，以及一个简单的 `jinja2` 模板，您可以根据需要进行调整。
-- 期望的输出格式。
-
-您可以通过向 `system_prompt` 参数传递自定义提示来最大程度地提高灵活性，从而覆盖整个系统提示模板。
-
-```python
-from transformers import ReactJsonAgent
-from transformers.agents import PythonInterpreterTool
-
-agent = ReactJsonAgent(tools=[PythonInterpreterTool()], system_prompt="{your_custom_prompt}")
-```
-
-> [WARNING]
-> 必须在`template`中定义 `<<tool_descriptions>>` 这个变量，以便智能体能够正确地识别并使用可用的工具
-
-
-### 检查智能体的运行
-
-以下是检查运行后发生了什么的一些有用属性：
-- `agent.logs` 存储了智能体的详细日志。每一步的所有内容都会存储在一个字典中，然后附加到 `agent.logs`。
-- 运行 `agent.write_inner_memory_from_logs()` 会从日志中创建智能体的内存，以便 LLM 查看，作为一系列聊天消息。此方法会遍历日志的每个步骤，只保存其感兴趣的消息：例如，它会单独保存系统提示和任务，然后为每个步骤保存 LLM 输出的消息，以及工具调用输出的消息。如果您想要更高层次的查看发生了什么，可以使用此方法 —— 但并不是每个日志都会被此方法转录。
-
-## 工具
-
-工具是智能体使用的基本功能。
-
-例如，您可以检查 [`PythonInterpreterTool`]：它有一个名称、描述、输入描述、输出类型和 `__call__` 方法来执行该操作。
-
-当智能体初始化时，工具属性会用来生成工具描述，然后将其嵌入到智能体的系统提示中，这让智能体知道可以使用哪些工具以及为什么使用它们。
-
-### 默认工具箱
-
-Transformers 提供了一个默认工具箱，用于增强智能体，您可以在初始化时通过 `add_base_tools=True` 参数将其添加到智能体中：
-
-- **文档问答**：给定一个文档（如图像格式的 PDF），回答关于该文档的问题([Donut](./model_doc/donut))
-- **图像问答**：给定一张图片，回答关于该图像的问题([VILT](./model_doc/vilt))
-- **语音转文本**：给定一个人讲述的音频录音，将其转录为文本（Whisper）
-- **文本转语音**：将文本转换为语音([SpeechT5](./model_doc/speecht5))
-- **翻译**：将给定的句子从源语言翻译为目标语言
-- **DuckDuckGo 搜索**：使用 `DuckDuckGo` 浏览器进行网络搜索
-- **Python 代码解释器**：在安全环境中运行 LLM 生成的 Python 代码。只有在初始化 [`ReactJsonAgent`] 时将 `add_base_tools=True` 时，代码智能体才会添加此工具，因为基于代码的智能体已经能够原生执行 Python 代码
-
-
-您可以通过调用 [`load_tool`] 函数来手动使用某个工具并执行任务。
-
-
-```python
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-
-### 创建新工具
-
-您可以为 `Hugging Face` 默认工具无法涵盖的用例创建自己的工具。 
-例如，假设我们要创建一个返回在 `Hugging Face Hub` 上某个任务中下载次数最多的模型的工具。
-
-您将从以下代码开始：
-
-```python
-from huggingface_hub import list_models
-
-task = "text-classification"
-
-model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-print(model.id)
-```
-
-这段代码可以很快转换为工具，只需将其包装成一个函数，并添加 `tool` 装饰器：
-
-
-```py
-from transformers import tool
-
-@tool
-def model_download_tool(task: str) -> str:
-    """
-    This is a tool that returns the most downloaded model of a given task on the Hugging Face Hub.
-    It returns the name of the checkpoint.
-
-    Args:
-        task: The task for which
-    """
-    model = next(iter(list_models(filter="text-classification", sort="downloads", direction=-1)))
-    return model.id
-```
-
-该函数需要：
-- 一个清晰的名称。名称通常描述工具的功能。由于代码返回某个任务中下载次数最多的模型，因此我们将其命名为 `model_download_tool`。
-- 对输入和输出进行类型提示
-- 描述，其中包括 "`Args`:" 部分，描述每个参数（这次不需要类型指示，它会从类型提示中获取）。 
-
-所有这些将自动嵌入到智能体的系统提示中，因此请尽量使它们尽可能清晰！
-
-> [TIP]
-> 这个定义格式与 apply_chat_template 中使用的工具模式相同，唯一的区别是添加了 tool 装饰器：可以在我们的工具使用 API 中[了解更多](https://huggingface.co/blog/unified-tool-use#passing-tools-to-a-chat-template).
-
-然后，您可以直接初始化您的智能体：
-```py
-from transformers import CodeAgent
-agent = CodeAgent(tools=[model_download_tool], llm_engine=llm_engine)
-agent.run(
-    "Can you give me the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?"
-)
-```
-
-您将得到以下输出：
-```text
-======== New task ========
-Can you give me the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?
-==== Agent is executing the code below:
-most_downloaded_model = model_download_tool(task="text-to-video")
-print(f"The most downloaded model for the 'text-to-video' task is {most_downloaded_model}.")
-====
-```
-
-输出：
-`"The most downloaded model for the 'text-to-video' task is ByteDance/AnimateDiff-Lightning."`
-
-### 管理智能体的工具箱
-
-如果您已经初始化了一个智能体，但想添加一个新的工具，重新初始化智能体会很麻烦。借助 Transformers，您可以通过添加或替换工具来管理智能体的工具箱。
-
-让我们将 `model_download_tool` 添加到一个仅初始化了默认工具箱的现有智能体中。
-
-```python
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-agent.toolbox.add_tool(model_download_tool)
-```
-现在，我们可以同时使用新工具和之前的文本到语音工具：
-
-```python
-agent.run(
-    "Can you read out loud the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub and return the audio?"
-)
-```
-
-
-| **Audio**                                                                                                                                            |
-|------------------------------------------------------------------------------------------------------------------------------------------------------|
-| <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/damo.wav" type="audio/wav"/> |
-
-
-> [WARNING]
-> 当向一个已经运行良好的代理添加工具时要小心，因为这可能会导致选择偏向你的工具，或者选择已经定义的工具之外的其他工具。
-
-
-使用 agent.toolbox.update_tool() 方法可以替换智能体工具箱中的现有工具。
-如果您的新工具完全替代了现有工具，这非常有用，因为智能体已经知道如何执行该特定任务。
-只需确保新工具遵循与替换工具相同的 API，或者调整系统提示模板，以确保所有使用替换工具的示例都得到更新。
-
-
-### 使用工具集合
-
-您可以通过使用 ToolCollection 对象来利用工具集合，指定您想要使用的工具集合的 slug。
-然后将这些工具作为列表传递给智能体进行初始化，并开始使用它们！
-
-```py
-from transformers import ToolCollection, ReactCodeAgent
-
-image_tool_collection = ToolCollection(collection_slug="huggingface-tools/diffusion-tools-6630bb19a942c2306a2cdb6f")
-agent = ReactCodeAgent(tools=[*image_tool_collection.tools], add_base_tools=True)
-
-agent.run("Please draw me a picture of rivers and lakes.")
-```
-
-为了加速启动，工具仅在智能体调用时加载。
-
-这将生成如下图像：
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png">

docs/source/zh/agents_advanced.md

@@ -1,250 +0,0 @@
-<!--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.
--->
-# 智能体，超强版 - 多智能体、外部工具等
-
-[[open-in-colab]]
-
-### 什么是智能体？
-
-> [!TIP]
-> 如果你是 `transformers.agents` 的新手，请先阅读主文档 [智能体文档 ](./agents).
-在本页面中，我们将重点介绍 `transformers.agents` 的几种高级用法。
-
-## 多智能体
-
-多智能体功能是微软框架 [Autogen](https://huggingface.co/papers/2308.08155) 中引入的。
-它的意思是让多个智能体一起工作来解决任务，而不是只有一个智能体。
-经验表明，在大多数基准测试中，这种方法能带来更好的性能。之所以有更好的性能，原因很简单：对于许多任务，通常我们更愿意让多个单独的单元专注于子任务，而不是让一个系统做所有事情。这里，拥有不同工具集和记忆的多个智能体可以实现高效的专业化。
-
-你可以轻松地用 `transformers.agents` 构建层次化的多智能体系统。
-
-为此，需要将智能体封装在 [`ManagedAgent`] 对象中。这个对象需要 `agent`、`name` 和 `description` 这几个参数，这些信息会嵌入到管理智能体的系统提示中，帮助它知道如何调用这个管理的智能体，就像我们对工具所做的那样。
-
-下面是一个通过使用我们的 [`DuckDuckGoSearchTool`] 创建一个管理特定网络搜索智能体的示例：
-
-
-```py
-from transformers.agents import ReactCodeAgent, HfApiEngine, DuckDuckGoSearchTool, ManagedAgent
-
-llm_engine = HfApiEngine()
-
-web_agent = ReactCodeAgent(tools=[DuckDuckGoSearchTool()], llm_engine=llm_engine)
-
-managed_web_agent = ManagedAgent(
-    agent=web_agent,
-    name="web_search",
-    description="Runs web searches for you. Give it your query as an argument."
-)
-
-manager_agent = ReactCodeAgent(
-    tools=[], llm_engine=llm_engine, managed_agents=[managed_web_agent]
-)
-
-manager_agent.run("Who is the CEO of Hugging Face?")
-```
-
-> [!TIP]
-> 如果你想深入了解如何高效地实现多智能体系统，请查看 [how we pushed our multi-agent system to the top of the GAIA leaderboard](https://huggingface.co/blog/beating-gaia).
-
-## 高级工具使用
-
-### 通过子类化 Tool 来直接定义工具，并将其共享到 Hub
-
-让我们再次使用主文档中的工具示例，我们已经实现了一个 `tool` 装饰器。
-
-如果你需要添加一些变化，比如为工具自定义属性，可以按照更细粒度的方法构建工具：构建一个继承自 [`Tool`] 超类的类。
-
-自定义工具需要：
-- `name` 属性：表示工具本身的名称，通常描述工具的作用。由于代码返回了针对任务下载量最多的模型，我们将其命名为 model_download_counter。
-- `description` 属性：用于填充智能体的系统提示。
-- `inputs` 属性：这是一个包含 "type" 和 "description" 键的字典。它包含了有助于 Python 解释器做出选择的输入信息。
-- `output_type` 属性：指定输出类型。
-- `forward` 方法：其中包含执行推理代码。
-
-`inputs` 和 `output_type` 的类型应当是 [Pydantic 格式](https://docs.pydantic.dev/latest/concepts/json_schema/#generating-json-schema)。
-
-```python
-from transformers import Tool
-from huggingface_hub import list_models
-
-class HFModelDownloadsTool(Tool):
-    name = "model_download_counter"
-    description = """
-    This is a tool that returns the most downloaded model of a given task on the Hugging Face Hub.
-    It returns the name of the checkpoint."""
-
-    inputs = {
-        "task": {
-            "type": "string",
-            "description": "the task category (such as text-classification, depth-estimation, etc)",
-        }
-    }
-    output_type = "string"
-
-    def forward(self, task: str):
-        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-        return model.id
-```
-
-现在，自定义的 `HfModelDownloadsTool` 类已经准备好，可以将其保存到名为 `model_downloads.py` 的文件中，并导入使用。
-
-
-```python
-from model_downloads import HFModelDownloadsTool
-
-tool = HFModelDownloadsTool()
-```
-
-你还可以通过调用 [`~Tool.push_to_hub`] 将自定义工具推送到 Hub。确保你已经为该工具创建了一个仓库，并使用具有读取访问权限的许可。
-
-```python
-tool.push_to_hub("{your_username}/hf-model-downloads")
-```
-
-通过 [`~Tool.load_tool`] 函数加载工具，并将其传递给智能体的 tools 参数。
-
-```python
-from transformers import load_tool, CodeAgent
-
-model_download_tool = load_tool("m-ric/hf-model-downloads")
-```
-
-### 将 Space 导入为工具 🚀
-
-你可以直接通过 [`Tool.from_space`] 方法将 Hub 上的 Space 导入为工具！
-
-只需要提供 Space 在 Hub 上的 ID、名称和描述，帮助智能体理解工具的作用。在幕后，这将使用 [`gradio-client`](https://pypi.org/project/gradio-client/) 库来调用 Space。
-
-例如，下面是从 Hub 导入 `FLUX.1-dev` Space 并用其生成图像的示例：
-
-```
-from transformers import Tool
-image_generation_tool = Tool.from_space(
-    "black-forest-labs/FLUX.1-dev",
-    name="image_generator",
-    description="Generate an image from a prompt")
-image_generation_tool("A sunny beach")
-```
-看！这就是你生成的图像！🏖️
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/sunny_beach.webp">
-
-然后，你可以像使用其他工具一样使用这个工具。例如，改进提示 `穿宇航服的兔子` 并生成其图像：
-
-```python
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[image_generation_tool])
-
-agent.run(
-    "Improve this prompt, then generate an image of it.", prompt='A rabbit wearing a space suit'
-)
-```
-
-```text
-=== Agent thoughts:
-improved_prompt could be "A bright blue space suit wearing rabbit, on the surface of the moon, under a bright orange sunset, with the Earth visible in the background"
-Now that I have improved the prompt, I can use the image generator tool to generate an image based on this prompt.
->>> Agent is executing the code below:
-image = image_generator(prompt="A bright blue space suit wearing rabbit, on the surface of the moon, under a bright orange sunset, with the Earth visible in the background")
-final_answer(image)
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit_spacesuit_flux.webp">
-
-这真酷吧？🤩
-
-### 使用 gradio-tools
-
-[gradio-tools](https://github.com/freddyaboulton/gradio-tools) 是一个强大的库，允许使用 Hugging Face Spaces 作为工具。它支持许多现有的 Spaces，也支持自定义 Spaces。
-
-transformers 支持通过 [`Tool.from_gradio`] 方法使用 `gradio_tools`。例如，下面是如何使用来自 `gradio-tools` 工具包的 [`StableDiffusionPromptGeneratorTool`](https://github.com/freddyaboulton/gradio-tools/blob/main/gradio_tools/tools/prompt_generator.py) 来改进提示，以生成更好的图像：
-
-导入和实例化工具，并将其传递给 `Tool.from_gradio` 方法:
-
-```python
-from gradio_tools import StableDiffusionPromptGeneratorTool
-from transformers import Tool, load_tool, CodeAgent
-
-gradio_prompt_generator_tool = StableDiffusionPromptGeneratorTool()
-prompt_generator_tool = Tool.from_gradio(gradio_prompt_generator_tool)
-```
-
-> [!WARNING]
-> gradio-tools 需要 **文本** 输入和输出，即使在处理像图像和音频这样的不同模态时也是如此。目前，图像和音频的输入输出与此不兼容。
-### 使用 LangChain 工具
-
-我们很喜欢 LangChain，并认为它有一套非常有吸引力的工具。
-要从 LangChain 导入工具，可以使用 `from_langchain()` 方法。
-
-例如，下面是如何使用它来重新创建上面介绍的搜索结果，使用一个 LangChain 网络搜索工具。该工具需要 `pip install google-search-results` 来正常工作。
-
-```python
-from langchain.agents import load_tools
-from transformers import Tool, ReactCodeAgent
-
-search_tool = Tool.from_langchain(load_tools(["serpapi"])[0])
-
-agent = ReactCodeAgent(tools=[search_tool])
-
-agent.run("How many more blocks (also denoted as layers) are in BERT base encoder compared to the encoder from the architecture proposed in Attention is All You Need?")
-```
-
-## 在酷炫的 Gradio 界面中展示智能体运行
-
-你可以利用 `gradio.Chatbot` 来展示智能体的思考过程，通过 `stream_to_gradio`，下面是一个示例：
-
-```py
-import gradio as gr
-from transformers import (
-    load_tool,
-    ReactCodeAgent,
-    HfApiEngine,
-    stream_to_gradio,
-)
-
-# Import tool from Hub
-image_generation_tool = load_tool("m-ric/text-to-image")
-
-llm_engine = HfApiEngine("meta-llama/Meta-Llama-3-70B-Instruct")
-
-# Initialize the agent with the image generation tool
-agent = ReactCodeAgent(tools=[image_generation_tool], llm_engine=llm_engine)
-
-
-def interact_with_agent(task):
-    messages = []
-    messages.append(gr.ChatMessage(role="user", content=task))
-    yield messages
-    for msg in stream_to_gradio(agent, task):
-        messages.append(msg)
-        yield messages + [
-            gr.ChatMessage(role="assistant", content="⏳ Task not finished yet!")
-        ]
-    yield messages
-
-
-with gr.Blocks() as demo:
-    text_input = gr.Textbox(lines=1, label="Chat Message", value="Make me a picture of the Statue of Liberty.")
-    submit = gr.Button("Run illustrator agent!")
-    chatbot = gr.Chatbot(
-        label="Agent",
-        type="messages",
-        avatar_images=(
-            None,
-            "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png",
-        ),
-    )
-    submit.click(interact_with_agent, [text_input], [chatbot])
-
-if __name__ == "__main__":
-    demo.launch()
-```

docs/source/zh/main_classes/agent.md

@@ -1,26 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Agents和工具
-
-<Tip warning={true}>
-
-The Agents framework has significantly changed in version v4.41.0.
-This document has been removed as it was referencing an older API.
-
-We eagerly welcome new contributions for the updated API.
-
-</Tip>

examples/modular-transformers/modeling_new_task_model.py

@@ -19,7 +19,6 @@ from ...utils import (
     add_start_docstrings_to_model_forward,
     replace_return_docstrings,
 )
-from ...utils.deprecation import deprecate_kwarg
 from ..auto import AutoModel, AutoModelForCausalLM
 from .configuration_new_task_model import NewTaskModelConfig
 
@@ -328,7 +327,6 @@ class NewTaskModelForNewTask(NewTaskModelPreTrainedModel, GenerationMixin):
         image_features = image_features / (self.config.text_config.hidden_size**0.5)
         return image_features
 
-    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
     @add_start_docstrings_to_model_forward(NEW_TASK_MODEL_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=NewTaskModelCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
     def forward(

hubconf.py

@@ -1,162 +0,0 @@
-# Copyright 2020 The HuggingFace Team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import os
-import sys
-
-
-SRC_DIR = os.path.join(os.path.dirname(__file__), "src")
-sys.path.append(SRC_DIR)
-
-
-from transformers import (
-    AutoConfig,
-    AutoModel,
-    AutoModelForCausalLM,
-    AutoModelForMaskedLM,
-    AutoModelForQuestionAnswering,
-    AutoModelForSequenceClassification,
-    AutoTokenizer,
-    add_start_docstrings,
-)
-
-
-dependencies = ["torch", "numpy", "tokenizers", "filelock", "requests", "tqdm", "regex", "sentencepiece", "sacremoses", "importlib_metadata", "huggingface_hub"]
-
-
-@add_start_docstrings(AutoConfig.__doc__)
-def config(*args, **kwargs):
-    r"""
-                # Using torch.hub !
-                import torch
-
-                config = torch.hub.load('huggingface/transformers', 'config', 'google-bert/bert-base-uncased')  # Download configuration from huggingface.co and cache.
-                config = torch.hub.load('huggingface/transformers', 'config', './test/bert_saved_model/')  # E.g. config (or model) was saved using `save_pretrained('./test/saved_model/')`
-                config = torch.hub.load('huggingface/transformers', 'config', './test/bert_saved_model/my_configuration.json')
-                config = torch.hub.load('huggingface/transformers', 'config', 'google-bert/bert-base-uncased', output_attentions=True, foo=False)
-                assert config.output_attentions == True
-                config, unused_kwargs = torch.hub.load('huggingface/transformers', 'config', 'google-bert/bert-base-uncased', output_attentions=True, foo=False, return_unused_kwargs=True)
-                assert config.output_attentions == True
-                assert unused_kwargs == {'foo': False}
-
-            """
-
-    return AutoConfig.from_pretrained(*args, **kwargs)
-
-
-@add_start_docstrings(AutoTokenizer.__doc__)
-def tokenizer(*args, **kwargs):
-    r"""
-        # Using torch.hub !
-        import torch
-
-        tokenizer = torch.hub.load('huggingface/transformers', 'tokenizer', 'google-bert/bert-base-uncased')    # Download vocabulary from huggingface.co and cache.
-        tokenizer = torch.hub.load('huggingface/transformers', 'tokenizer', './test/bert_saved_model/')  # E.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`
-
-    """
-
-    return AutoTokenizer.from_pretrained(*args, **kwargs)
-
-
-@add_start_docstrings(AutoModel.__doc__)
-def model(*args, **kwargs):
-    r"""
-            # Using torch.hub !
-            import torch
-
-            model = torch.hub.load('huggingface/transformers', 'model', 'google-bert/bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
-            model = torch.hub.load('huggingface/transformers', 'model', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
-            model = torch.hub.load('huggingface/transformers', 'model', 'google-bert/bert-base-uncased', output_attentions=True)  # Update configuration during loading
-            assert model.config.output_attentions == True
-            # Loading from a TF checkpoint file instead of a PyTorch model (slower)
-            config = AutoConfig.from_pretrained('./tf_model/bert_tf_model_config.json')
-            model = torch.hub.load('huggingface/transformers', 'model', './tf_model/bert_tf_checkpoint.ckpt.index', from_tf=True, config=config)
-
-        """
-
-    return AutoModel.from_pretrained(*args, **kwargs)
-
-
-@add_start_docstrings(AutoModelForCausalLM.__doc__)
-def modelForCausalLM(*args, **kwargs):
-    r"""
-        # Using torch.hub !
-        import torch
-
-        model = torch.hub.load('huggingface/transformers', 'modelForCausalLM', 'openai-community/gpt2')    # Download model and configuration from huggingface.co and cache.
-        model = torch.hub.load('huggingface/transformers', 'modelForCausalLM', './test/saved_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
-        model = torch.hub.load('huggingface/transformers', 'modelForCausalLM', 'openai-community/gpt2', output_attentions=True)  # Update configuration during loading
-        assert model.config.output_attentions == True
-        # Loading from a TF checkpoint file instead of a PyTorch model (slower)
-        config = AutoConfig.from_pretrained('./tf_model/gpt_tf_model_config.json')
-        model = torch.hub.load('huggingface/transformers', 'modelForCausalLM', './tf_model/gpt_tf_checkpoint.ckpt.index', from_tf=True, config=config)
-
-    """
-    return AutoModelForCausalLM.from_pretrained(*args, **kwargs)
-
-
-@add_start_docstrings(AutoModelForMaskedLM.__doc__)
-def modelForMaskedLM(*args, **kwargs):
-    r"""
-            # Using torch.hub !
-            import torch
-
-            model = torch.hub.load('huggingface/transformers', 'modelForMaskedLM', 'google-bert/bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
-            model = torch.hub.load('huggingface/transformers', 'modelForMaskedLM', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
-            model = torch.hub.load('huggingface/transformers', 'modelForMaskedLM', 'google-bert/bert-base-uncased', output_attentions=True)  # Update configuration during loading
-            assert model.config.output_attentions == True
-            # Loading from a TF checkpoint file instead of a PyTorch model (slower)
-            config = AutoConfig.from_pretrained('./tf_model/bert_tf_model_config.json')
-            model = torch.hub.load('huggingface/transformers', 'modelForMaskedLM', './tf_model/bert_tf_checkpoint.ckpt.index', from_tf=True, config=config)
-
-        """
-
-    return AutoModelForMaskedLM.from_pretrained(*args, **kwargs)
-
-
-@add_start_docstrings(AutoModelForSequenceClassification.__doc__)
-def modelForSequenceClassification(*args, **kwargs):
-    r"""
-            # Using torch.hub !
-            import torch
-
-            model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', 'google-bert/bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
-            model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
-            model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', 'google-bert/bert-base-uncased', output_attentions=True)  # Update configuration during loading
-            assert model.config.output_attentions == True
-            # Loading from a TF checkpoint file instead of a PyTorch model (slower)
-            config = AutoConfig.from_pretrained('./tf_model/bert_tf_model_config.json')
-            model = torch.hub.load('huggingface/transformers', 'modelForSequenceClassification', './tf_model/bert_tf_checkpoint.ckpt.index', from_tf=True, config=config)
-
-        """
-
-    return AutoModelForSequenceClassification.from_pretrained(*args, **kwargs)
-
-
-@add_start_docstrings(AutoModelForQuestionAnswering.__doc__)
-def modelForQuestionAnswering(*args, **kwargs):
-    r"""
-        # Using torch.hub !
-        import torch
-
-        model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', 'google-bert/bert-base-uncased')    # Download model and configuration from huggingface.co and cache.
-        model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
-        model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', 'google-bert/bert-base-uncased', output_attentions=True)  # Update configuration during loading
-        assert model.config.output_attentions == True
-        # Loading from a TF checkpoint file instead of a PyTorch model (slower)
-        config = AutoConfig.from_pretrained('./tf_model/bert_tf_model_config.json')
-        model = torch.hub.load('huggingface/transformers', 'modelForQuestionAnswering', './tf_model/bert_tf_checkpoint.ckpt.index', from_tf=True, config=config)
-
-    """
-    return AutoModelForQuestionAnswering.from_pretrained(*args, **kwargs)

i18n/README_ar.md

@@ -245,7 +245,7 @@ limitations under the License.
 
 ### باستخدام pip
 
-تم اختبار هذا المستودع على Python 3.9+، Flax 0.4.1+، PyTorch 2.0+، و TensorFlow 2.6+.
+تم اختبار هذا المستودع على Python 3.9+، Flax 0.4.1+، PyTorch 2.1+، و TensorFlow 2.6+.
 
 يجب تثبيت 🤗 Transformers في [بيئة افتراضية](https://docs.python.org/3/library/venv.html). إذا كنت غير معتاد على البيئات الافتراضية Python، فراجع [دليل المستخدم](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 

i18n/README_de.md

@@ -246,7 +246,7 @@ Das Modell selbst ist ein reguläres [PyTorch `nn.Module`](https://pytorch.org/d
 
 ### Mit pip
 
-Dieses Repository wurde mit Python 3.9+, Flax 0.4.1+, PyTorch 2.0+ und TensorFlow 2.6+ getestet.
+Dieses Repository wurde mit Python 3.9+, Flax 0.4.1+, PyTorch 2.1+ und TensorFlow 2.6+ getestet.
 
 Sie sollten 🤗 Transformers in einer [virtuellen Umgebung](https://docs.python.org/3/library/venv.html) installieren. Wenn Sie mit virtuellen Python-Umgebungen nicht vertraut sind, schauen Sie sich den [Benutzerleitfaden](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) an.
 

i18n/README_es.md

@@ -222,7 +222,7 @@ El modelo en si es un [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.h
 
 ### Con pip
 
-Este repositorio está probado en Python 3.9+, Flax 0.4.1+, PyTorch 2.0+ y TensorFlow 2.6+.
+Este repositorio está probado en Python 3.9+, Flax 0.4.1+, PyTorch 2.1+ y TensorFlow 2.6+.
 
 Deberías instalar 🤗 Transformers en un [entorno virtual](https://docs.python.org/3/library/venv.html). Si no estas familiarizado con los entornos virtuales de Python, consulta la [guía de usuario](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 

i18n/README_fr.md

@@ -243,7 +243,7 @@ Le modèle lui-même est un module [`nn.Module` PyTorch](https://pytorch.org/doc
 
 ### Avec pip
 
-Ce référentiel est testé sur Python 3.9+, Flax 0.4.1+, PyTorch 2.0+ et TensorFlow 2.6+.
+Ce référentiel est testé sur Python 3.9+, Flax 0.4.1+, PyTorch 2.1+ et TensorFlow 2.6+.
 
 Vous devriez installer 🤗 Transformers dans un [environnement virtuel](https://docs.python.org/3/library/venv.html). Si vous n'êtes pas familier avec les environnements virtuels Python, consultez le [guide utilisateur](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 

i18n/README_hd.md

@@ -198,7 +198,7 @@ checkpoint: जाँच बिंदु
 
 ### पिप का उपयोग करना
 
-इस रिपॉजिटरी का परीक्षण Python 3.9+, Flax 0.4.1+, PyTorch 2.0+ और TensorFlow 2.6+ के तहत किया गया है।
+इस रिपॉजिटरी का परीक्षण Python 3.9+, Flax 0.4.1+, PyTorch 2.1+ और TensorFlow 2.6+ के तहत किया गया है।
 
 आप [वर्चुअल एनवायरनमेंट](https://docs.python.org/3/library/venv.html) में 🤗 ट्रांसफॉर्मर इंस्टॉल कर सकते हैं। यदि आप अभी तक पायथन के वर्चुअल एनवायरनमेंट से परिचित नहीं हैं, तो कृपया इसे [उपयोगकर्ता निर्देश](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) पढ़ें।
 

i18n/README_ja.md

@@ -256,7 +256,7 @@ Hugging Faceチームによって作られた **[トランスフォーマーを
 
 ### pipにて
 
-このリポジトリは、Python 3.9+, Flax 0.4.1+, PyTorch 2.0+, TensorFlow 2.6+ でテストされています。
+このリポジトリは、Python 3.9+, Flax 0.4.1+, PyTorch 2.1+, TensorFlow 2.6+ でテストされています。
 
 🤗Transformersは[仮想環境](https://docs.python.org/3/library/venv.html)にインストールする必要があります。Pythonの仮想環境に慣れていない場合は、[ユーザーガイド](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)を確認してください。
 

i18n/README_ko.md

@@ -242,7 +242,7 @@ Transformers에 달린 100,000개의 별을 축하하기 위해, 우리는 커
 
 ### pip로 설치하기
 
-이 저장소는 Python 3.9+, Flax 0.4.1+, PyTorch 2.0+, TensorFlow 2.6+에서 테스트 되었습니다.
+이 저장소는 Python 3.9+, Flax 0.4.1+, PyTorch 2.1+, TensorFlow 2.6+에서 테스트 되었습니다.
 
 [가상 환경](https://docs.python.org/3/library/venv.html)에 🤗 Transformers를 설치하세요. Python 가상 환경에 익숙하지 않다면, [사용자 가이드](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)를 확인하세요.
 

i18n/README_pt-br.md

@@ -253,7 +253,7 @@ O modelo em si é um [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.ht
 
 ### Com pip
 
-Este repositório é testado no Python 3.9+, Flax 0.4.1+, PyTorch 2.0+ e TensorFlow 2.6+.
+Este repositório é testado no Python 3.9+, Flax 0.4.1+, PyTorch 2.1+ e TensorFlow 2.6+.
 
 Você deve instalar o 🤗 Transformers em um [ambiente virtual](https://docs.python.org/3/library/venv.html). Se você não está familiarizado com ambientes virtuais em Python, confira o [guia do usuário](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 

i18n/README_ru.md

@@ -244,7 +244,7 @@ Hugging Face Hub. Мы хотим, чтобы Transformers позволил ра
 
 ### С помощью pip
 
-Данный репозиторий протестирован на Python 3.9+, Flax 0.4.1+, PyTorch 2.0+ и TensorFlow 2.6+.
+Данный репозиторий протестирован на Python 3.9+, Flax 0.4.1+, PyTorch 2.1+ и TensorFlow 2.6+.
 
 Устанавливать 🤗 Transformers следует в [виртуальной среде](https://docs.python.org/3/library/venv.html). Если вы не знакомы с виртуальными средами Python, ознакомьтесь с [руководством пользователя](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 

i18n/README_te.md

@@ -246,7 +246,7 @@ limitations under the License.
 
 ### పిప్ తో
 
-ఈ రిపోజిటరీ పైథాన్ 3.9+, ఫ్లాక్స్ 0.4.1+, PyTorch 2.0+ మరియు TensorFlow 2.6+లో పరీక్షించబడింది.
+ఈ రిపోజిటరీ పైథాన్ 3.9+, ఫ్లాక్స్ 0.4.1+, PyTorch 2.1+ మరియు TensorFlow 2.6+లో పరీక్షించబడింది.
 
 మీరు [వర్చువల్ వాతావరణం](https://docs.python.org/3/library/venv.html)లో 🤗 ట్రాన్స్‌ఫార్మర్‌లను ఇన్‌స్టాల్ చేయాలి. మీకు పైథాన్ వర్చువల్ పరిసరాల గురించి తెలియకుంటే, [యూజర్ గైడ్](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) చూడండి.