run

.circleci/config.yml

@@ -217,9 +217,6 @@ workflows:
         when:
             not: <<pipeline.parameters.nightly>>
         jobs:
-            - check_circleci_user
-            - check_code_quality
-            - check_repository_consistency
             - fetch_tests
 
     nightly:

.circleci/create_circleci_config.py

@@ -127,8 +127,6 @@ class CircleCIJob:
             },
         ]
         steps.extend([{"run": l} for l in self.install_steps])
-        steps.extend([{"run": 'pip install "fsspec>=2023.5.0,<2023.10.0"'}])
-        steps.extend([{"run": "pip install pytest-subtests"}])
         steps.append(
             {
                 "save_cache": {
@@ -539,15 +537,15 @@ doc_test_job = CircleCIJob(
 )
 
 REGULAR_TESTS = [
-    torch_and_tf_job,
-    torch_and_flax_job,
+    # torch_and_tf_job,
+    # torch_and_flax_job,
     torch_job,
-    tf_job,
-    flax_job,
-    custom_tokenizers_job,
-    hub_job,
-    onnx_job,
-    exotic_models_job,
+    # tf_job,
+    # flax_job,
+    # custom_tokenizers_job,
+    # hub_job,
+    # onnx_job,
+    # exotic_models_job,
 ]
 EXAMPLES_TESTS = [
     examples_torch_job,
@@ -556,7 +554,7 @@ EXAMPLES_TESTS = [
 ]
 PIPELINE_TESTS = [
     pipelines_torch_job,
-    pipelines_tf_job,
+    # pipelines_tf_job,
 ]
 REPO_UTIL_TESTS = [repo_utils_job]
 DOC_TESTS = [doc_test_job]
@@ -625,8 +623,8 @@ def create_circleci_config(folder=None):
             else:
                 job.tests_to_run = [f for f in example_tests.split(" ") if f.startswith(f"examples/{framework}")]
 
-            if len(job.tests_to_run) > 0:
-                jobs.append(job)
+            # if len(job.tests_to_run) > 0:
+            #     jobs.append(job)
 
     doctest_file = os.path.join(folder, "doctest_list.txt")
     if os.path.exists(doctest_file):
@@ -634,12 +632,12 @@ def create_circleci_config(folder=None):
             doctest_list = f.read()
     else:
         doctest_list = []
-    if len(doctest_list) > 0:
-        jobs.extend(DOC_TESTS)
+    # if len(doctest_list) > 0:
+    #     jobs.extend(DOC_TESTS)
 
     repo_util_file = os.path.join(folder, "test_repo_utils.txt")
-    if os.path.exists(repo_util_file) and os.path.getsize(repo_util_file) > 0:
-        jobs.extend(REPO_UTIL_TESTS)
+    # if os.path.exists(repo_util_file) and os.path.getsize(repo_util_file) > 0:
+    #     jobs.extend(REPO_UTIL_TESTS)
 
     if len(jobs) == 0:
         jobs = [EmptyJob()]

.github/workflows/build_documentation.yml

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

.github/workflows/build_pr_documentation.yml

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

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

@@ -21,8 +21,36 @@ env:
   RUN_PT_TF_CROSS_TESTS: 1
 
 jobs:
+  check_runner_status:
+    name: Check Runner Status
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout transformers
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 2
+
+      - name: Check Runner Status
+        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-past-ci-runner-docker,multi-gpu-past-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+
+  check_runners:
+    name: Check Runners
+    needs: check_runner_status
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
+    container:
+      image: huggingface/transformers-all-latest-torch-nightly-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
   setup:
     name: Setup
+    needs: check_runners
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
@@ -248,6 +276,8 @@ jobs:
     runs-on: ubuntu-latest
     if: always()
     needs: [
+      check_runner_status,
+      check_runners,
       setup,
       run_tests_single_gpu,
       run_tests_multi_gpu,
@@ -258,6 +288,8 @@ jobs:
         shell: bash
         # For the meaning of these environment variables, see the job `Setup`
         run: |
+          echo "Runner availability: ${{ needs.check_runner_status.result }}"
+          echo "Runner status: ${{ needs.check_runners.result }}"
           echo "Setup status: ${{ needs.setup.result }}"
 
       - uses: actions/checkout@v3
@@ -271,6 +303,8 @@ jobs:
           CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
           CI_EVENT: Nightly CI
+          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
+          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
           SETUP_STATUS: ${{ needs.setup.result }}
         # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
         # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.

.github/workflows/self-past.yml

@@ -32,8 +32,36 @@ env:
   RUN_PT_TF_CROSS_TESTS: 1
 
 jobs:
+  check_runner_status:
+    name: Check Runner Status
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout transformers
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 2
+
+      - name: Check Runner Status
+        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-past-ci-runner-docker,multi-gpu-past-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+
+  check_runners:
+    name: Check Runners
+    needs: check_runner_status
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
+    container:
+      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
   setup:
     name: Setup
+    needs: check_runners
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
@@ -291,6 +319,8 @@ jobs:
     runs-on: ubuntu-latest
     if: always()
     needs: [
+      check_runner_status,
+      check_runners,
       setup,
       run_tests_single_gpu,
       run_tests_multi_gpu,
@@ -301,6 +331,8 @@ jobs:
         shell: bash
         # For the meaning of these environment variables, see the job `Setup`
         run: |
+          echo "Runner availability: ${{ needs.check_runner_status.result }}"
+          echo "Runner status: ${{ needs.check_runners.result }}"
           echo "Setup status: ${{ needs.setup.result }}"
 
       - uses: actions/checkout@v3
@@ -319,6 +351,8 @@ jobs:
           CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
           CI_EVENT: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
+          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
           SETUP_STATUS: ${{ needs.setup.result }}
         # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
         # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.

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

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

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

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

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

@@ -1,11 +1,21 @@
 name: Self-hosted runner AMD GPU (push)
 
 on:
-  workflow_call:
-    inputs:
-      gpu_flavor:
-        required: true
-        type: string
+  workflow_run:
+    workflows: ["Self-hosted runner (push-caller)"]
+    branches: ["main"]
+    types: [completed]
+  push:
+    branches:
+      - ci_*
+      - ci-*
+    paths:
+      - "src/**"
+      - "tests/**"
+      - ".github/**"
+      - "templates/**"
+      - "utils/**"
+  repository_dispatch:
 
 env:
   HF_HOME: /mnt/cache
@@ -35,7 +45,8 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+        gpu_flavor: [mi210]
+    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ matrix.gpu_flavor }}']
     container:
       image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
       options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -54,7 +65,8 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+        gpu_flavor: [mi210]
+    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ matrix.gpu_flavor }}']
     container:
       image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
       options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -152,7 +164,8 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup_gpu.outputs.matrix) }}
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+        gpu_flavor: [mi210]
+    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ matrix.gpu_flavor }}']
     container:
       image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
       options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -308,7 +321,7 @@ jobs:
           CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
           CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: Push CI (AMD) - ${{ inputs.gpu_flavor }}
+          CI_EVENT: push
           CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
           CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
           CI_SHA: ${{ env.CI_SHA }}

.github/workflows/self-push.yml

@@ -27,8 +27,36 @@ env:
   RUN_PT_TF_CROSS_TESTS: 1
 
 jobs:
+  check_runner_status:
+    name: Check Runner Status
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout transformers
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 2
+
+      - name: Check Runner Status
+        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-ci-runner-docker,multi-gpu-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+
+  check_runners:
+    name: Check Runners
+    needs: check_runner_status
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
+    container:
+      image: huggingface/transformers-all-latest-gpu-push-ci
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
   setup:
     name: Setup
+    needs: check_runners
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
@@ -493,6 +521,8 @@ jobs:
     runs-on: ubuntu-latest
     if: always()
     needs: [
+        check_runner_status,
+        check_runners,
         setup,
         run_tests_single_gpu,
         run_tests_multi_gpu,
@@ -504,7 +534,9 @@ jobs:
         shell: bash
         # For the meaning of these environment variables, see the job `Setup`
         run: |
+          echo "Runner availability: ${{ needs.check_runner_status.result }}"
           echo "Setup status: ${{ needs.setup.result }}"
+          echo "Runner status: ${{ needs.check_runners.result }}"
 
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
       # We also take into account the `push` event (we might want to test some changes in a branch)
@@ -557,6 +589,8 @@ jobs:
           CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
           CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
           CI_SHA: ${{ env.CI_SHA }}
+          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
+          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
           SETUP_STATUS: ${{ needs.setup.result }}
 
         # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change

.github/workflows/self-scheduled.yml

@@ -25,8 +25,36 @@ env:
   RUN_PT_TF_CROSS_TESTS: 1
 
 jobs:
+  check_runner_status:
+    name: Check Runner Status
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout transformers
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 2
+
+      - name: Check Runner Status
+        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-scheduled-ci-runner-docker,multi-gpu-scheduled-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+
+  check_runners:
+    name: Check Runners
+    needs: check_runner_status
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    container:
+      image: huggingface/transformers-all-latest-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
   setup:
     name: Setup
+    needs: check_runners
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
@@ -402,6 +430,8 @@ jobs:
     runs-on: ubuntu-latest
     if: always()
     needs: [
+      check_runner_status,
+      check_runners,
       setup,
       run_tests_single_gpu,
       run_tests_multi_gpu,
@@ -450,6 +480,8 @@ jobs:
     runs-on: ubuntu-latest
     if: always()
     needs: [
+      check_runner_status,
+      check_runners,
       setup,
       run_tests_single_gpu,
       run_tests_multi_gpu,
@@ -464,6 +496,8 @@ jobs:
         shell: bash
         # For the meaning of these environment variables, see the job `Setup`
         run: |
+          echo "Runner availability: ${{ needs.check_runner_status.result }}"
+          echo "Runner status: ${{ needs.check_runners.result }}"
           echo "Setup status: ${{ needs.setup.result }}"
 
       - uses: actions/checkout@v3
@@ -479,6 +513,8 @@ jobs:
           CI_EVENT: scheduled
           CI_SHA: ${{ github.sha }}
           CI_WORKFLOW_REF: ${{ github.workflow_ref }}
+          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
+          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
           SETUP_STATUS: ${{ needs.setup.result }}
         # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
         # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.

CONTRIBUTING.md

@@ -40,8 +40,8 @@ There are several ways you can contribute to 🤗 Transformers:
 
 If you don't know where to start, there is a special [Good First
 Issue](https://github.com/huggingface/transformers/contribute) listing. It will give you a list of
-open issues that are beginner-friendly and help you start contributing to open-source. Just comment on the issue that you'd like to work
-on. 
+open issues that are beginner-friendly and help you start contributing to open-source. Just comment in the issue that you'd like to work
+on it. 
 
 For something slightly more challenging, you can also take a look at the [Good Second Issue](https://github.com/huggingface/transformers/labels/Good%20Second%20Issue) list. In general though, if you feel like you know what you're doing, go for it and we'll help you get there! 🚀
 
@@ -62,7 +62,7 @@ feedback.
 The 🤗 Transformers library is robust and reliable thanks to users who report the problems they encounter.
 
 Before you report an issue, we would really appreciate it if you could **make sure the bug was not
-already reported** (use the search bar on GitHub under Issues). Your issue should also be related to bugs in the library itself, and not your code. If you're unsure whether the bug is in your code or the library, please ask in the [forum](https://discuss.huggingface.co/) first. This helps us respond quicker to fixing issues related to the library versus general questions.
+already reported** (use the search bar on GitHub under Issues). Your issue should also be related to bugs in the library itself, and not your code. If you're unsure whether the bug is in your code or the library, please ask on the [forum](https://discuss.huggingface.co/) first. This helps us respond quicker to fixing issues related to the library versus general questions.
 
 Once you've confirmed the bug hasn't already been reported, please include the following information in your issue so we can quickly resolve it:
 
@@ -105,7 +105,7 @@ We have added [templates](https://github.com/huggingface/transformers/tree/main/
 
 New models are constantly released and if you want to implement a new model, please provide the following information
 
-* A short description of the model and a link to the paper.
+* A short description of the model and link to the paper.
 * Link to the implementation if it is open-sourced.
 * Link to the model weights if they are available.
 
@@ -172,7 +172,7 @@ You'll need **[Python 3.8]((https://github.com/huggingface/transformers/blob/mai
 
    which should be enough for most use cases.
 
-5. Develop the features in your branch.
+5. Develop the features on your branch.
 
    As you work on your code, you should make sure the test suite
    passes. Run the tests impacted by your changes like this:
@@ -208,7 +208,7 @@ You'll need **[Python 3.8]((https://github.com/huggingface/transformers/blob/mai
    make quality
    ```
 
-   Finally, we have a lot of scripts to make sure we don't forget to update
+   Finally, we have a lot of scripts to make sure we didn't forget to update
    some files when adding a new model. You can run these scripts with:
 
    ```bash
@@ -218,7 +218,7 @@ You'll need **[Python 3.8]((https://github.com/huggingface/transformers/blob/mai
    To learn more about those checks and how to fix any issues with them, check out the
    [Checks on a Pull Request](https://huggingface.co/docs/transformers/pr_checks) guide.
 
-   If you're modifying documents under the `docs/source` directory, make sure the documentation can still be built. This check will also run in the CI when you open a pull request. To run a local check
+   If you're modifying documents under `docs/source` directory, make sure the documentation can still be built. This check will also run in the CI when you open a pull request. To run a local check
    make sure you install the documentation builder:
    
    ```bash
@@ -234,7 +234,7 @@ You'll need **[Python 3.8]((https://github.com/huggingface/transformers/blob/mai
    This will build the documentation in the `~/tmp/test-build` folder where you can inspect the generated
    Markdown files with your favorite editor. You can also preview the docs on GitHub when you open a pull request.
 
-   Once you're happy with your changes, add the changed files with `git add` and
+   Once you're happy with your changes, add changed files with `git add` and
    record your changes locally with `git commit`:
 
    ```bash
@@ -261,7 +261,7 @@ You'll need **[Python 3.8]((https://github.com/huggingface/transformers/blob/mai
 
    If you've already opened a pull request, you'll need to force push with the `--force` flag. Otherwise, if the pull request hasn't been opened yet, you can just push your changes normally.
 
-6. Now you can go to your fork of the repository on GitHub and click on **Pull Request** to open a pull request. Make sure you tick off all the boxes on our [checklist](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md/#pull-request-checklist) below. When you're ready, you can send your changes to the project maintainers for review.
+6. Now you can go to your fork of the repository on GitHub and click on **Pull request** to open a pull request. Make sure you tick off all the boxes in our [checklist](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md/#pull-request-checklist) below. When you're ready, you can send your changes to the project maintainers for review.
 
 7. It's ok if maintainers request changes, it happens to our core contributors
    too! So everyone can see the changes in the pull request, work in your local

README.md

@@ -52,9 +52,7 @@ limitations under the License.
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_pt-br.md">Рortuguês</a> |
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a>
     </p>
 </h4>
 
@@ -70,7 +68,7 @@ limitations under the License.
 
 These models can be applied on:
 
-* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, and text generation, in over 100 languages.
+* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, text generation, in over 100 languages.
 * 🖼️ Images, for tasks like image classification, object detection, and segmentation.
 * 🗣️ Audio, for tasks like speech recognition and audio classification.
 
@@ -148,7 +146,7 @@ To immediately use a model on a given input (text, image, audio, ...), we provid
 [{'label': 'POSITIVE', 'score': 0.9996980428695679}]
 ```
 
-The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here, the answer is "positive" with a confidence of 99.97%.
+The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here the answer is "positive" with a confidence of 99.97%.
 
 Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in computer vision and speech. For example, we can easily extract detected objects in an image:
 
@@ -182,7 +180,7 @@ Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in compute
   'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
 ```
 
-Here, we get a list of objects detected in the image, with a box surrounding the object and a confidence score. Here is the original image on the left, with the predictions displayed on the right:
+Here we get a list of objects detected in the image, with a box surrounding the object and a confidence score. Here is the original image on the left, with the predictions displayed on the right:
 
 <h3 align="center">
     <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
@@ -213,7 +211,7 @@ And here is the equivalent code for TensorFlow:
 >>> outputs = model(**inputs)
 ```
 
-The tokenizer is responsible for all the preprocessing the pretrained model expects and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
+The tokenizer is responsible for all the preprocessing the pretrained model expects, and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
 
 The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use as usual. [This tutorial](https://huggingface.co/docs/transformers/training) explains how to integrate such a model into a classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune on a new dataset.
 
@@ -233,7 +231,7 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 1. Choose the right framework for every part of a model's lifetime:
     - Train state-of-the-art models in 3 lines of code.
     - Move a single model between TF2.0/PyTorch/JAX frameworks at will.
-    - Seamlessly pick the right framework for training, evaluation, and production.
+    - Seamlessly pick the right framework for training, evaluation and production.
 
 1. Easily customize a model or an example to your needs:
     - We provide examples for each architecture to reproduce the results published by its original authors.
@@ -244,19 +242,19 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 
 - This library is not a modular toolbox of building blocks for neural nets. The code in the model files is not refactored with additional abstractions on purpose, so that researchers can quickly iterate on each of the models without diving into additional abstractions/files.
 - The training API is not intended to work on any model but is optimized to work with the models provided by the library. For generic machine learning loops, you should use another library (possibly, [Accelerate](https://huggingface.co/docs/accelerate)).
-- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/main/examples) are just that: examples. It is expected that they won't work out-of-the-box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
+- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/main/examples) are just that: examples. It is expected that they won't work out-of-the box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
 
 ## Installation
 
 ### With pip
 
-This repository is tested on Python 3.8+, Flax 0.4.1+, PyTorch 1.10+, and TensorFlow 2.6+.
+This repository is tested on Python 3.8+, Flax 0.4.1+, PyTorch 1.10+ and TensorFlow 2.6+.
 
 You should install 🤗 Transformers in a [virtual environment](https://docs.python.org/3/library/venv.html). If you're unfamiliar with Python virtual environments, check out the [user guide](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 
 First, create a virtual environment with the version of Python you're going to use and activate it.
 
-Then, you will need to install at least one of Flax, PyTorch, or TensorFlow.
+Then, you will need to install at least one of Flax, PyTorch or TensorFlow.
 Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax](https://github.com/google/flax#quick-install) and [Jax](https://github.com/google/jax#installation) installation pages regarding the specific installation command for your platform.
 
 When one of those backends has been installed, 🤗 Transformers can be installed using pip as follows:
@@ -283,7 +281,7 @@ Follow the installation pages of Flax, PyTorch or TensorFlow to see how to insta
 
 ## Model architectures
 
-**[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co/models), where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
+**[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co/models) where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
 
 Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
 
@@ -295,11 +293,11 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
 1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
 1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov, and Luke Zettlemoyer.
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
 1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
 1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
 1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova.
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
 1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
 1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
 1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
@@ -364,14 +362,13 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://openai.com/research/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
 1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
 1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
 1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://openai.com/research/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
 1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
 1. **[GPT-Sw3](https://huggingface.co/docs/transformers/model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
 1. **[GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
@@ -386,7 +383,6 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
@@ -395,7 +391,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
 1. **[LiLT](https://huggingface.co/docs/transformers/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
 1. **[LLaMA](https://huggingface.co/docs/transformers/model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
-1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
+1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/XXX) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
 1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
 1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
 1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
@@ -434,10 +430,9 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
@@ -461,7 +456,6 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng), released on [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
@@ -500,7 +494,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
 1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
 1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) released with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
+1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) rreleased with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
 1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
@@ -522,7 +516,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
 1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-1. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedback before starting your PR.
+1. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
 To check if each model has an implementation in Flax, PyTorch or TensorFlow, or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/docs/transformers/index#supported-frameworks).
 

README_es.md

@@ -46,8 +46,7 @@ limitations under the License.
         <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
         <b>Español</b> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
     </p>
 </h4>
 
@@ -339,7 +338,6 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -361,7 +359,6 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
@@ -409,10 +406,9 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
@@ -436,7 +432,6 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng) released with the paper [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
@@ -531,4 +526,4 @@ Ahora nosotros tenemos un [papel](https://www.aclweb.org/anthology/2020.emnlp-de
     url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
     pages = "38--45"
 }
-```
+```

README_hd.md

@@ -72,7 +72,6 @@ checkpoint: जाँच बिंदु
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <b>हिन्दी</b> |
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
     </p>
 </h4>
 
@@ -86,13 +85,13 @@ checkpoint: जाँच बिंदु
 
 🤗 Transformers 100 से अधिक भाषाओं में पाठ वर्गीकरण, सूचना निष्कर्षण, प्रश्न उत्तर, सारांशीकरण, अनुवाद, पाठ निर्माण का समर्थन करने के लिए हजारों पूर्व-प्रशिक्षित मॉडल प्रदान करता है। इसका उद्देश्य सबसे उन्नत एनएलपी तकनीक को सभी के लिए सुलभ बनाना है।
 
-🤗 Transformers त्वरित डाउनलोड और उपयोग के लिए एक एपीआई प्रदान करता है, जिससे आप किसी दिए गए पाठ पर एक पूर्व-प्रशिक्षित मॉडल ले सकते हैं, इसे अपने डेटासेट पर ठीक कर सकते हैं और इसे [मॉडल हब](https://huggingface.co/models) के माध्यम से समुदाय के साथ साझा कर सकते हैं। इसी समय, प्रत्येक परिभाषित पायथन मॉड्यूल पूरी तरह से स्वतंत्र है, जो संशोधन और तेजी से अनुसंधान प्रयोगों के लिए सुविधाजनक है।
+🤗 Transformers त्वरित डाउनलोड और उपयोग के लिए एक एपीआई प्रदान करता है, जिससे आप किसी दिए गए पाठ पर एक पूर्व-प्रशिक्षित मॉडल ले सकते हैं, इसे अपने डेटासेट पर ठीक कर सकते हैं और इसे [मॉडल हब] (https://huggingface.co/models) के माध्यम से समुदाय के साथ साझा कर सकते हैं। ) . इसी समय, प्रत्येक परिभाषित पायथन मॉड्यूल पूरी तरह से स्वतंत्र है, जो संशोधन और तेजी से अनुसंधान प्रयोगों के लिए सुविधाजनक है।
 
 🤗 Transformers तीन सबसे लोकप्रिय गहन शिक्षण पुस्तकालयों का समर्थन करता है： [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — और इसके साथ निर्बाध रूप से एकीकृत होता है। आप अपने मॉडल को सीधे एक ढांचे के साथ प्रशिक्षित कर सकते हैं और दूसरे के साथ लोड और अनुमान लगा सकते हैं।
 
 ## ऑनलाइन डेमो
 
-आप सबसे सीधे मॉडल पृष्ठ पर परीक्षण कर सकते हैं [model hub](https://huggingface.co/models) मॉडल पर। हम [निजी मॉडल होस्टिंग, मॉडल संस्करण, और अनुमान एपीआई](https://huggingface.co/pricing) भी प्रदान करते हैं।。
+आप सबसे सीधे मॉडल पृष्ठ पर परीक्षण कर सकते हैं [model hub](https://huggingface.co/models) मॉडल पर। हम [निजी मॉडल होस्टिंग, मॉडल संस्करण, और अनुमान एपीआई] भी प्रदान करते हैं।(https://huggingface.co/pricing)。
 
 यहाँ कुछ उदाहरण हैं：
 - [शब्द को भरने के लिए मास्क के रूप में BERT का प्रयोग करें](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
@@ -166,7 +165,7 @@ checkpoint: जाँच बिंदु
 
 टोकननाइज़र सभी पूर्व-प्रशिक्षित मॉडलों के लिए प्रीप्रोसेसिंग प्रदान करता है और इसे सीधे एक स्ट्रिंग (जैसे ऊपर दिए गए उदाहरण) या किसी सूची पर बुलाया जा सकता है। यह एक डिक्शनरी (तानाशाही) को आउटपुट करता है जिसे आप डाउनस्ट्रीम कोड में उपयोग कर सकते हैं या `**` अनपैकिंग एक्सप्रेशन के माध्यम से सीधे मॉडल को पास कर सकते हैं।
 
-मॉडल स्वयं एक नियमित [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) या [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (आपके बैकएंड के आधार पर), जो हो सकता है सामान्य तरीके से उपयोग किया जाता है। [यह ट्यूटोरियल](https://huggingface.co/transformers/training.html) बताता है कि इस तरह के मॉडल को क्लासिक PyTorch या TensorFlow प्रशिक्षण लूप में कैसे एकीकृत किया जाए, या हमारे `ट्रेनर` एपीआई का उपयोग कैसे करें ताकि इसे जल्दी से फ़ाइन ट्यून किया जा सके।एक नया डेटासेट पे।
+मॉडल स्वयं एक नियमित [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) या [TensorFlow `tf.keras.Model`](https ://pytorch.org/docs/stable/nn.html#torch.nn.Module) ://www.tensorflow.org/api_docs/python/tf/keras/Model) (आपके बैकएंड के आधार पर), जो हो सकता है सामान्य तरीके से उपयोग किया जाता है। [यह ट्यूटोरियल](https://huggingface.co/transformers/training.html) बताता है कि इस तरह के मॉडल को क्लासिक PyTorch या TensorFlow प्रशिक्षण लूप में कैसे एकीकृत किया जाए, या हमारे `ट्रेनर` एपीआई का उपयोग कैसे करें ताकि इसे जल्दी से फ़ाइन ट्यून किया जा सके।एक नया डेटासेट पे।
 
 ## ट्रांसफार्मर का उपयोग क्यों करें?
 
@@ -195,7 +194,7 @@ checkpoint: जाँच बिंदु
 
 - यह लाइब्रेरी मॉड्यूलर न्यूरल नेटवर्क टूलबॉक्स नहीं है। मॉडल फ़ाइल में कोड जानबूझकर अल्पविकसित है, बिना अतिरिक्त सार इनकैप्सुलेशन के, ताकि शोधकर्ता अमूर्तता और फ़ाइल जंपिंग में शामिल हुए जल्दी से पुनरावृति कर सकें।
 - `ट्रेनर` एपीआई किसी भी मॉडल के साथ संगत नहीं है, यह केवल इस पुस्तकालय के मॉडल के लिए अनुकूलित है। यदि आप सामान्य मशीन लर्निंग के लिए उपयुक्त प्रशिक्षण लूप कार्यान्वयन की तलाश में हैं, तो कहीं और देखें।
-- हमारे सर्वोत्तम प्रयासों के बावजूद, [उदाहरण निर्देशिका](https://github.com/huggingface/transformers/tree/main/examples) में स्क्रिप्ट केवल उपयोग के मामले हैं। आपकी विशिष्ट समस्या के लिए, वे जरूरी नहीं कि बॉक्स से बाहर काम करें, और आपको कोड की कुछ पंक्तियों को सूट करने की आवश्यकता हो सकती है।
+- हमारे सर्वोत्तम प्रयासों के बावजूद, [उदाहरण निर्देशिका] (https://github.com/huggingface/transformers/tree/main/examples) में स्क्रिप्ट केवल उपयोग के मामले हैं। आपकी विशिष्ट समस्या के लिए, वे जरूरी नहीं कि बॉक्स से बाहर काम करें, और आपको कोड की कुछ पंक्तियों को सूट करने की आवश्यकता हो सकती है।
 
 ## स्थापित करना
 
@@ -203,13 +202,11 @@ checkpoint: जाँच बिंदु
 
 इस रिपॉजिटरी का परीक्षण Python 3.8+, Flax 0.4.1+, PyTorch 1.10+ और TensorFlow 2.6+ के तहत किया गया है।
 
-आप [वर्चुअल एनवायरनमेंट](https://docs.python.org/3/library/venv.html) में 🤗 ट्रांसफॉर्मर इंस्टॉल कर सकते हैं। यदि आप अभी तक पायथन के वर्चुअल एनवायरनमेंट से परिचित नहीं हैं, तो कृपया इसे [उपयोगकर्ता निर्देश](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) पढ़ें।
+आप [वर्चुअल एनवायरनमेंट] (https://docs.python.org/3/library/venv.html) में 🤗 ट्रांसफॉर्मर इंस्टॉल कर सकते हैं। यदि आप अभी तक पायथन के वर्चुअल एनवायरनमेंट से परिचित नहीं हैं, तो कृपया इसे [उपयोगकर्ता निर्देश] (https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/) पढ़ें।
 
 सबसे पहले, पायथन के उस संस्करण के साथ एक आभासी वातावरण बनाएं जिसका आप उपयोग करने और उसे सक्रिय करने की योजना बना रहे हैं।
 
-फिर, आपको Flax, PyTorch या TensorFlow में से किसी एक को स्थापित करने की आवश्यकता है। अपने प्लेटफ़ॉर्म पर इन फ़्रेमवर्क को स्थापित करने के लिए, [TensorFlow स्थापना पृष्ठ](https://www.tensorflow.org/install/), [PyTorch स्थापना पृष्ठ](https://pytorch.org/get-started/locally) 
-
-देखें start-locally या [Flax स्थापना पृष्ठ](https://github.com/google/flax#quick-install).
+फिर, आपको Flax, PyTorch या TensorFlow में से किसी एक को स्थापित करने की आवश्यकता है। अपने प्लेटफ़ॉर्म पर इन फ़्रेमवर्क को स्थापित करने के लिए, [TensorFlow स्थापना पृष्ठ](https://www.tensorflow.org/install/), [PyTorch स्थापना पृष्ठ](https://pytorch.org/get-started /locally/# देखें) start-locally) या [Flax स्थापना पृष्ठ](https://github.com/google/flax#quick-install).
 
 जब इनमें से कोई एक बैकएंड सफलतापूर्वक स्थापित हो जाता है, तो ट्रांसफॉर्मर निम्नानुसार स्थापित किए जा सकते हैं:
 
@@ -217,7 +214,7 @@ checkpoint: जाँच बिंदु
 pip install transformers
 ```
 
-यदि आप उपयोग के मामलों को आज़माना चाहते हैं या आधिकारिक रिलीज़ से पहले नवीनतम इन-डेवलपमेंट कोड का उपयोग करना चाहते हैं, तो आपको [सोर्स से इंस्टॉल करना होगा](https://huggingface.co/docs/transformers/installation#installing-from-) स्रोत।
+यदि आप उपयोग के मामलों को आज़माना चाहते हैं या आधिकारिक रिलीज़ से पहले नवीनतम इन-डेवलपमेंट कोड का उपयोग करना चाहते हैं, तो आपको [सोर्स से इंस्टॉल करना होगा](https://huggingface.co/docs/transformers/installation#installing-from- स्रोत)।
 
 ### कोंडा का उपयोग करना
 
@@ -232,7 +229,7 @@ conda install -c huggingface transformers
 कोंडा के माध्यम से Flax, PyTorch, या TensorFlow में से किसी एक को स्थापित करने के लिए, निर्देशों के लिए उनके संबंधित स्थापना पृष्ठ देखें।
 
 ## मॉडल आर्किटेक्चर
-[उपयोगकर्ता](https://huggingface.co/users) और [organization](https://huggingface.co) द्वारा ट्रांसफॉर्मर समर्थित [**सभी मॉडल चौकियों**](https://huggingface.co/models/users) हगिंगफेस.को/ऑर्गनाइजेशन), सभी को बिना किसी बाधा के हगिंगफेस.को [मॉडल हब](https://huggingface.co) के साथ एकीकृत किया गया है।
+[उपयोगकर्ता](https://huggingface.co/users) और [organization](https://huggingface.co) द्वारा ट्रांसफॉर्मर समर्थित [**सभी मॉडल चौकियों**](https://huggingface.co/models) /users) हगिंगफेस.को/ऑर्गनाइजेशन), सभी को बिना किसी बाधा के हगिंगफेस.को [मॉडल हब](https://huggingface.co) के साथ एकीकृत किया गया है।
 
 चौकियों की वर्तमान संख्या: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
 
@@ -244,13 +241,13 @@ conda install -c huggingface transformers
 1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
 1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
 1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (फेसबुक) साथ थीसिस [बार्ट: प्राकृतिक भाषा निर्माण, अनुवाद के लिए अनुक्रम-से-अनुक्रम पूर्व प्रशिक्षण , और समझ](https://arxiv.org/pdf/1910.13461.pdf) पर निर्भर माइक लुईस, यिनहान लियू, नमन गोयल, मार्जन ग़ज़विनिनेजाद, अब्देलरहमान मोहम्मद, ओमर लेवी, वेस स्टोयानोव और ल्यूक ज़ेटलमॉयर
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (फेसबुक) साथ थीसिस [बार्ट: प्राकृतिक भाषा निर्माण, अनुवाद के लिए अनुक्रम-से-अनुक्रम पूर्व प्रशिक्षण , और समझ] (https://arxiv.org/pdf/1910.13461.pdf) पर निर्भर माइक लुईस, यिनहान लियू, नमन गोयल, मार्जन ग़ज़विनिनेजाद, अब्देलरहमान मोहम्मद, ओमर लेवी, वेस स्टोयानोव और ल्यूक ज़ेटलमॉयर
 1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (से École polytechnique) साथ थीसिस [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) पर निर्भर Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis रिहाई।
 1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (VinAI Research से) साथ में पेपर [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701)गुयेन लुओंग ट्रान, डुओंग मिन्ह ले और डाट क्वोक गुयेन द्वारा पोस्ट किया गया।
 1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (Microsoft से) साथ में कागज [BEiT: BERT इमेज ट्रांसफॉर्मर्स का प्री-ट्रेनिंग](https://arxiv.org/abs/2106.08254) Hangbo Bao, Li Dong, Furu Wei द्वारा।
 1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (गूगल से) साथ वाला पेपर [बीईआरटी: प्री-ट्रेनिंग ऑफ डीप बिडायरेक्शनल ट्रांसफॉर्मर्स फॉर लैंग्वेज अंडरस्टैंडिंग](https://arxiv.org/abs/1810.04805) जैकब डेवलिन, मिंग-वेई चांग, ​​केंटन ली और क्रिस्टीना टौटानोवा द्वारा प्रकाशित किया गया था। .
 1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (गूगल से) साथ देने वाला पेपर [सीक्वेंस जेनरेशन टास्क के लिए प्री-ट्रेंड चेकपॉइंट का इस्तेमाल करना](https ://arxiv.org/abs/1907.12461) साशा रोठे, शशि नारायण, अलियाक्सि सेवेरिन द्वारा।
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (VinAI Research से) साथ में पेपर [BERTweet: अंग्रेजी ट्वीट्स के लिए एक पूर्व-प्रशिक्षित भाषा मॉडल](https://aclanthology.org/2020.emnlp-demos.2/) डाट क्वोक गुयेन, थान वु और अन्ह तुआन गुयेन द्वारा प्रकाशित।
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (VinAI Research से) साथ में पेपर [BERTweet: अंग्रेजी ट्वीट्स के लिए एक पूर्व-प्रशिक्षित भाषा मॉडल] (https://aclanthology.org/2020.emnlp-demos.2/) डाट क्वोक गुयेन, थान वु और अन्ह तुआन गुयेन द्वारा प्रकाशित।
 1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (गूगल रिसर्च से) साथ वाला पेपर [बिग बर्ड: ट्रांसफॉर्मर्स फॉर लॉन्गर सीक्वेंस](https://arxiv .org/abs/2007.14062) मंज़िल ज़हीर, गुरु गुरुगणेश, अविनावा दुबे, जोशुआ आइंस्ली, क्रिस अल्बर्टी, सैंटियागो ओंटानोन, फिलिप फाम, अनिरुद्ध रावुला, किफ़ान वांग, ली यांग, अमर अहमद द्वारा।
 1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (गूगल रिसर्च से) साथ में पेपर [बिग बर्ड: ट्रांसफॉर्मर्स फॉर लॉन्गर सीक्वेंस](https://arxiv.org/abs/2007.14062) मंज़िल ज़हीर, गुरु गुरुगणेश, अविनावा दुबे, जोशुआ आइंस्ली, क्रिस अल्बर्टी, सैंटियागो ओंटानन, फिलिप फाम द्वारा , अनिरुद्ध रावुला, किफ़ान वांग, ली यांग, अमर अहमद द्वारा पोस्ट किया गया।
 1. **[BioGpt](https://huggingface.co/docs/transformers/model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
@@ -313,7 +310,6 @@ conda install -c huggingface transformers
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (गूगल रिसर्च से) साथ वाला पेपर [FNet: मिक्सिंग टोकन विद फूरियर ट्रांसफॉर्म्स](https://arxiv.org /abs/2105.03824) जेम्स ली-थॉर्प, जोशुआ आइंस्ली, इल्या एकस्टीन, सैंटियागो ओंटानन द्वारा।
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (Microsoft Research से) Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao. द्वाराअनुसंधान पत्र [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) के साथ जारी किया गया
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (सीएमयू/गूगल ब्रेन से) साथ में कागज [फ़नल-ट्रांसफॉर्मर: कुशल भाषा प्रसंस्करण के लिए अनुक्रमिक अतिरेक को छानना](https://arxiv.org/abs/2006.03236) जिहांग दाई, गुओकुन लाई, यिमिंग यांग, क्वोक वी. ले ​​द्वारा रिहाई।
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (ADEPT से) रोहन बाविशी, एरिच एलसेन, कर्टिस हॉथोर्न, मैक्सवेल नी, ऑगस्टस ओडेना, अरुशी सोमानी, सागनाक तासिरलार  [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (KAIST से) साथ वाला पेपर [वर्टिकल कटडेप्थ के साथ मोनोकुलर डेप्थ एस्टीमेशन के लिए ग्लोबल-लोकल पाथ नेटवर्क्स](https:/ /arxiv.org/abs/2201.07436) डोयोन किम, वूंगह्युन गा, प्युंगवान आह, डोंगग्यू जू, सेहवान चुन, जुनमो किम द्वारा।
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (OpenAI से) साथ में दिया गया पेपर [जेनरेटिव प्री-ट्रेनिंग द्वारा भाषा की समझ में सुधार](https://blog .openai.com/language-unsupervised/) एलेक रैडफोर्ड, कार्तिक नरसिम्हन, टिम सालिमन्स और इल्या सुत्स्केवर द्वारा।
@@ -335,7 +331,6 @@ conda install -c huggingface transformers
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (Salesforce से) Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi. द्वाराअनुसंधान पत्र [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) के साथ जारी किया गया
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (माइक्रोसॉफ्ट रिसर्च एशिया से) साथ देने वाला पेपर [लेआउटएलएमवी3: यूनिफाइड टेक्स्ट और इमेज मास्किंग के साथ दस्तावेज़ एआई के लिए पूर्व-प्रशिक्षण](https://arxiv.org/abs/2204.08387) युपन हुआंग, टेंगचाओ लव, लेई कुई, युटोंग लू, फुरु वेई द्वारा पोस्ट किया गया।
@@ -383,10 +378,9 @@ conda install -c huggingface transformers
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (Meta AI से) Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. द्वाराअनुसंधान पत्र [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) के साथ जारी किया गया
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (विस्कॉन्सिन विश्वविद्यालय - मैडिसन से) साथ में कागज [Nyströmformer: A Nyström- आधारित एल्गोरिथम आत्म-ध्यान का अनुमान लगाने के लिए ](https://arxiv.org/abs/2102.03902) युनयांग ज़िओंग, झानपेंग ज़ेंग, रुद्रसिस चक्रवर्ती, मिंगक्सिंग टैन, ग्लेन फंग, यिन ली, विकास सिंह द्वारा पोस्ट किया गया।
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs से) पेपर [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) जितेश जैन, जिआचेन ली, मांगटिक चिउ, अली हसनी, निकिता ओरलोव, हम्फ्री शि के द्वारा जारी किया गया है।
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (Google AI से) साथ में कागज [विज़न ट्रांसफॉर्मर्स के साथ सिंपल ओपन-वोकैबुलरी ऑब्जेक्ट डिटेक्शन](https:/ /arxiv.org/abs/2205.06230) मैथियास मिंडरर, एलेक्सी ग्रिट्सेंको, ऑस्टिन स्टोन, मैक्सिम न्यूमैन, डिर्क वीसेनबोर्न, एलेक्सी डोसोवित्स्की, अरविंद महेंद्रन, अनुराग अर्नब, मुस्तफा देहघानी, ज़ुओरन शेन, जिओ वांग, ज़ियाओहुआ झाई, थॉमस किफ़, और नील हॉल्सबी द्वारा पोस्ट किया गया।
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (Google AI से) Matthias Minderer, Alexey Gritsenko, Neil Houlsby. द्वाराअनुसंधान पत्र [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) के साथ जारी किया गया
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (Google की ओर से) साथ में दिया गया पेपर [लंबे इनपुट सारांश के लिए ट्रांसफ़ॉर्मरों को बेहतर तरीके से एक्सटेंड करना](https://arxiv .org/abs/2208.04347) जेसन फांग, याओ झाओ, पीटर जे लियू द्वारा।
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (दीपमाइंड से) साथ में पेपर [पर्सीवर आईओ: संरचित इनपुट और आउटपुट के लिए एक सामान्य वास्तुकला] (https://arxiv.org/abs/2107.14795) एंड्रयू जेगल, सेबेस्टियन बोरग्यूड, जीन-बैप्टिस्ट अलायराक, कार्ल डोर्श, कैटलिन इओनेस्कु, डेविड द्वारा डिंग, स्कंद कोप्पुला, डैनियल ज़ोरान, एंड्रयू ब्रॉक, इवान शेलहैमर, ओलिवियर हेनाफ, मैथ्यू एम। बोट्विनिक, एंड्रयू ज़िसरमैन, ओरिओल विनियल्स, जोआओ कैरेरा द्वारा पोस्ट किया गया।
@@ -410,7 +404,6 @@ conda install -c huggingface transformers
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (झुईई टेक्नोलॉजी से), साथ में पेपर [रोफॉर्मर: रोटरी पोजिशन एंबेडिंग के साथ एन्हांस्ड ट्रांसफॉर्मर] (https://arxiv.org/pdf/2104.09864v1.pdf) जियानलिन सु और यू लू और शेंगफेंग पैन और बो वेन और युनफेंग लियू द्वारा प्रकाशित।
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (Bo Peng से) Bo Peng. द्वाराअनुसंधान पत्र [this repo](https://github.com/BlinkDL/RWKV-LM) के साथ जारी किया गया
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (Meta AI से) Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick. द्वाराअनुसंधान पत्र [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) के साथ जारी किया गया
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP से) साथ देने वाला पेपर [भाषण पहचान के लिए अनसुपरवाइज्ड प्री-ट्रेनिंग में परफॉर्मेंस-एफिशिएंसी ट्रेड-ऑफ्स](https ://arxiv.org/abs/2109.06870) फेलिक्स वू, क्वांगयुन किम, जिंग पैन, क्यू हान, किलियन क्यू. वेनबर्गर, योव आर्टज़ी द्वारा।
@@ -453,12 +446,12 @@ conda install -c huggingface transformers
 1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (मेटा एआई से) साथ में कागज [लेबल-कुशल सीखने के लिए मास्क्ड स्याम देश के नेटवर्क](https://arxiv. org/abs/2204.07141) महमूद असरान, मथिल्डे कैरन, ईशान मिश्रा, पियोट्र बोजानोवस्की, फ्लोरियन बोर्डेस, पास्कल विंसेंट, आर्मंड जौलिन, माइकल रब्बत, निकोलस बल्लास द्वारा।
 1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (Kakao Enterprise से) Jaehyeon Kim, Jungil Kong, Juhee Son. द्वाराअनुसंधान पत्र [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) के साथ जारी किया गया
 1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (फेसबुक एआई से) साथ में पेपर [wav2vec 2.0: ए फ्रेमवर्क फॉर सेल्फ-सुपरवाइज्ड लर्निंग ऑफ स्पीच रिप्रेजेंटेशन](https://arxiv.org/abs/2006.11477) एलेक्सी बेवस्की, हेनरी झोउ, अब्देलरहमान मोहम्मद, माइकल औली द्वारा।
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (फेसबुक एआई से) साथ में पेपर [wav2vec 2.0: ए फ्रेमवर्क फॉर सेल्फ-सुपरवाइज्ड लर्निंग ऑफ स्पीच रिप्रेजेंटेशन] (https://arxiv.org/abs/2006.11477) एलेक्सी बेवस्की, हेनरी झोउ, अब्देलरहमान मोहम्मद, माइकल औली द्वारा।
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (Facebook AI से) साथ वाला पेपर [FAIRSEQ S2T: FAIRSEQ के साथ फास्ट स्पीच-टू-टेक्स्ट मॉडलिंग ](https://arxiv.org/abs/2010.05171) चांगहान वांग, यूं तांग, जुताई मा, ऐनी वू, सरव्या पोपुरी, दिमित्रो ओखोनको, जुआन पिनो द्वारा पोस्ट किया गया।
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (Facebook AI से) साथ वाला पेपर [सरल और प्रभावी जीरो-शॉट क्रॉस-लिंगुअल फोनेम रिकॉग्निशन](https://arxiv.org/abs/2109.11680) कियानटोंग जू, एलेक्सी बाएव्स्की, माइकल औली द्वारा।
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (माइक्रोसॉफ्ट रिसर्च से) पेपर के साथ जारी किया गया [WavLM: फुल स्टैक के लिए बड़े पैमाने पर स्व-पर्यवेक्षित पूर्व-प्रशिक्षण स्पीच प्रोसेसिंग](https://arxiv.org/abs/2110.13900) सानयुआन चेन, चेंगयी वांग, झेंगयांग चेन, यू वू, शुजी लियू, ज़ुओ चेन, जिन्यु ली, नाओयुकी कांडा, ताकुया योशियोका, ज़िओंग जिओ, जियान वू, लॉन्ग झोउ, शुओ रेन, यानमिन कियान, याओ कियान, जियान वू, माइकल ज़ेंग, फुरु वेई।
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (Facebook AI से) साथ वाला पेपर [सरल और प्रभावी जीरो-शॉट क्रॉस-लिंगुअल फोनेम रिकॉग्निशन](https:/ /arxiv.org/abs/2109.11680) कियानटोंग जू, एलेक्सी बाएव्स्की, माइकल औली द्वारा।
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (माइक्रोसॉफ्ट रिसर्च से) पेपर के साथ जारी किया गया [WavLM: फुल स्टैक के लिए बड़े पैमाने पर स्व-पर्यवेक्षित पूर्व-प्रशिक्षण स्पीच प्रोसेसिंग] (https://arxiv.org/abs/2110.13900) सानयुआन चेन, चेंगयी वांग, झेंगयांग चेन, यू वू, शुजी लियू, ज़ुओ चेन, जिन्यु ली, नाओयुकी कांडा, ताकुया योशियोका, ज़िओंग जिओ, जियान वू, लॉन्ग झोउ, शुओ रेन, यानमिन कियान, याओ कियान, जियान वू, माइकल ज़ेंग, फुरु वेई।
 1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (OpenAI से) साथ में कागज [बड़े पैमाने पर कमजोर पर्यवेक्षण के माध्यम से मजबूत भाषण पहचान](https://cdn. openai.com/papers/whisper.pdf) एलेक रैडफोर्ड, जोंग वूक किम, ताओ जू, ग्रेग ब्रॉकमैन, क्रिस्टीन मैकलीवे, इल्या सुत्स्केवर द्वारा।
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (माइक्रोसॉफ्ट रिसर्च से) कागज के साथ [एक्सपैंडिंग लैंग्वेज-इमेज प्रीट्रेन्ड मॉडल फॉर जनरल वीडियो रिकग्निशन](https://arxiv.org/abs/2208.02816) बोलिन नी, होउवेन पेंग, मिंगाओ चेन, सोंगयांग झांग, गाओफेंग मेंग, जियानलोंग फू, शिमिंग जियांग, हैबिन लिंग द्वारा।
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (माइक्रोसॉफ्ट रिसर्च से) कागज के साथ [एक्सपैंडिंग लैंग्वेज-इमेज प्रीट्रेन्ड मॉडल फॉर जनरल वीडियो रिकग्निशन](https: //arxiv.org/abs/2208.02816) बोलिन नी, होउवेन पेंग, मिंगाओ चेन, सोंगयांग झांग, गाओफेंग मेंग, जियानलोंग फू, शिमिंग जियांग, हैबिन लिंग द्वारा।
 1. **[X-MOD](https://huggingface.co/docs/transformers/model_doc/xmod)** (Meta AI से) Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe. द्वाराअनुसंधान पत्र [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) के साथ जारी किया गया
 1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
 1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (फेसबुक से) साथ में पेपर [क्रॉस-लिंगुअल लैंग्वेज मॉडल प्रीट्रेनिंग] (https://arxiv.org/abs/1901.07291) गिलाउम लैम्पल और एलेक्सिस कोनो द्वारा।
@@ -471,7 +464,7 @@ conda install -c huggingface transformers
 1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (फेसबुक एआई से) साथ में पेपर [अनसुपरवाइज्ड क्रॉस-लिंगुअल रिप्रेजेंटेशन लर्निंग फॉर स्पीच रिकग्निशन] (https://arxiv.org/abs/2006.13979) एलेक्सिस कोन्यू, एलेक्सी बेवस्की, रोनन कोलोबर्ट, अब्देलरहमान मोहम्मद, माइकल औली द्वारा।
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (हुआझोंग यूनिवर्सिटी ऑफ साइंस एंड टेक्नोलॉजी से) साथ में पेपर [यू ओनली लुक एट वन सीक्वेंस: रीथिंकिंग ट्रांसफॉर्मर इन विज़न थ्रू ऑब्जेक्ट डिटेक्शन](https://arxiv.org/abs/2106.00666) युक्सिन फेंग, बेनचेंग लियाओ, जिंगगैंग वांग, जेमिन फेंग, जियांग क्यूई, रुई वू, जियानवेई नीयू, वेन्यू लियू द्वारा पोस्ट किया गया।
 1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (विस्कॉन्सिन विश्वविद्यालय - मैडिसन से) साथ में पेपर [यू ओनली सैंपल (लगभग) ज़ानपेंग ज़ेंग, युनयांग ज़िओंग द्वारा , सत्य एन. रवि, शैलेश आचार्य, ग्लेन फंग, विकास सिंह द्वारा पोस्ट किया गया।
-1. एक नए मॉडल में योगदान देना चाहते हैं? नए मॉडल जोड़ने में आपका मार्गदर्शन करने के लिए हमारे पास एक **विस्तृत मार्गदर्शिका और टेम्प्लेट** है। आप उन्हें [`टेम्पलेट्स`](./templates) निर्देशिका में पा सकते हैं। पीआर शुरू करने से पहले [योगदान दिशानिर्देश](./CONTRIBUTING.md) देखना और अनुरक्षकों से संपर्क करना या प्रतिक्रिया प्राप्त करने के लिए एक नया मुद्दा खोलना याद रखें।
+1. एक नए मॉडल में योगदान देना चाहते हैं? नए मॉडल जोड़ने में आपका मार्गदर्शन करने के लिए हमारे पास एक **विस्तृत मार्गदर्शिका और टेम्प्लेट** है। आप उन्हें [`टेम्पलेट्स`](./templates) निर्देशिका में पा सकते हैं। पीआर शुरू करने से पहले [योगदान दिशानिर्देश] (./CONTRIBUTING.md) देखना और अनुरक्षकों से संपर्क करना या प्रतिक्रिया प्राप्त करने के लिए एक नया मुद्दा खोलना याद रखें।
 
 यह जांचने के लिए कि क्या किसी मॉडल में पहले से ही Flax, PyTorch या TensorFlow का कार्यान्वयन है, या यदि उसके पास Tokenizers लाइब्रेरी में संबंधित टोकन है, तो [यह तालिका](https://huggingface.co/docs/transformers/index#supported) देखें। -फ्रेमवर्क)।
 

README_ja.md

@@ -82,7 +82,6 @@ user: ユーザ
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <b>日本語</b> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
     </p>
 </h4>
 
@@ -211,7 +210,7 @@ Hugging Faceチームによって作られた **[トランスフォーマーを
 >>> outputs = model(**inputs)
 ```
 
-そしてこちらはTensorFlowと同等のコードとなります:
+And here is the equivalent code for TensorFlow:
 ```python
 >>> from transformers import AutoTokenizer, TFAutoModel
 
@@ -373,7 +372,6 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (Google Research から) James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon から公開された研究論文: [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824)
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (Microsoft Research から) Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao. から公開された研究論文 [Focal Modulation Networks](https://arxiv.org/abs/2203.11926)
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (CMU/Google Brain から) Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le から公開された研究論文: [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236)
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (ADEPT から) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. から公開された研究論文 [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (Microsoft Research から) Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang. から公開された研究論文 [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100)
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (KAIST から) Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim から公開された研究論文: [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436)
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (OpenAI から) Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever から公開された研究論文: [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/)
@@ -395,7 +393,6 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (Salesforce から) Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi. から公開された研究論文 [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500)
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (OpenAI から) Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever から公開された研究論文: [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf)
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (Microsoft Research Asia から) Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou から公開された研究論文: [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318)
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (Microsoft Research Asia から) Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou から公開された研究論文: [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740)
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (Microsoft Research Asia から) Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei から公開された研究論文: [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387)
@@ -443,10 +440,9 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (Meta AI から) Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. から公開された研究論文 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418)
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (the University of Wisconsin - Madison から) Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh から公開された研究論文: [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902)
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs から) Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi から公開された研究論文: [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220)
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (Meta AI から) Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al から公開された研究論文: [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068)
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (Google AI から) Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby から公開された研究論文: [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230)
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (Google AI から) Matthias Minderer, Alexey Gritsenko, Neil Houlsby. から公開された研究論文 [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683)
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (Google から) Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu から公開された研究論文: [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777)
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (Google から) Jason Phang, Yao Zhao, and Peter J. Liu から公開された研究論文: [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347)
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (Deepmind から) Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira から公開された研究論文: [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795)
@@ -470,7 +466,6 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (WeChatAI から) HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou から公開された研究論文: [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf)
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (ZhuiyiTechnology から), Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu から公開された研究論文: [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864)
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (Bo Peng から) Bo Peng. から公開された研究論文 [this repo](https://github.com/BlinkDL/RWKV-LM)
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (NVIDIA から) Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo から公開された研究論文: [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203)
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (Meta AI から) Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick. から公開された研究論文 [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf)
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP から) Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi から公開された研究論文: [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870)

README_ko.md

@@ -47,7 +47,6 @@ limitations under the License.
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
     </p>
 </h4>
 
@@ -288,7 +287,6 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. 논문과 함께 공개 [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -310,7 +308,6 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (Salesforce 에서 제공)은 Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.의 [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500)논문과 함께 발표했습니다.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (OpenAI 에서) Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever 의 [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) 논문과 함께 발표했습니다.
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (Microsoft Research Asia 에서) Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou 의 [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) 논문과 함께 발표했습니다.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (Microsoft Research Asia 에서) Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou 의 [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) 논문과 함께 발표했습니다.
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (Microsoft Research Asia 에서) Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei 의 [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) 논문과 함께 발표했습니다.
@@ -358,10 +355,9 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (Meta AI 에서 제공)은 Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.의 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418)논문과 함께 발표했습니다.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (the University of Wisconsin - Madison 에서) Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 의 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 논문과 함께 발표했습니다.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs 에서) Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi 의 [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) 논문과 함께 발표했습니다.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (Meta AI 에서) Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 의 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 논문과 함께 발표했습니다.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (Google AI 에서) Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 의 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 논문과 함께 발표했습니다.
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (Google AI 에서 제공)은 Matthias Minderer, Alexey Gritsenko, Neil Houlsby.의 [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683)논문과 함께 발표했습니다.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (Google 에서) Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 의 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 논문과 함께 발표했습니다.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (Google 에서) Jason Phang, Yao Zhao, Peter J. Liu 의 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 논문과 함께 발표했습니다.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (Deepmind 에서) Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira 의 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 논문과 함께 발표했습니다.
@@ -385,7 +381,6 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (WeChatAI 에서) HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou 의 [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) 논문과 함께 발표했습니다.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (ZhuiyiTechnology 에서) Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu 의 a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) 논문과 함께 발표했습니다.
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (Bo Peng 에서 제공)은 Bo Peng.의 [this repo](https://github.com/BlinkDL/RWKV-LM)논문과 함께 발표했습니다.
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (NVIDIA 에서) Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 의 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 논문과 함께 발표했습니다.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (Meta AI 에서 제공)은 Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.의 [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf)논문과 함께 발표했습니다.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP 에서) Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 의 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 논문과 함께 발표했습니다.

README_pt-br.md

@@ -1,565 +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.
--->
-
-<p align="center">
-  <picture>
-    <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-dark.svg">
-    <source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg">
-    <img alt="Hugging Face Transformers Library" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg" width="352" height="59" style="max-width: 100%;">
-  </picture>
-  <br/>
-  <br/>
-</p>
-
-<p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
-        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
-    </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
-    </a>
-    <a href="https://github.com/huggingface/transformers/releases">
-        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
-        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
-    </a>
-    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
-</p>
-
-<h4 align="center">
-    <p>
-        <b>English</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_pt-br.md">Рortuguês</a> |
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
-    </p>
-</h4>
-
-<h3 align="center">
-    <p>Aprendizado de máquina de última geração para JAX, PyTorch e TensorFlow</p>
-</h3>
-
-<h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
-</h3>
-
-
-A biblioteca 🤗 Transformers oferece milhares de modelos pré-treinados para executar tarefas em diferentes modalidades, como texto, visão e áudio.
-
-Esses modelos podem ser aplicados a:
-
-* 📝 Texto, para tarefas como classificação de texto, extração de informações, resposta a perguntas, sumarização, tradução, geração de texto, em mais de 100 idiomas.
-* 🖼️ Imagens, para tarefas como classificação de imagens, detecção de objetos e segmentação.
-* 🗣️ Áudio, para tarefas como reconhecimento de fala e classificação de áudio.
-
-Os modelos Transformer também podem executar tarefas em diversas modalidades combinadas, como responder a perguntas em tabelas, reconhecimento óptico de caracteres, extração de informações de documentos digitalizados, classificação de vídeo e resposta a perguntas visuais.
-
-
-A biblioteca 🤗 Transformers oferece APIs para baixar e usar rapidamente esses modelos pré-treinados em um texto específico, ajustá-los em seus próprios conjuntos de dados e, em seguida, compartilhá-los com a comunidade em nosso [model hub](https://huggingface.co/models). Ao mesmo tempo, cada módulo Python que define uma arquitetura é totalmente independente e pode ser modificado para permitir experimentos de pesquisa rápidos.
-
-A biblioteca 🤗 Transformers é respaldada pelas três bibliotecas de aprendizado profundo mais populares — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) e [TensorFlow](https://www.tensorflow.org/) — com uma integração perfeita entre elas. É simples treinar seus modelos com uma delas antes de carregá-los para inferência com a outra
-
-## Demonstração Online
-
-Você pode testar a maioria de nossos modelos diretamente em suas páginas a partir do [model hub](https://huggingface.co/models). Também oferecemos [hospedagem de modelos privados, versionamento e uma API de inferência](https://huggingface.co/pricing)
-para modelos públicos e privados.
-
-Aqui estão alguns exemplos:
-
-Em Processamento de Linguagem Natural:
-
-- [Completar palavra mascarada com BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
--  [Reconhecimento de Entidades Nomeadas com Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [Geração de texto com GPT-2](https://huggingface.co/gpt2?text=A+long+time+ago%2C)
-- [Inferência de Linguagem Natural com RoBERTa](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
-- [Sumarização com BART](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [Resposta a perguntas com DistilBERT](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [Tradução com T5](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-
-Em Visão Computacional:
-- [Classificação de Imagens com ViT](https://huggingface.co/google/vit-base-patch16-224)
-- [Detecção de Objetos com DETR](https://huggingface.co/facebook/detr-resnet-50)
-- [Segmentação Semântica com SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
-- [Segmentação Panóptica com MaskFormer](https://huggingface.co/facebook/maskformer-swin-small-coco)
-- [Estimativa de Profundidade com DPT](https://huggingface.co/docs/transformers/model_doc/dpt)
-- [Classificação de Vídeo com VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)
-- [Segmentação Universal com OneFormer](https://huggingface.co/shi-labs/oneformer_ade20k_dinat_large)
-
-
-Em Áudio:
-- [Reconhecimento Automático de Fala com Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base-960h)
-- [Detecção de Palavras-Chave com Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
-- [Classificação de Áudio com Transformer de Espectrograma de Áudio](https://huggingface.co/MIT/ast-finetuned-audioset-10-10-0.4593)
-
-Em Tarefas Multimodais:
-- [Respostas de Perguntas em Tabelas com TAPAS](https://huggingface.co/google/tapas-base-finetuned-wtq)
-- [Respostas de Perguntas Visuais com ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
-- [Classificação de Imagens sem Anotação com CLIP](https://huggingface.co/openai/clip-vit-large-patch14)
-- [Respostas de Perguntas em Documentos com LayoutLM](https://huggingface.co/impira/layoutlm-document-qa)
-- [Classificação de Vídeo sem Anotação com X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)
-
-## 100 Projetos Usando Transformers
-
-Transformers é mais do que um conjunto de ferramentas para usar modelos pré-treinados: é uma comunidade de projetos construídos ao seu redor e o Hugging Face Hub. Queremos que o Transformers permita que desenvolvedores, pesquisadores, estudantes, professores, engenheiros e qualquer outra pessoa construa seus projetos dos sonhos.
-
-Para celebrar as 100.000 estrelas do Transformers, decidimos destacar a comunidade e criamos a página [awesome-transformers](./awesome-transformers.md), que lista 100 projetos incríveis construídos nas proximidades dos Transformers.
-
-Se você possui ou utiliza um projeto que acredita que deveria fazer parte da lista, abra um PR para adicioná-lo!
-
-## Se você está procurando suporte personalizado da equipe Hugging Face
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-
-## Tour Rápido
-
-Para usar imediatamente um modelo em uma entrada específica (texto, imagem, áudio, ...), oferecemos a API `pipeline`. Os pipelines agrupam um modelo pré-treinado com o pré-processamento que foi usado durante o treinamento desse modelo. Aqui está como usar rapidamente um pipeline para classificar textos como positivos ou negativos:
-
-```python
-from transformers import pipeline
-
-# Carregue o pipeline de classificação de texto
->>> classifier = pipeline("sentiment-analysis")
-
-# Classifique o texto como positivo ou negativo
->>> classifier("Estamos muito felizes em apresentar o pipeline no repositório dos transformers.")
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
-```
-
-A segunda linha de código baixa e armazena em cache o modelo pré-treinado usado pelo pipeline, enquanto a terceira linha o avalia no texto fornecido. Neste exemplo, a resposta é "positiva" com uma confiança de 99,97%.
-
-Muitas tarefas têm um `pipeline` pré-treinado pronto para uso, não apenas em PNL, mas também em visão computacional e processamento de áudio. Por exemplo, podemos facilmente extrair objetos detectados em uma imagem:
-
-``` python
->>> import requests
->>> from PIL import Image
->>> from transformers import pipeline
-
-# Download an image with cute cats
->>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
->>> image_data = requests.get(url, stream=True).raw
->>> image = Image.open(image_data)
-
-# Allocate a pipeline for object detection
->>> object_detector = pipeline('object-detection')
->>> object_detector(image)
-[{'score': 0.9982201457023621,
-  'label': 'remote',
-  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
- {'score': 0.9960021376609802,
-  'label': 'remote',
-  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
- {'score': 0.9954745173454285,
-  'label': 'couch',
-  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
- {'score': 0.9988006353378296,
-  'label': 'cat',
-  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
- {'score': 0.9986783862113953,
-  'label': 'cat',
-  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
-```
-
-
-Aqui obtemos uma lista de objetos detectados na imagem, com uma caixa envolvendo o objeto e uma pontuação de confiança. Aqui está a imagem original à esquerda, com as previsões exibidas à direita:
-
-<h3 align="center">
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
-</h3>
-
-Você pode aprender mais sobre as tarefas suportadas pela API `pipeline` em [este tutorial](https://huggingface.co/docs/transformers/task_summary).
-
-
-Além do `pipeline`, para baixar e usar qualquer um dos modelos pré-treinados em sua tarefa específica, tudo o que é necessário são três linhas de código. Aqui está a versão em PyTorch:
-
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = AutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
-```
-
-E aqui está o código equivalente para TensorFlow:
-
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
-```
-
-O tokenizador é responsável por todo o pré-processamento que o modelo pré-treinado espera, e pode ser chamado diretamente em uma única string (como nos exemplos acima) ou em uma lista. Ele produzirá um dicionário que você pode usar no código subsequente ou simplesmente passar diretamente para o seu modelo usando o operador de descompactação de argumentos **.
-
-O modelo em si é um [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)  ou um [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)(dependendo do seu back-end) que você pode usar como de costume. [Este tutorial](https://huggingface.co/docs/transformers/training) explica como integrar esse modelo em um ciclo de treinamento clássico do PyTorch ou TensorFlow, ou como usar nossa API `Trainer` para ajuste fino rápido em um novo conjunto de dados.
-
-## Por que devo usar transformers?
-
-1. Modelos state-of-the-art fáceis de usar:
-    - Alto desempenho em compreensão e geração de linguagem natural, visão computacional e tarefas de áudio.
-    - Barreira de entrada baixa para educadores e profissionais.
-    - Poucas abstrações visíveis para o usuário, com apenas três classes para aprender.
-    - Uma API unificada para usar todos os nossos modelos pré-treinados.
-
-1. Menores custos de computação, menor pegada de carbono:
-    - Pesquisadores podem compartilhar modelos treinados em vez de treinar sempre do zero.
-    - Profissionais podem reduzir o tempo de computação e os custos de produção.
-    - Dezenas de arquiteturas com mais de 60.000 modelos pré-treinados em todas as modalidades.
-
-1. Escolha o framework certo para cada parte da vida de um modelo:
-    - Treine modelos state-of-the-art em 3 linhas de código.
-    - Mova um único modelo entre frameworks TF2.0/PyTorch/JAX à vontade.
-    - Escolha o framework certo de forma contínua para treinamento, avaliação e produção.
-
-1. Personalize facilmente um modelo ou um exemplo para atender às suas necessidades:
-    - Fornecemos exemplos para cada arquitetura para reproduzir os resultados publicados pelos autores originais.
-    - Os detalhes internos do modelo são expostos de maneira consistente.
-    - Os arquivos do modelo podem ser usados de forma independente da biblioteca para experimentos rápidos.
-
-## Por que não devo usar transformers?
-
-- Esta biblioteca não é uma caixa de ferramentas modular para construir redes neurais. O código nos arquivos do modelo não é refatorado com abstrações adicionais de propósito, para que os pesquisadores possam iterar rapidamente em cada um dos modelos sem se aprofundar em abstrações/arquivos adicionais.
-- A API de treinamento não é projetada para funcionar com qualquer modelo, mas é otimizada para funcionar com os modelos fornecidos pela biblioteca. Para loops de aprendizado de máquina genéricos, você deve usar outra biblioteca (possivelmente, [Accelerate](https://huggingface.co/docs/accelerate)).
-- Embora nos esforcemos para apresentar o maior número possível de casos de uso, os scripts em nossa [pasta de exemplos](https://github.com/huggingface/transformers/tree/main/examples) são apenas isso: exemplos. É esperado que eles não funcionem prontos para uso em seu problema específico e que seja necessário modificar algumas linhas de código para adaptá-los às suas necessidades.
-
-
-
-### Com pip
-
-Este repositório é testado no Python 3.8+, Flax 0.4.1+, PyTorch 1.10+ e TensorFlow 2.6+.
-
-Você deve instalar o 🤗 Transformers em um [ambiente virtual](https://docs.python.org/3/library/venv.html). Se você não está familiarizado com ambientes virtuais em Python, confira o [guia do usuário](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
-
-Primeiro, crie um ambiente virtual com a versão do Python que você vai usar e ative-o.
-
-Em seguida, você precisará instalar pelo menos um dos back-ends Flax, PyTorch ou TensorFlow.
-Consulte a [página de instalação do TensorFlow](https://www.tensorflow.org/install/), a [página de instalação do PyTorch](https://pytorch.org/get-started/locally/#start-locally) e/ou [Flax](https://github.com/google/flax#quick-install) e [Jax](https://github.com/google/jax#installation) páginas de instalação para obter o comando de instalação específico para a sua plataforma.
-
-Quando um desses back-ends estiver instalado, o 🤗 Transformers pode ser instalado usando pip da seguinte forma:
-
-```bash
-pip install transformers
-```
-Se você deseja experimentar com os exemplos ou precisa da versão mais recente do código e não pode esperar por um novo lançamento, você deve instalar a [biblioteca a partir do código-fonte](https://huggingface.co/docs/transformers/installation#installing-from-source).
-
-### Com conda
-
-Desde a versão v4.0.0 do Transformers, agora temos um canal conda: `huggingface`.
-
-O 🤗 Transformers pode ser instalado com conda da seguinte forma:
-
-```bash
-conda install -c huggingface transformers
-```
-
-Siga as páginas de instalação do Flax, PyTorch ou TensorFlow para ver como instalá-los com conda. 
-
-Siga as páginas de instalação do Flax, PyTorch ou TensorFlow para ver como instalá-los com o conda.
-
-> **_NOTA:_**  No Windows, você pode ser solicitado a ativar o Modo de Desenvolvedor para aproveitar o cache. Se isso não for uma opção para você, por favor nos avise [neste problema](https://github.com/huggingface/huggingface_hub/issues/1062).
-
-## Arquiteturas de Modelos
-
-**[Todos os pontos de verificação de modelo](https://huggingface.co/models)** fornecidos pelo 🤗 Transformers são integrados de forma transparente do [model hub](https://huggingface.co/models) do huggingface.co, onde são carregados diretamente por [usuários](https://huggingface.co/users) e [organizações](https://huggingface.co/organizations).
-
-Número atual de pontos de verificação: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 Transformers atualmente fornece as seguintes arquiteturas (veja [aqui](https://huggingface.co/docs/transformers/model_summary) para um resumo de alto nível de cada uma delas):
-
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (from Google Research) released with the paper [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.
-1. **[AltCLIP](https://huggingface.co/docs/transformers/model_doc/altclip)** (from BAAI) released with the paper [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679) by Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell.
-1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
-1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
-1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BioGpt](https://huggingface.co/docs/transformers/model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
-1. **[BiT](https://huggingface.co/docs/transformers/model_doc/bit)** (from Google AI) released with the paper [Big Transfer (BiT): General Visual Representation Learning](https://arxiv.org/abs/1912.11370) by Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil Houlsby.
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLIP](https://huggingface.co/docs/transformers/model_doc/blip)** (from Salesforce) released with the paper [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://arxiv.org/abs/2201.12086) by Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi.
-1. **[BLIP-2](https://huggingface.co/docs/transformers/model_doc/blip-2)** (from Salesforce) released with the paper [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://arxiv.org/abs/2301.12597) by Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi.
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
-1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[Chinese-CLIP](https://huggingface.co/docs/transformers/model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
-1. **[CLAP](https://huggingface.co/docs/transformers/model_doc/clap)** (from LAION-AI) released with the paper [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://arxiv.org/abs/2211.06687) by Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, Shlomo Dubnov.
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)** (from University of Göttingen) released with the paper [Image Segmentation Using Text and Image Prompts](https://arxiv.org/abs/2112.10003) by Timo Lüddecke and Alexander Ecker.
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[CodeLlama](https://huggingface.co/docs/transformers/model_doc/llama_code)** (from MetaAI) released with the paper [Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/) by Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong, Alexandre Défossez, Jade Copet, Faisal Azhar, Hugo Touvron, Louis Martin, Nicolas Usunier, Thomas Scialom, Gabriel Synnaeve.
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[ConvNeXTV2](https://huggingface.co/docs/transformers/model_doc/convnextv2)** (from Facebook AI) released with the paper [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) by Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, Saining Xie.
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CPM-Ant](https://huggingface.co/docs/transformers/model_doc/cpmant)** (from OpenBMB) released by the [OpenBMB](https://www.openbmb.org/).
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
-1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DiNAT](https://huggingface.co/docs/transformers/model_doc/dinat)** (from SHI Labs) released with the paper [Dilated Neighborhood Attention Transformer](https://arxiv.org/abs/2209.15001) by Ali Hassani and Humphrey Shi.
-1. **[DINOv2](https://huggingface.co/docs/transformers/model_doc/dinov2)** (from Meta AI) released with the paper [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[EfficientFormer](https://huggingface.co/docs/transformers/model_doc/efficientformer)** (from Snap Research) released with the paper [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191) by Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren.
-1. **[EfficientNet](https://huggingface.co/docs/transformers/model_doc/efficientnet)** (from Google Brain) released with the paper [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946) by Mingxing Tan, Quoc V. Le.
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EnCodec](https://huggingface.co/docs/transformers/model_doc/encodec)** (from Meta AI) released with the paper [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) by Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (from Baidu) released with the paper [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) by Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang.
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2 and ESMFold** were released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[Falcon](https://huggingface.co/docs/transformers/model_doc/falcon)** (from Technology Innovation Institute) by Almazrouei, Ebtesam and Alobeidli, Hamza and Alshamsi, Abdulaziz and Cappelli, Alessandro and Cojocaru, Ruxandra and Debbah, Merouane and Goffinet, Etienne and Heslow, Daniel and Launay, Julien and Malartic, Quentin and Noune, Badreddine and Pannier, Baptiste and Penedo, Guilherme.
-1. **[FLAN-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-t5-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
-1. **[FLAN-UL2](https://huggingface.co/docs/transformers/model_doc/flan-ul2)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-ul2-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://openai.com/research/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://openai.com/research/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GPT-Sw3](https://huggingface.co/docs/transformers/model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
-1. **[GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
-1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
-1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh.
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
-1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
-1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](https://huggingface.co/docs/transformers/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[LLaMA](https://huggingface.co/docs/transformers/model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
-1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[Mask2Former](https://huggingface.co/docs/transformers/model_doc/mask2former)** (from FAIR and UIUC) released with the paper [Masked-attention Mask Transformer for Universal Image Segmentation](https://arxiv.org/abs/2112.01527) by Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, Rohit Girdhar.
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[MatCha](https://huggingface.co/docs/transformers/model_doc/matcha)** (from Google AI) released with the paper [MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering](https://arxiv.org/abs/2212.09662) by Fangyu Liu, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Yasemin Altun, Nigel Collier, Julian Martin Eisenschlos.
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[MEGA](https://huggingface.co/docs/transformers/model_doc/mega)** (from Meta/USC/CMU/SJTU) released with the paper [Mega: Moving Average Equipped Gated Attention](https://arxiv.org/abs/2209.10655) by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer.
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
-1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The [Mistral AI](https://mistral.ai) team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
-1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MPT](https://huggingface.co/docs/transformers/model_doc/mpt)** (from MosaiML) released with the repository [llm-foundry](https://github.com/mosaicml/llm-foundry/) by the MosaicML NLP Team.
-1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (from ADEPT) released in a [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng), released on [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechT5](https://huggingface.co/docs/transformers/model_doc/speecht5)** (from Microsoft Research) released with the paper [SpeechT5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing](https://arxiv.org/abs/2110.07205) by Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, Zhihua Wei, Yao Qian, Jinyu Li, Furu Wei.
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (from MBZUAI) released with the paper [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) by Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (from University of Würzburg) released with the paper [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) by Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte.
-1. **[SwitchTransformers](https://huggingface.co/docs/transformers/model_doc/switch_transformers)** (from Google) released with the paper [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961) by William Fedus, Barret Zoph, Noam Shazeer.
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[Table Transformer](https://huggingface.co/docs/transformers/model_doc/table-transformer)** (from Microsoft Research) released with the paper [PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents](https://arxiv.org/abs/2110.00061) by Brandon Smock, Rohith Pesala, Robin Abraham.
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
-1. **[TimeSformer](https://huggingface.co/docs/transformers/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[TVLT](https://huggingface.co/docs/transformers/model_doc/tvlt)** (from UNC Chapel Hill) released with the paper [TVLT: Textless Vision-Language Transformer](https://arxiv.org/abs/2209.14156) by Zineng Tang, Jaemin Cho, Yixin Nie, Mohit Bansal.
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UMT5](https://huggingface.co/docs/transformers/model_doc/umt5)** (from Google Research) released with the paper [UniMax: Fairer and More Effective Language Sampling for Large-Scale Multilingual Pretraining](https://openreview.net/forum?id=kXwdL1cWOAi) by Hyung Won Chung, Xavier Garcia, Adam Roberts, Yi Tay, Orhan Firat, Sharan Narang, Noah Constant.
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[UPerNet](https://huggingface.co/docs/transformers/model_doc/upernet)** (from Peking University) released with the paper [Unified Perceptual Parsing for Scene Understanding](https://arxiv.org/abs/1807.10221) by Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, Jian Sun.
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) rreleased with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
-1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[X-MOD](https://huggingface.co/docs/transformers/model_doc/xmod)** (from Meta AI) released with the paper [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) by Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe.
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLM-V](https://huggingface.co/docs/transformers/model_doc/xlm-v)** (from Meta AI) released with the paper [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://arxiv.org/abs/2301.10472) by Davis Liang, Hila Gonen, Yuning Mao, Rui Hou, Naman Goyal, Marjan Ghazvininejad, Luke Zettlemoyer, Madian Khabsa.
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, 
-Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-
-1. Quer contribuir com um novo modelo? Adicionamos um **guia detalhado e modelos de exemplo** para orientar você no processo de adição de um novo modelo. Você pode encontrá-los na pasta [`templates`](./templates) do repositório. Certifique-se de verificar as [diretrizes de contribuição](./CONTRIBUTING.md) e entrar em contato com os mantenedores ou abrir uma issue para coletar feedback antes de iniciar sua PR.
-
-Para verificar se cada modelo tem uma implementação em Flax, PyTorch ou TensorFlow, ou possui um tokenizador associado com a biblioteca 🤗 Tokenizers, consulte [esta tabela](https://huggingface.co/docs/transformers/index#supported-frameworks).
-
-Essas implementações foram testadas em vários conjuntos de dados (veja os scripts de exemplo) e devem corresponder ao desempenho das implementações originais. Você pode encontrar mais detalhes sobre o desempenho na seção de Exemplos da [documentação](https://github.com/huggingface/transformers/tree/main/examples).
-
-
-## Saiba mais
-
-| Seção | Descrição |
-|-|-|
-| [Documentação](https://huggingface.co/docs/transformers/) | Documentação completa da API e tutoriais |
-| [Resumo de Tarefas](https://huggingface.co/docs/transformers/task_summary) | Tarefas suportadas pelo 🤗 Transformers |
-| [Tutorial de Pré-processamento](https://huggingface.co/docs/transformers/preprocessing) | Usando a classe `Tokenizer` para preparar dados para os modelos |
-| [Treinamento e Ajuste Fino](https://huggingface.co/docs/transformers/training) | Usando os modelos fornecidos pelo 🤗 Transformers em um loop de treinamento PyTorch/TensorFlow e a API `Trainer` |
-| [Tour Rápido: Scripts de Ajuste Fino/Utilização](https://github.com/huggingface/transformers/tree/main/examples) | Scripts de exemplo para ajuste fino de modelos em uma ampla gama de tarefas |
-| [Compartilhamento e Envio de Modelos](https://huggingface.co/docs/transformers/model_sharing) | Envie e compartilhe seus modelos ajustados com a comunidade |
-
-## Citação
-
-Agora temos um [artigo](https://www.aclweb.org/anthology/2020.emnlp-demos.6/) que você pode citar para a biblioteca 🤗 Transformers:
-```bibtex
-@inproceedings{wolf-etal-2020-transformers,
-    title = "Transformers: State-of-the-Art Natural Language Processing",
-    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
-    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
-    month = out,
-    year = "2020",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
-    pages = "38--45"
-}
-```

README_ru.md

@@ -53,7 +53,6 @@ limitations under the License.
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
         <b>Русский</b>
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
     <p>
 </h4>
 
@@ -67,11 +66,11 @@ limitations under the License.
 
 🤗 Transformers предоставляет тысячи предварительно обученных моделей для выполнения различных задач, таких как текст, зрение и аудио.
 
-Эти модели могут быть применены к:
+Эти модели могут быть применены на:
 
-* 📝 Тексту для таких задач, как классификация текстов, извлечение информации, ответы на вопросы, обобщение, перевод, генерация текстов на более чем 100 языках.
-* 🖼️ Изображениям для задач классификации изображений, обнаружения объектов и сегментации.
-* 🗣️ Аудио для задач распознавания речи и классификации аудио.
+* 📝 Текст, для таких задач, как классификация текстов, извлечение информации, ответы на вопросы, обобщение, перевод, генерация текстов, на более чем 100 языках.
+* 🖼️ Изображения - для задач классификации изображений, обнаружения объектов и сегментации.
+* 🗣️ Аудио - для задач распознавания речи и классификации аудио.
 
 Модели transformers также могут выполнять несколько задач, такие как ответы на табличные вопросы, распознавание оптических символов, извлечение информации из отсканированных документов, классификация видео и ответы на визуальные вопросы.
 
@@ -362,7 +361,6 @@ conda install -c huggingface transformers
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -429,7 +427,7 @@ conda install -c huggingface transformers
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.

README_te.md

@@ -1,557 +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.
--->
-
-<p align="center">
-  <picture>
-    <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-dark.svg">
-    <source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg">
-    <img alt="Hugging Face Transformers Library" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg" width="352" height="59" style="max-width: 100%;">
-  </picture>
-  <br/>
-  <br/>
-</p>
-
-
-<p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers">
-        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
-        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
-    </a>
-    <a href="https://huggingface.co/docs/transformers/index">
-        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
-    </a>
-    <a href="https://github.com/huggingface/transformers/releases">
-        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
-    </a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
-        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
-    </a>
-    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
-</p>
-
-
-<h4 align="center">
-    <p>
-        <a href="https://github.com/huggingface/transformers/">English</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_ru.md">Русский</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/README_pt-br.md">Рortuguês</a> |
-        <b>తెలుగు</b> |
-    </p>
-</h4>
-
-<h3 align="center">
-    <p>JAX, PyTorch మరియు TensorFlow కోసం అత్యాధునిక యంత్ర అభ్యాసం</p>
-</h3>
-
-<h3 align="center">
-    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
-</h3>
-
-🤗 ట్రాన్స్‌ఫార్మర్లు టెక్స్ట్, విజన్ మరియు ఆడియో వంటి విభిన్న పద్ధతులపై టాస్క్‌లను నిర్వహించడానికి వేలాది ముందుగా శిక్షణ పొందిన మోడల్‌లను అందిస్తాయి.
-
-ఈ నమూనాలు వర్తించవచ్చు:
-
-* 📝 టెక్స్ట్, 100కి పైగా భాషల్లో టెక్స్ట్ క్లాసిఫికేషన్, ఇన్ఫర్మేషన్ ఎక్స్‌ట్రాక్షన్, ప్రశ్నలకు సమాధానాలు, సారాంశం, అనువాదం, టెక్స్ట్ జనరేషన్ వంటి పనుల కోసం.
-* 🖼️ ఇమేజ్‌లు, ఇమేజ్ వర్గీకరణ, ఆబ్జెక్ట్ డిటెక్షన్ మరియు సెగ్మెంటేషన్ వంటి పనుల కోసం.
-* 🗣️ ఆడియో, స్పీచ్ రికగ్నిషన్ మరియు ఆడియో వర్గీకరణ వంటి పనుల కోసం.
-
-ట్రాన్స్‌ఫార్మర్ మోడల్‌లు టేబుల్ క్వశ్చన్ ఆన్సర్ చేయడం, ఆప్టికల్ క్యారెక్టర్ రికగ్నిషన్, స్కాన్ చేసిన డాక్యుమెంట్‌ల నుండి ఇన్ఫర్మేషన్ ఎక్స్‌ట్రాక్షన్, వీడియో క్లాసిఫికేషన్ మరియు విజువల్ క్వశ్చన్ ఆన్సర్ చేయడం వంటి **అనేక పద్ధతులతో కలిపి** పనులను కూడా చేయగలవు.
-
-🤗 ట్రాన్స్‌ఫార్మర్లు అందించిన టెక్స్ట్‌లో ప్రీట్రైన్డ్ మోడల్‌లను త్వరగా డౌన్‌లోడ్ చేయడానికి మరియు ఉపయోగించడానికి, వాటిని మీ స్వంత డేటాసెట్‌లలో ఫైన్-ట్యూన్ చేయడానికి మరియు వాటిని మా [మోడల్ హబ్](https://huggingface.co/models)లో సంఘంతో భాగస్వామ్యం చేయడానికి API లను అందిస్తుంది. అదే సమయంలో, ఆర్కిటెక్చర్‌ని నిర్వచించే ప్రతి పైథాన్ మాడ్యూల్ పూర్తిగా స్వతంత్రంగా ఉంటుంది మరియు త్వరిత పరిశోధన ప్రయోగాలను ప్రారంభించడానికి సవరించవచ్చు.
-
-🤗 ట్రాన్స్‌ఫార్మర్‌లకు మూడు అత్యంత ప్రజాదరణ పొందిన డీప్ లెర్నింగ్ లైబ్రరీలు ఉన్నాయి — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) మరియు [TensorFlow](https://www.tensorflow.org/) — వాటి మధ్య అతుకులు లేని ఏకీకరణతో. మీ మోడల్‌లను ఒకదానితో మరొకదానితో అనుమితి కోసం లోడ్ చేసే ముందు వాటికి శిక్షణ ఇవ్వడం చాలా సులభం.
-
-## ఆన్‌లైన్ డెమోలు
-
-మీరు [మోడల్ హబ్](https://huggingface.co/models) నుండి మా మోడళ్లలో చాలా వరకు వాటి పేజీలలో నేరుగా పరీక్షించవచ్చు. మేము పబ్లిక్ మరియు ప్రైవేట్ మోడల్‌ల కోసం [ప్రైవేట్ మోడల్ హోస్టింగ్, సంస్కరణ & అనుమితి API](https://huggingface.co/pricing)ని కూడా అందిస్తాము.
-
-ఇక్కడ కొన్ని ఉదాహరణలు ఉన్నాయి:
-
-సహజ భాషా ప్రాసెసింగ్‌లో:
-- [BERT తో మాస్క్‌డ్ వర్డ్ కంప్లీషన్](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [Electra తో పేరు ఎంటిటీ గుర్తింపు](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [GPT-2 తో టెక్స్ట్ జనరేషన్](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
-- [RoBERTa తో సహజ భాషా అనుమితి](https://huggingface.co/roberta-large-mnli?text=The+dog+was+Lost.+Nobody+lost+any+animal)
-- [BART తో సారాంశం](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [DistilBERT తో ప్రశ్న సమాధానం](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [T5 తో అనువాదం](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-కంప్యూటర్ దృష్టిలో:
-- [VIT తో చిత్ర వర్గీకరణ](https://huggingface.co/google/vit-base-patch16-224)
-- [DETR తో ఆబ్జెక్ట్ డిటెక్షన్](https://huggingface.co/facebook/detr-resnet-50)
-- [SegFormer తో సెమాంటిక్ సెగ్మెంటేషన్](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
-- [MaskFormer తో పానోప్టిక్ సెగ్మెంటేషన్](https://huggingface.co/facebook/maskformer-swin-small-coco)
-- [DPT తో లోతు అంచనా](https://huggingface.co/docs/transformers/model_doc/dpt)
-- [VideoMAE తో వీడియో వర్గీకరణ](https://huggingface.co/docs/transformers/model_doc/videomae)
-- [OneFormer తో యూనివర్సల్ సెగ్మెంటేషన్](https://huggingface.co/shi-labs/oneformer_ade20k_dinat_large)
-
-ఆడియోలో:
-- [Wav2Vec2 తో ఆటోమేటిక్ స్పీచ్ రికగ్నిషన్](https://huggingface.co/facebook/wav2vec2-base-960h)
-- [Wav2Vec2 తో కీవర్డ్ స్పాటింగ్](https://huggingface.co/superb/wav2vec2-base-superb-ks)
-- [ఆడియో స్పెక్ట్రోగ్రామ్ ట్రాన్స్‌ఫార్మర్‌తో ఆడియో వర్గీకరణ](https://huggingface.co/MIT/ast-finetuned-audioset-10-10-0.4593)
-
-మల్టీమోడల్ టాస్క్‌లలో:
-- [TAPAS తో టేబుల్ ప్రశ్న సమాధానాలు](https://huggingface.co/google/tapas-base-finetuned-wtq)
-- [ViLT తో దృశ్యమాన ప్రశ్నకు సమాధానం](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
-- [CLIP తో జీరో-షాట్ ఇమేజ్ వర్గీకరణ](https://huggingface.co/openai/clip-vit-large-patch14)
-- [LayoutLM తో డాక్యుమెంట్ ప్రశ్నకు సమాధానం](https://huggingface.co/impira/layoutlm-document-qa)
-- [X-CLIP తో జీరో-షాట్ వీడియో వర్గీకరణ](https://huggingface.co/docs/transformers/model_doc/xclip)
-
-## ట్రాన్స్‌ఫార్మర్‌లను ఉపయోగించి 100 ప్రాజెక్టులు
-
-ట్రాన్స్‌ఫార్మర్లు ప్రీట్రైన్డ్ మోడల్‌లను ఉపయోగించడానికి టూల్‌కిట్ కంటే ఎక్కువ: ఇది దాని చుట్టూ నిర్మించిన ప్రాజెక్ట్‌ల సంఘం మరియు
-హగ్గింగ్ ఫేస్ హబ్. డెవలపర్‌లు, పరిశోధకులు, విద్యార్థులు, ప్రొఫెసర్‌లు, ఇంజనీర్లు మరియు ఎవరినైనా అనుమతించేలా ట్రాన్స్‌ఫార్మర్‌లను మేము కోరుకుంటున్నాము
-వారి కలల ప్రాజెక్టులను నిర్మించడానికి.
-
-ట్రాన్స్‌ఫార్మర్‌ల 100,000 నక్షత్రాలను జరుపుకోవడానికి, మేము స్పాట్‌లైట్‌ని ఉంచాలని నిర్ణయించుకున్నాము
-సంఘం, మరియు మేము 100 జాబితాలను కలిగి ఉన్న [awesome-transformers](./awesome-transformers.md) పేజీని సృష్టించాము.
-ట్రాన్స్‌ఫార్మర్ల పరిసరాల్లో అద్భుతమైన ప్రాజెక్టులు నిర్మించబడ్డాయి.
-
-జాబితాలో భాగమని మీరు విశ్వసించే ప్రాజెక్ట్‌ను మీరు కలిగి ఉంటే లేదా ఉపయోగిస్తుంటే, దయచేసి దానిని జోడించడానికి PRని తెరవండి!
-
-## మీరు హగ్గింగ్ ఫేస్ టీమ్ నుండి అనుకూల మద్దతు కోసం చూస్తున్నట్లయితే
-
-<a target="_blank" href="https://huggingface.co/support">
-    <img alt="HuggingFace Expert Acceleration Program" src="https://cdn-media.huggingface.co/marketing/transformers/new-support-improved.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
-</a><br>
-
-## త్వరిత పర్యటన
-
-ఇచ్చిన ఇన్‌పుట్ (టెక్స్ట్, ఇమేజ్, ఆడియో, ...)పై తక్షణమే మోడల్‌ను ఉపయోగించడానికి, మేము `pipeline` API ని అందిస్తాము. పైప్‌లైన్‌లు ఆ మోడల్ శిక్షణ సమయంలో ఉపయోగించిన ప్రీప్రాసెసింగ్‌తో కూడిన ప్రీట్రైన్డ్ మోడల్‌ను సమూహపరుస్తాయి. సానుకూల మరియు ప్రతికూల పాఠాలను వర్గీకరించడానికి పైప్‌లైన్‌ను త్వరగా ఎలా ఉపయోగించాలో ఇక్కడ ఉంది:
-
-```python
->>> from transformers import pipeline
-
-# Allocate a pipeline for sentiment-analysis
->>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to introduce pipeline to the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
-```
-
-రెండవ లైన్ కోడ్ డౌన్‌లోడ్ మరియు పైప్‌లైన్ ఉపయోగించే ప్రీట్రైన్డ్ మోడల్‌ను కాష్ చేస్తుంది, మూడవది ఇచ్చిన టెక్స్ట్‌పై మూల్యాంకనం చేస్తుంది. ఇక్కడ సమాధానం 99.97% విశ్వాసంతో "పాజిటివ్".
-
-చాలా పనులు NLPలో కానీ కంప్యూటర్ విజన్ మరియు స్పీచ్‌లో కూడా ముందుగా శిక్షణ పొందిన `pipeline` సిద్ధంగా ఉన్నాయి. ఉదాహరణకు, మనం చిత్రంలో గుర్తించిన వస్తువులను సులభంగా సంగ్రహించవచ్చు:
-
-``` python
->>> import requests
->>> from PIL import Image
->>> from transformers import pipeline
-
-# Download an image with cute cats
->>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
->>> image_data = requests.get(url, stream=True).raw
->>> image = Image.open(image_data)
-
-# Allocate a pipeline for object detection
->>> object_detector = pipeline('object-detection')
->>> object_detector(image)
-[{'score': 0.9982201457023621,
-  'label': 'remote',
-  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
- {'score': 0.9960021376609802,
-  'label': 'remote',
-  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
- {'score': 0.9954745173454285,
-  'label': 'couch',
-  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
- {'score': 0.9988006353378296,
-  'label': 'cat',
-  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
- {'score': 0.9986783862113953,
-  'label': 'cat',
-  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
-```
-
-ఇక్కడ మనం ఆబ్జెక్ట్ చుట్టూ ఉన్న బాక్స్ మరియు కాన్ఫిడెన్స్ స్కోర్‌తో చిత్రంలో గుర్తించబడిన వస్తువుల జాబితాను పొందుతాము. ఇక్కడ ఎడమవైపున ఉన్న అసలు చిత్రం, కుడివైపున అంచనాలు ప్రదర్శించబడతాయి:
-
-<h3 align="center">
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
-</h3>
-
-మీరు [ఈ ట్యుటోరియల్](https://huggingface.co/docs/transformers/task_summary)లో `pipeline` API ద్వారా సపోర్ట్ చేసే టాస్క్‌ల గురించి మరింత తెలుసుకోవచ్చు.
-
-`pipeline`తో పాటు, మీరు ఇచ్చిన టాస్క్‌లో ఏదైనా ప్రీట్రైన్డ్ మోడల్‌లను డౌన్‌లోడ్ చేయడానికి మరియు ఉపయోగించడానికి, దీనికి మూడు లైన్ల కోడ్ సరిపోతుంది. ఇక్కడ PyTorch వెర్షన్ ఉంది:
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = AutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
-```
-
-మరియు TensorFlow కి సమానమైన కోడ్ ఇక్కడ ఉంది:
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
-```
-
-ప్రిట్రైన్డ్ మోడల్ ఆశించే అన్ని ప్రీప్రాసెసింగ్‌లకు టోకెనైజర్ బాధ్యత వహిస్తుంది మరియు నేరుగా ఒకే స్ట్రింగ్ (పై ఉదాహరణలలో వలె) లేదా జాబితాపై కాల్ చేయవచ్చు. ఇది మీరు డౌన్‌స్ట్రీమ్ కోడ్‌లో ఉపయోగించగల నిఘంటువుని అవుట్‌పుట్ చేస్తుంది లేదా ** ఆర్గ్యుమెంట్ అన్‌ప్యాకింగ్ ఆపరేటర్‌ని ఉపయోగించి నేరుగా మీ మోడల్‌కి పంపుతుంది.
-
-మోడల్ కూడా సాధారణ [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) లేదా [TensorFlow `tf.keras.Model`]( https://www.tensorflow.org/api_docs/python/tf/keras/Model) (మీ బ్యాకెండ్‌ని బట్టి) మీరు మామూలుగా ఉపయోగించవచ్చు. [ఈ ట్యుటోరియల్](https://huggingface.co/docs/transformers/training) అటువంటి మోడల్‌ని క్లాసిక్ PyTorch లేదా TensorFlow ట్రైనింగ్ లూప్‌లో ఎలా ఇంటిగ్రేట్ చేయాలో లేదా మా `Trainer` API ని ఎలా ఉపయోగించాలో వివరిస్తుంది కొత్త డేటాసెట్.
-
-## నేను ట్రాన్స్‌ఫార్మర్‌లను ఎందుకు ఉపయోగించాలి?
-
-1. ఉపయోగించడానికి సులభమైన స్టేట్ ఆఫ్ ది ఆర్ట్ మోడల్‌లు:
-    - సహజ భాషా అవగాహన & ఉత్పత్తి, కంప్యూటర్ దృష్టి మరియు ఆడియో పనులపై అధిక పనితీరు.
-    - విద్యావేత్తలు మరియు అభ్యాసకుల ప్రవేశానికి తక్కువ అవరోధం.
-    - తెలుసుకోవడానికి కేవలం మూడు తరగతులతో కొన్ని వినియోగదారు-ముఖ సంగ్రహణలు.
-    - మా అన్ని ప్రీట్రైన్డ్ మోడల్‌లను ఉపయోగించడం కోసం ఏకీకృత API.
-  
-2. తక్కువ గణన ఖర్చులు, చిన్న కార్బన్ పాదముద్ర:
-    - పరిశోధకులు ఎల్లప్పుడూ మళ్లీ శిక్షణ పొందే బదులు శిక్షణ పొందిన నమూనాలను పంచుకోవచ్చు.
-    - అభ్యాసకులు గణన సమయాన్ని మరియు ఉత్పత్తి ఖర్చులను తగ్గించగలరు.
-    - అన్ని పద్ధతుల్లో 60,000 కంటే ఎక్కువ ప్రీట్రైన్డ్ మోడల్‌లతో డజన్ల కొద్దీ ఆర్కిటెక్చర్‌లు.
-  
-3. మోడల్ జీవితకాలంలో ప్రతి భాగానికి సరైన ఫ్రేమ్‌వర్క్‌ను ఎంచుకోండి:
-    - 3 లైన్ల కోడ్‌లో స్టేట్ ఆఫ్ ది ఆర్ట్ మోడల్‌లకు శిక్షణ ఇవ్వండి.
-    - TF2.0/PyTorch/JAX ఫ్రేమ్‌వర్క్‌ల మధ్య ఒకే మోడల్‌ను ఇష్టానుసారంగా తరలించండి.
-    - శిక్షణ, మూల్యాంకనం మరియు ఉత్పత్తి కోసం సరైన ఫ్రేమ్‌వర్క్‌ను సజావుగా ఎంచుకోండి.
-
-4. మీ అవసరాలకు అనుగుణంగా మోడల్ లేదా ఉదాహరణను సులభంగా అనుకూలీకరించండి:
-    - ప్రతి ఆర్కిటెక్చర్ దాని అసలు రచయితలు ప్రచురించిన ఫలితాలను పునరుత్పత్తి చేయడానికి మేము ఉదాహరణలను అందిస్తాము.
-    - మోడల్ ఇంటర్నల్‌లు వీలైనంత స్థిరంగా బహిర్గతమవుతాయి.
-    - శీఘ్ర ప్రయోగాల కోసం లైబ్రరీ నుండి స్వతంత్రంగా మోడల్ ఫైల్‌లను ఉపయోగించవచ్చు.
-
-## నేను ట్రాన్స్‌ఫార్మర్‌లను ఎందుకు ఉపయోగించకూడదు?
-
-- ఈ లైబ్రరీ న్యూరల్ నెట్‌ల కోసం బిల్డింగ్ బ్లాక్‌ల మాడ్యులర్ టూల్‌బాక్స్ కాదు. మోడల్ ఫైల్‌లలోని కోడ్ ఉద్దేశపూర్వకంగా అదనపు సంగ్రహణలతో రీఫ్యాక్టరింగ్ చేయబడదు, తద్వారా పరిశోధకులు అదనపు సంగ్రహణలు/ఫైళ్లలోకి ప్రవేశించకుండా ప్రతి మోడల్‌పై త్వరగా మళ్లించగలరు.
-- శిక్షణ API ఏ మోడల్‌లో పని చేయడానికి ఉద్దేశించబడలేదు కానీ లైబ్రరీ అందించిన మోడల్‌లతో పని చేయడానికి ఆప్టిమైజ్ చేయబడింది. సాధారణ మెషిన్ లెర్నింగ్ లూప్‌ల కోసం, మీరు మరొక లైబ్రరీని ఉపయోగించాలి (బహుశా, [Accelerate](https://huggingface.co/docs/accelerate)).
-- మేము వీలైనన్ని ఎక్కువ వినియోగ సందర్భాలను ప్రదర్శించడానికి ప్రయత్నిస్తున్నప్పుడు, మా [ఉదాహరణల ఫోల్డర్](https://github.com/huggingface/transformers/tree/main/examples)లోని స్క్రిప్ట్‌లు కేవలం: ఉదాహరణలు. మీ నిర్దిష్ట సమస్యపై అవి పని చేయవు మరియు వాటిని మీ అవసరాలకు అనుగుణంగా మార్చుకోవడానికి మీరు కొన్ని కోడ్ లైన్‌లను మార్చవలసి ఉంటుంది.
-
-## సంస్థాపన
-
-### పిప్ తో
-
-ఈ రిపోజిటరీ పైథాన్ 3.8+, ఫ్లాక్స్ 0.4.1+, PyTorch 1.10+ 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.
-
-ముందుగా, మీరు ఉపయోగించబోతున్న పైథాన్ వెర్షన్‌తో వర్చువల్ వాతావరణాన్ని సృష్టించండి మరియు దానిని సక్రియం చేయండి.
-
-అప్పుడు, మీరు ఫ్లాక్స్, పైటార్చ్ లేదా టెన్సర్‌ఫ్లోలో కనీసం ఒకదానిని ఇన్‌స్టాల్ చేయాలి.
-దయచేసి [TensorFlow ఇన్‌స్టాలేషన్ పేజీ](https://www.tensorflow.org/install/), [PyTorch ఇన్‌స్టాలేషన్ పేజీ](https://pytorch.org/get-started/locally/#start-locally) మరియు/ని చూడండి లేదా మీ ప్లాట్‌ఫారమ్ కోసం నిర్దిష్ట ఇన్‌స్టాలేషన్ కమాండ్‌కు సంబంధించి [Flax](https://github.com/google/flax#quick-install) మరియు [Jax](https://github.com/google/jax#installation) ఇన్‌స్టాలేషన్ పేజీలు .
-
-ఆ బ్యాకెండ్‌లలో ఒకటి ఇన్‌స్టాల్ చేయబడినప్పుడు, 🤗 ట్రాన్స్‌ఫార్మర్‌లను ఈ క్రింది విధంగా పిప్‌ని ఉపయోగించి ఇన్‌స్టాల్ చేయవచ్చు:
-
-```bash
-pip install transformers
-```
-
-మీరు ఉదాహరణలతో ప్లే చేయాలనుకుంటే లేదా కోడ్ యొక్క బ్లీడింగ్ ఎడ్జ్ అవసరం మరియు కొత్త విడుదల కోసం వేచి ఉండలేకపోతే, మీరు తప్పనిసరిగా [మూలం నుండి లైబ్రరీని ఇన్‌స్టాల్ చేయాలి](https://huggingface.co/docs/transformers/installation#installing-from-source).
-
-### కొండా తో
-
-ట్రాన్స్‌ఫార్మర్స్ వెర్షన్ v4.0.0 నుండి, మేము ఇప్పుడు కొండా ఛానెల్‌ని కలిగి ఉన్నాము: `huggingface`.
-
-🤗 కింది విధంగా కొండా ఉపయోగించి ట్రాన్స్‌ఫార్మర్‌లను ఇన్‌స్టాల్ చేయవచ్చు:
-
-```shell script
-conda install -c huggingface transformers
-```
-
-Flax, PyTorch లేదా TensorFlow యొక్క ఇన్‌స్టాలేషన్ పేజీలను కొండాతో ఎలా ఇన్‌స్టాల్ చేయాలో చూడటానికి వాటిని అనుసరించండి.
-
-> **_గమనిక:_** Windowsలో, కాషింగ్ నుండి ప్రయోజనం పొందేందుకు మీరు డెవలపర్ మోడ్‌ని సక్రియం చేయమని ప్రాంప్ట్ చేయబడవచ్చు. ఇది మీకు ఎంపిక కాకపోతే, దయచేసి [ఈ సంచిక](https://github.com/huggingface/huggingface_hub/issues/1062)లో మాకు తెలియజేయండి.
-
-## మోడల్ ఆర్కిటెక్చర్లు
-
-**[అన్ని మోడల్ చెక్‌పాయింట్‌లు](https://huggingface.co/models)** 🤗 అందించిన ట్రాన్స్‌ఫార్మర్లు huggingface.co [model hub](https://huggingface.co/models) నుండి సజావుగా ఏకీకృతం చేయబడ్డాయి [users](https://huggingface.co/users) మరియు [organizations](https://huggingface.co/organizations) ద్వారా నేరుగా అప్‌లోడ్ చేయబడతాయి.
-
-ప్రస్తుత తనిఖీ కేంద్రాల సంఖ్య: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 ట్రాన్స్‌ఫార్మర్లు ప్రస్తుతం కింది ఆర్కిటెక్చర్‌లను అందజేస్తున్నాయి (వాటిలో ప్రతి ఒక్కటి ఉన్నత స్థాయి సారాంశం కోసం [ఇక్కడ](https://huggingface.co/docs/transformers/model_summary) చూడండి):
-
-1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
-1. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (from Google Research) released with the paper [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.
-1. **[AltCLIP](https://huggingface.co/docs/transformers/model_doc/altclip)** (from BAAI) released with the paper [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679) by Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell.
-1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (from MIT) released with the paper [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) by Yuan Gong, Yu-An Chung, James Glass.
-1. **[Autoformer](https://huggingface.co/docs/transformers/model_doc/autoformer)** (from Tsinghua University) released with the paper [Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting](https://arxiv.org/abs/2106.13008) by Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long.
-1. **[Bark](https://huggingface.co/docs/transformers/model_doc/bark)** (from Suno) released in the repository [suno-ai/bark](https://github.com/suno-ai/bark) by Suno AI team.
-1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov, and Luke Zettlemoyer.
-1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
-1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
-1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
-1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
-1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
-1. **[BioGpt](https://huggingface.co/docs/transformers/model_doc/biogpt)** (from Microsoft Research AI4Science) released with the paper [BioGPT: generative pre-trained transformer for biomedical text generation and mining](https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbac409/6713511?guestAccessKey=a66d9b5d-4f83-4017-bb52-405815c907b9) by Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon and Tie-Yan Liu.
-1. **[BiT](https://huggingface.co/docs/transformers/model_doc/bit)** (from Google AI) released with the paper [Big Transfer (BiT): General Visual Representation Learning](https://arxiv.org/abs/1912.11370) by Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil Houlsby.
-1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-1. **[BLIP](https://huggingface.co/docs/transformers/model_doc/blip)** (from Salesforce) released with the paper [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://arxiv.org/abs/2201.12086) by Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi.
-1. **[BLIP-2](https://huggingface.co/docs/transformers/model_doc/blip-2)** (from Salesforce) released with the paper [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://arxiv.org/abs/2301.12597) by Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi.
-1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigScience Workshop](https://bigscience.huggingface.co/).
-1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
-1. **[BridgeTower](https://huggingface.co/docs/transformers/model_doc/bridgetower)** (from Harbin Institute of Technology/Microsoft Research Asia/Intel Labs) released with the paper [BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan.
-1. **[BROS](https://huggingface.co/docs/transformers/model_doc/bros)** (from NAVER CLOVA) released with the paper [BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents](https://arxiv.org/abs/2108.04539) by Teakgyu Hong, Donghyun Kim, Mingi Ji, Wonseok Hwang, Daehyun Nam, Sungrae Park.
-1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
-1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
-1. **[Chinese-CLIP](https://huggingface.co/docs/transformers/model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
-1. **[CLAP](https://huggingface.co/docs/transformers/model_doc/clap)** (from LAION-AI) released with the paper [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://arxiv.org/abs/2211.06687) by Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, Shlomo Dubnov.
-1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
-1. **[CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)** (from University of Göttingen) released with the paper [Image Segmentation Using Text and Image Prompts](https://arxiv.org/abs/2112.10003) by Timo Lüddecke and Alexander Ecker.
-1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
-1. **[CodeLlama](https://huggingface.co/docs/transformers/model_doc/llama_code)** (from MetaAI) released with the paper [Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/) by Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong, Alexandre Défossez, Jade Copet, Faisal Azhar, Hugo Touvron, Louis Martin, Nicolas Usunier, Thomas Scialom, Gabriel Synnaeve.
-1. **[Conditional DETR](https://huggingface.co/docs/transformers/model_doc/conditional_detr)** (from Microsoft Research Asia) released with the paper [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang.
-1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
-1. **[ConvNeXTV2](https://huggingface.co/docs/transformers/model_doc/convnextv2)** (from Facebook AI) released with the paper [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) by Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, Saining Xie.
-1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-1. **[CPM-Ant](https://huggingface.co/docs/transformers/model_doc/cpmant)** (from OpenBMB) released by the [OpenBMB](https://www.openbmb.org/).
-1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
-1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
-1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
-1. **[Deformable DETR](https://huggingface.co/docs/transformers/model_doc/deformable_detr)** (from SenseTime Research) released with the paper [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://arxiv.org/abs/2010.04159) by Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.
-1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-1. **[DePlot](https://huggingface.co/docs/transformers/model_doc/deplot)** (from Google AI) released with the paper [DePlot: One-shot visual language reasoning by plot-to-table translation](https://arxiv.org/abs/2212.10505) by Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, Yasemin Altun.
-1. **[DETA](https://huggingface.co/docs/transformers/model_doc/deta)** (from The University of Texas at Austin) released with the paper [NMS Strikes Back](https://arxiv.org/abs/2212.06137) by Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, Philipp Krähenbühl.
-1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
-1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DiNAT](https://huggingface.co/docs/transformers/model_doc/dinat)** (from SHI Labs) released with the paper [Dilated Neighborhood Attention Transformer](https://arxiv.org/abs/2209.15001) by Ali Hassani and Humphrey Shi.
-1. **[DINOv2](https://huggingface.co/docs/transformers/model_doc/dinov2)** (from Meta AI) released with the paper [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
-1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
-1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
-1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
-1. **[EfficientFormer](https://huggingface.co/docs/transformers/model_doc/efficientformer)** (from Snap Research) released with the paper [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191) by Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren.
-1. **[EfficientNet](https://huggingface.co/docs/transformers/model_doc/efficientnet)** (from Google Brain) released with the paper [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946) by Mingxing Tan, Quoc V. Le.
-1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-1. **[EnCodec](https://huggingface.co/docs/transformers/model_doc/encodec)** (from Meta AI) released with the paper [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) by Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.
-1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
-1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (from Baidu) released with the paper [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) by Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang.
-1. **[ESM](https://huggingface.co/docs/transformers/model_doc/esm)** (from Meta AI) are transformer protein language models.  **ESM-1b** was released with the paper [Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. **ESM-1v** was released with the paper [Language models enable zero-shot prediction of the effects of mutations on protein function](https://doi.org/10.1101/2021.07.09.450648) by Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu and Alexander Rives. **ESM-2 and ESMFold** were released with the paper [Language models of protein sequences at the scale of evolution enable accurate structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives.
-1. **[Falcon](https://huggingface.co/docs/transformers/model_doc/falcon)** (from Technology Innovation Institute) by Almazrouei, Ebtesam and Alobeidli, Hamza and Alshamsi, Abdulaziz and Cappelli, Alessandro and Cojocaru, Ruxandra and Debbah, Merouane and Goffinet, Etienne and Heslow, Daniel and Launay, Julien and Malartic, Quentin and Noune, Badreddine and Pannier, Baptiste and Penedo, Guilherme.
-1. **[FLAN-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-t5-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
-1. **[FLAN-UL2](https://huggingface.co/docs/transformers/model_doc/flan-ul2)** (from Google AI) released in the repository [google-research/t5x](https://github.com/google-research/t5x/blob/main/docs/models.md#flan-ul2-checkpoints) by Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei
-1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
-1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
-1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
-1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
-1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
-1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
-1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://openai.com/research/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
-1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
-1. **[GPT NeoX Japanese](https://huggingface.co/docs/transformers/model_doc/gpt_neox_japanese)** (from ABEJA) released by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori.
-1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://openai.com/research/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
-1. **[GPT-Sw3](https://huggingface.co/docs/transformers/model_doc/gpt-sw3)** (from AI-Sweden) released with the paper [Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.376.pdf) by Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren.
-1. **[GPTBigCode](https://huggingface.co/docs/transformers/model_doc/gpt_bigcode)** (from BigCode) released with the paper [SantaCoder: don't reach for the stars!](https://arxiv.org/abs/2301.03988) by Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi, Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin, Dmitry Abulkhanov, Manuel Romero, Michael Lappert, Francesco De Toni, Bernardo García del Río, Qian Liu, Shamik Bose, Urvashi Bhattacharyya, Terry Yue Zhuo, Ian Yu, Paulo Villegas, Marco Zocca, Sourab Mangrulkar, David Lansky, Huu Nguyen, Danish Contractor, Luis Villa, Jia Li, Dzmitry Bahdanau, Yacine Jernite, Sean Hughes, Daniel Fried, Arjun Guha, Harm de Vries, Leandro von Werra.
-1. **[GPTSAN-japanese](https://huggingface.co/docs/transformers/model_doc/gptsan-japanese)** released in the repository [tanreinama/GPTSAN](https://github.com/tanreinama/GPTSAN/blob/main/report/model.md) by Toshiyuki Sakamoto(tanreinama).
-1. **[Graphormer](https://huggingface.co/docs/transformers/model_doc/graphormer)** (from Microsoft) released with the paper [Do Transformers Really Perform Bad for Graph Representation?](https://arxiv.org/abs/2106.05234) by Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, Tie-Yan Liu.
-1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
-1. **[HerBERT](https://huggingface.co/docs/transformers/model_doc/herbert)** (from Allegro.pl, AGH University of Science and Technology) released with the paper [KLEJ: Comprehensive Benchmark for Polish Language Understanding](https://www.aclweb.org/anthology/2020.acl-main.111.pdf) by Piotr Rybak, Robert Mroczkowski, Janusz Tracz, Ireneusz Gawlik.
-1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
-1. **[IDEFICS](https://huggingface.co/docs/transformers/model_doc/idefics)** (from HuggingFace) released with the paper [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://huggingface.co/papers/2306.16527) by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh.
-1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
-1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
-1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
-1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
-1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
-1. **[LiLT](https://huggingface.co/docs/transformers/model_doc/lilt)** (from South China University of Technology) released with the paper [LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding](https://arxiv.org/abs/2202.13669) by Jiapeng Wang, Lianwen Jin, Kai Ding.
-1. **[LLaMA](https://huggingface.co/docs/transformers/model_doc/llama)** (from The FAIR team of Meta AI) released with the paper [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971) by Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, Guillaume Lample.
-1. **[Llama2](https://huggingface.co/docs/transformers/model_doc/llama2)** (from The FAIR team of Meta AI) released with the paper [Llama2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/) by Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushka rMishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing EllenTan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom.
-1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
-1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
-1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
-1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (from Microsoft Research Asia) released with the paper [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) by Junlong Li, Yiheng Xu, Lei Cui, Furu Wei.
-1. **[Mask2Former](https://huggingface.co/docs/transformers/model_doc/mask2former)** (from FAIR and UIUC) released with the paper [Masked-attention Mask Transformer for Universal Image Segmentation](https://arxiv.org/abs/2112.01527) by Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, Rohit Girdhar.
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
-1. **[MatCha](https://huggingface.co/docs/transformers/model_doc/matcha)** (from Google AI) released with the paper [MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering](https://arxiv.org/abs/2212.09662) by Fangyu Liu, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Yasemin Altun, Nigel Collier, Julian Martin Eisenschlos.
-1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-1. **[MEGA](https://huggingface.co/docs/transformers/model_doc/mega)** (from Meta/USC/CMU/SJTU) released with the paper [Mega: Moving Average Equipped Gated Attention](https://arxiv.org/abs/2209.10655) by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer.
-1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-1. **[MGP-STR](https://huggingface.co/docs/transformers/model_doc/mgp-str)** (from Alibaba Research) released with the paper [Multi-Granularity Prediction for Scene Text Recognition](https://arxiv.org/abs/2209.03592) by Peng Wang, Cheng Da, and Cong Yao.
-1. **[Mistral](https://huggingface.co/docs/transformers/model_doc/mistral)** (from Mistral AI) by The [Mistral AI](https://mistral.ai) team: Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
-1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
-1. **[MMS](https://huggingface.co/docs/transformers/model_doc/mms)** (from Facebook) released with the paper [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516) by Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, Michael Auli.
-1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
-1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
-1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
-1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
-1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
-1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
-1. **[MPT](https://huggingface.co/docs/transformers/model_doc/mpt)** (from MosaiML) released with the repository [llm-foundry](https://github.com/mosaicml/llm-foundry/) by the MosaicML NLP Team.
-1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
-1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
-1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
-1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
-1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
-1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
-1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
-1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
-1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
-1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
-1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
-1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
-1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
-1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
-1. **[Persimmon](https://huggingface.co/docs/transformers/model_doc/persimmon)** (from ADEPT) released in a [blog post](https://www.adept.ai/blog/persimmon-8b) by Erich Elsen, Augustus Odena, Maxwell Nye, Sağnak Taşırlar, Tri Dao, Curtis Hawthorne, Deepak Moparthi, Arushi Somani.
-1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
-1. **[Pix2Struct](https://huggingface.co/docs/transformers/model_doc/pix2struct)** (from Google) released with the paper [Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding](https://arxiv.org/abs/2210.03347) by Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, Kristina Toutanova.
-1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
-1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
-1. **[Pop2Piano](https://huggingface.co/docs/transformers/model_doc/pop2piano)** released with the paper [Pop2Piano : Pop Audio-based Piano Cover Generation](https://arxiv.org/abs/2211.00895) by Jongho Choi and Kyogu Lee.
-1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[PVT](https://huggingface.co/docs/transformers/model_doc/pvt)** (from Nanjing University, The University of Hong Kong etc.) released with the paper [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://arxiv.org/pdf/2102.12122.pdf) by Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, Ling Shao.
-1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
-1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
-1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
-1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
-1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
-1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
-1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
-1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng), released on [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
-1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
-1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
-1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
-1. **[SpeechT5](https://huggingface.co/docs/transformers/model_doc/speecht5)** (from Microsoft Research) released with the paper [SpeechT5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing](https://arxiv.org/abs/2110.07205) by Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, Zhihua Wei, Yao Qian, Jinyu Li, Furu Wei.
-1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
-1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
-1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-1. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (from MBZUAI) released with the paper [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) by Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.
-1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
-1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
-1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (from University of Würzburg) released with the paper [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) by Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte.
-1. **[SwitchTransformers](https://huggingface.co/docs/transformers/model_doc/switch_transformers)** (from Google) released with the paper [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961) by William Fedus, Barret Zoph, Noam Shazeer.
-1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-1. **[Table Transformer](https://huggingface.co/docs/transformers/model_doc/table-transformer)** (from Microsoft Research) released with the paper [PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents](https://arxiv.org/abs/2110.00061) by Brandon Smock, Rohith Pesala, Robin Abraham.
-1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
-1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
-1. **[TimeSformer](https://huggingface.co/docs/transformers/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
-1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
-1. **[TVLT](https://huggingface.co/docs/transformers/model_doc/tvlt)** (from UNC Chapel Hill) released with the paper [TVLT: Textless Vision-Language Transformer](https://arxiv.org/abs/2209.14156) by Zineng Tang, Jaemin Cho, Yixin Nie, Mohit Bansal.
-1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
-1. **[UMT5](https://huggingface.co/docs/transformers/model_doc/umt5)** (from Google Research) released with the paper [UniMax: Fairer and More Effective Language Sampling for Large-Scale Multilingual Pretraining](https://openreview.net/forum?id=kXwdL1cWOAi) by Hyung Won Chung, Xavier Garcia, Adam Roberts, Yi Tay, Orhan Firat, Sharan Narang, Noah Constant.
-1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
-1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
-1. **[UPerNet](https://huggingface.co/docs/transformers/model_doc/upernet)** (from Peking University) released with the paper [Unified Perceptual Parsing for Scene Understanding](https://arxiv.org/abs/1807.10221) by Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, Jian Sun.
-1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
-1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
-1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
-1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-1. **[ViT Hybrid](https://huggingface.co/docs/transformers/model_doc/vit_hybrid)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-1. **[VitDet](https://huggingface.co/docs/transformers/model_doc/vitdet)** (from Meta AI) released with the paper [Exploring Plain Vision Transformer Backbones for Object Detection](https://arxiv.org/abs/2203.16527) by Yanghao Li, Hanzi Mao, Ross Girshick, Kaiming He.
-1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
-1. **[ViTMatte](https://huggingface.co/docs/transformers/model_doc/vitmatte)** (from HUST-VL) released with the paper [ViTMatte: Boosting Image Matting with Pretrained Plain Vision Transformers](https://arxiv.org/abs/2305.15272) by Jingfeng Yao, Xinggang Wang, Shusheng Yang, Baoyuan Wang.
-1. **[ViTMSN](https://huggingface.co/docs/transformers/model_doc/vit_msn)** (from Meta AI) released with the paper [Masked Siamese Networks for Label-Efficient Learning](https://arxiv.org/abs/2204.07141) by Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Florian Bordes, Pascal Vincent, Armand Joulin, Michael Rabbat, Nicolas Ballas.
-1. **[VITS](https://huggingface.co/docs/transformers/model_doc/vits)** (from Kakao Enterprise) released with the paper [Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech](https://arxiv.org/abs/2106.06103) by Jaehyeon Kim, Jungil Kong, Juhee Son.
-1. **[ViViT](https://huggingface.co/docs/transformers/model_doc/vivit)** (from Google Research) released with the paper [ViViT: A Video Vision Transformer](https://arxiv.org/abs/2103.15691) by Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lučić, Cordelia Schmid.
-1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
-1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
-1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
-1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
-1. **[Whisper](https://huggingface.co/docs/transformers/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
-1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
-1. **[X-MOD](https://huggingface.co/docs/transformers/model_doc/xmod)** (from Meta AI) released with the paper [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](http://dx.doi.org/10.18653/v1/2022.naacl-main.255) by Jonas Pfeiffer, Naman Goyal, Xi Lin, Xian Li, James Cross, Sebastian Riedel, Mikel Artetxe.
-1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
-1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
-1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
-1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
-1. **[XLM-V](https://huggingface.co/docs/transformers/model_doc/xlm-v)** (from Meta AI) released with the paper [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://arxiv.org/abs/2301.10472) by Davis Liang, Hila Gonen, Yuning Mao, Rui Hou, Naman Goyal, Marjan Ghazvininejad, Luke Zettlemoyer, Madian Khabsa.
-1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
-1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
-1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
-1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
-1. కొత్త మోడల్‌ను అందించాలనుకుంటున్నారా? కొత్త మోడల్‌ను జోడించే ప్రక్రియలో మీకు మార్గనిర్దేశం చేసేందుకు మేము **వివరణాత్మక గైడ్ మరియు టెంప్లేట్‌లను** జోడించాము. మీరు వాటిని రిపోజిటరీ యొక్క [`టెంప్లేట్లు`](./టెంప్లేట్లు) ఫోల్డర్‌లో కనుగొనవచ్చు. మీ PRని ప్రారంభించడానికి ముందు [సహకార మార్గదర్శకాలు](./CONTRIBUTING.md)ని తనిఖీ చేసి, నిర్వహణదారులను సంప్రదించండి లేదా అభిప్రాయాన్ని సేకరించడానికి సమస్యను తెరవండి.
-
-ప్రతి మోడల్ ఫ్లాక్స్, పైటార్చ్ లేదా టెన్సర్‌ఫ్లోలో అమలు చేయబడిందా లేదా 🤗 Tokenizers లైబ్రరీ ద్వారా అనుబంధించబడిన టోకెనైజర్‌ని కలిగి ఉందో లేదో తనిఖీ చేయడానికి, [ఈ పట్టిక](https://huggingface.co/docs/transformers/index#supported-frameworks).
-
-ఈ అమలులు అనేక డేటాసెట్‌లలో పరీక్షించబడ్డాయి (ఉదాహరణ స్క్రిప్ట్‌లను చూడండి) మరియు అసలైన అమలుల పనితీరుతో సరిపోలాలి. మీరు [డాక్యుమెంటేషన్](https://github.com/huggingface/transformers/tree/main/examples) యొక్క ఉదాహరణల విభాగంలో పనితీరుపై మరిన్ని వివరాలను కనుగొనవచ్చు.
-
-## ఇంకా నేర్చుకో
-
-| విభాగం | వివరణ |
-|-|-|
-| [డాక్యుమెంటేషన్](https://huggingface.co/docs/transformers/) | పూర్తి API డాక్యుమెంటేషన్ మరియు ట్యుటోరియల్స్ |
-| [టాస్క్ సారాంశం](https://huggingface.co/docs/transformers/task_summary) | 🤗 ట్రాన్స్‌ఫార్మర్‌ల ద్వారా సపోర్ట్ చేయబడిన విధులు |
-| [ప్రీప్రాసెసింగ్ ట్యుటోరియల్](https://huggingface.co/docs/transformers/preprocessing) | మోడల్‌ల కోసం డేటాను సిద్ధం చేయడానికి `Tokenizer` క్లాస్‌ని ఉపయోగించడం |
-| [ట్రైనింగ్ మరియు ఫైన్-ట్యూనింగ్](https://huggingface.co/docs/transformers/training) | PyTorch/TensorFlow ట్రైనింగ్ లూప్ మరియు `Trainer` APIలో 🤗 ట్రాన్స్‌ఫార్మర్లు అందించిన మోడల్‌లను ఉపయోగించడం |
-| [త్వరిత పర్యటన: ఫైన్-ట్యూనింగ్/యూసేజ్ స్క్రిప్ట్‌లు](https://github.com/huggingface/transformers/tree/main/examples) | విస్తృత శ్రేణి టాస్క్‌లపై ఫైన్-ట్యూనింగ్ మోడల్స్ కోసం ఉదాహరణ స్క్రిప్ట్‌లు |
-| [మోడల్ భాగస్వామ్యం మరియు అప్‌లోడ్ చేయడం](https://huggingface.co/docs/transformers/model_sharing) | కమ్యూనిటీతో మీ ఫైన్-ట్యూన్డ్ మోడల్‌లను అప్‌లోడ్ చేయండి మరియు భాగస్వామ్యం చేయండి |
-
-## అనులేఖనం
-
-🤗 ట్రాన్స్‌ఫార్మర్స్ లైబ్రరీ కోసం మీరు ఉదహరించగల [పేపర్](https://www.aclweb.org/anthology/2020.emnlp-demos.6/) ఇప్పుడు మా వద్ద ఉంది:
-```bibtex
-@inproceedings{wolf-etal-2020-transformers,
-    title = "Transformers: State-of-the-Art Natural Language Processing",
-    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
-    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
-    month = oct,
-    year = "2020",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
-    pages = "38--45"
-}
-```

README_zh-hans.md

@@ -72,7 +72,6 @@ checkpoint: 检查点
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
     </p>
 </h4>
 
@@ -312,7 +311,6 @@ conda install -c huggingface transformers
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (来自 Google Research) 伴随论文 [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) 由 James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon 发布。
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (来自 Microsoft Research) 伴随论文 [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) 由 Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao 发布。
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (来自 CMU/Google Brain) 伴随论文 [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) 由 Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le 发布。
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (来自 ADEPT) 伴随论文 [blog post](https://www.adept.ai/blog/fuyu-8b 由 Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar 发布。)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (来自 Microsoft Research) 伴随论文 [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) 由 Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang 发布。
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (来自 KAIST) 伴随论文 [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) 由 Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim 发布。
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (来自 OpenAI) 伴随论文 [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) 由 Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever 发布。
@@ -334,7 +332,6 @@ conda install -c huggingface transformers
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (来自 Salesforce) 伴随论文 [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) 由 Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi 发布。
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) 由 Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou 发布。
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) 由 Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou 发布。
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) 由 Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei 发布。
@@ -382,10 +379,9 @@ conda install -c huggingface transformers
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (来自 Meta AI) 伴随论文 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) 由 Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic 发布。
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (来自 the University of Wisconsin - Madison) 伴随论文 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 由 Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 发布。
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (来自 SHI Labs)  伴随论文 [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) 由 Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi 发布。
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (来自 [s-JoL](https://huggingface.co/s-JoL)) 由 GitHub (现已删除).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (来自 [s-JoL](https://huggingface.co/s-JoL)) 由 [Open-Llama](https://github.com/s-JoL/Open-Llama) 发布.
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (来自 Meta AI) 伴随论文 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 由 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 发布。
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (来自 Google AI) 伴随论文 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 由 Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 发布。
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (来自 Google AI) 伴随论文 [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) 由 Matthias Minderer, Alexey Gritsenko, Neil Houlsby 发布。
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (来自 Google) 伴随论文 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 由 Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 发布。
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (来自 Google) 伴随论文 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 由 Jason Phang, Yao Zhao, Peter J. Liu 发布。
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (来自 Deepmind) 伴随论文 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 由 Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira 发布。
@@ -409,7 +405,6 @@ conda install -c huggingface transformers
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (来自 WeChatAI), 伴随论文 [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) 由 HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou 发布。
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (来自 ZhuiyiTechnology), 伴随论文 [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) 由 Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu 发布。
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (来自 Bo Peng) 伴随论文 [this repo](https://github.com/BlinkDL/RWKV-LM) 由 Bo Peng 发布。
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (来自 NVIDIA) 伴随论文 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 由 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 发布。
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (来自 Meta AI) 伴随论文 [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) 由 Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick 发布。
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。

README_zh-hant.md

@@ -84,7 +84,6 @@ user: 使用者
         <a href="https://github.com/huggingface/transformers/blob/main/README_es.md">Español</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_ja.md">日本語</a> |
         <a href="https://github.com/huggingface/transformers/blob/main/README_hd.md">हिन्दी</a>
-        <a href="https://github.com/huggingface/transformers//blob/main/README_te.md">తెలుగు</a> |
     </p>
 </h4>
 
@@ -324,7 +323,6 @@ conda install -c huggingface transformers
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
 1. **[FocalNet](https://huggingface.co/docs/transformers/model_doc/focalnet)** (from Microsoft Research) released with the paper [Focal Modulation Networks](https://arxiv.org/abs/2203.11926) by Jianwei Yang, Chunyuan Li, Xiyang Dai, Lu Yuan, Jianfeng Gao.
 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-1. **[Fuyu](https://huggingface.co/docs/transformers/model_doc/fuyu)** (from ADEPT) Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. Released with the paper [blog post](https://www.adept.ai/blog/fuyu-8b)
 1. **[GIT](https://huggingface.co/docs/transformers/model_doc/git)** (from Microsoft Research) released with the paper [GIT: A Generative Image-to-text Transformer for Vision and Language](https://arxiv.org/abs/2205.14100) by Jianfeng Wang, Zhengyuan Yang, Xiaowei Hu, Linjie Li, Kevin Lin, Zhe Gan, Zicheng Liu, Ce Liu, Lijuan Wang.
 1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -346,7 +344,6 @@ conda install -c huggingface transformers
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
 1. **[InstructBLIP](https://huggingface.co/docs/transformers/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
-1. **[KOSMOS-2](https://huggingface.co/docs/transformers/main/model_doc/kosmos-2)** (from Microsoft Research Asia) released with the paper [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
 1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
@@ -394,10 +391,9 @@ conda install -c huggingface transformers
 1. **[Nougat](https://huggingface.co/docs/transformers/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
-1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released on GitHub (now removed).
+1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
 1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
 1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
-1. **[OWLv2](https://huggingface.co/docs/transformers/main/model_doc/owlv2)** (from Google AI) released with the paper [Scaling Open-Vocabulary Object Detection](https://arxiv.org/abs/2306.09683) by Matthias Minderer, Alexey Gritsenko, Neil Houlsby.
 1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
 1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
@@ -421,7 +417,6 @@ conda install -c huggingface transformers
 1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)** (from Bo Peng) released with the paper [this repo](https://github.com/BlinkDL/RWKV-LM) by Bo Peng.
-1. **[SeamlessM4T](https://huggingface.co/docs/transformers/main/model_doc/seamless_m4t)** (from Meta AI) released with the paper [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[Segment Anything](https://huggingface.co/docs/transformers/model_doc/sam)** (from Meta AI) released with the paper [Segment Anything](https://arxiv.org/pdf/2304.02643v1.pdf) by Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alex Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

SECURITY.md

@@ -1,6 +0,0 @@
-# Security Policy
-
-## Reporting a Vulnerability
-
-🤗 We have our bug bounty program set up with HackerOne. Please feel free to submit vulnerability reports to our private program at https://hackerone.com/hugging_face.
-Note that you'll need to be invited to our program, so send us a quick email at security@huggingface.co if you've found a vulnerability.

docker/transformers-all-latest-gpu/Dockerfile

@@ -9,7 +9,7 @@ SHELL ["sh", "-lc"]
 # The following `ARG` are mainly used to specify the versions explicitly & directly in this docker file, and not meant
 # to be used as arguments for docker build (so far).
 
-ARG PYTORCH='2.1.0'
+ARG PYTORCH='2.0.1'
 # (not always a valid torch version)
 ARG INTEL_TORCH_EXT='1.11.0'
 # Example: `cu102`, `cu113`, etc.
@@ -55,9 +55,6 @@ RUN python3 -m pip install --no-cache-dir auto-gptq --extra-index-url https://hu
 # Add einops for additional model testing
 RUN python3 -m pip install --no-cache-dir einops
 
-# Add autoawq for quantization testing
-RUN python3 -m pip install --no-cache-dir autoawq
-
 # For bettertransformer + gptq 
 RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/optimum@main#egg=optimum
 

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

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

docs/source/en/_redirects.yml

@@ -1,3 +0,0 @@
-# Optimizing inference
-
-perf_infer_gpu_many: perf_infer_gpu_one

docs/source/en/_toctree.yml

@@ -71,10 +71,6 @@
       title: Zero-shot image classification
     - local: tasks/monocular_depth_estimation
       title: Depth estimation
-    - local: tasks/image_to_image
-      title: Image-to-Image
-    - local: tasks/knowledge_distillation_for_image_classification
-      title: Knowledge Distillation for Computer Vision
     title: Computer Vision
   - isExpanded: false
     sections:
@@ -96,8 +92,6 @@
     sections:
     - local: tasks/idefics
       title: Image tasks with IDEFICS
-    - local: tasks/prompting
-      title: LLM prompting guide
     title: Prompting
   title: Task Guides
 - sections:
@@ -155,9 +149,13 @@
     title: Efficient training techniques
   - sections:
     - local: perf_infer_cpu
-      title: CPU inference
+      title: Inference on CPU
     - local: perf_infer_gpu_one
-      title: GPU inference
+      title: Inference on one GPU
+    - local: perf_infer_gpu_many
+      title: Inference on many GPUs
+    - local: perf_infer_special
+      title: Inference on Specialized Hardware
     title: Optimizing inference
   - local: big_models
     title: Instantiating a big model
@@ -207,8 +205,6 @@
     title: Pipelines for webserver inference
   - local: model_memory_anatomy
     title: Model training anatomy
-  - local: llm_tutorial_optimization
-    title: Getting the most out of LLMs
   title: Conceptual guides
 - sections:
   - sections:
@@ -338,8 +334,6 @@
         title: FSMT
       - local: model_doc/funnel
         title: Funnel Transformer
-      - local: model_doc/fuyu
-        title: Fuyu
       - local: model_doc/openai-gpt
         title: GPT
       - local: model_doc/gpt_neo
@@ -514,7 +508,7 @@
       - local: model_doc/dinat
         title: DiNAT
       - local: model_doc/dinov2
-        title: DINOV2
+        title: DINO V2
       - local: model_doc/dit
         title: DiT
       - local: model_doc/dpt
@@ -610,8 +604,6 @@
         title: MusicGen
       - local: model_doc/pop2piano
         title: Pop2Piano
-      - local: model_doc/seamless_m4t
-        title: Seamless-M4T
       - local: model_doc/sew
         title: SEW
       - local: model_doc/sew-d
@@ -679,8 +671,6 @@
         title: IDEFICS
       - local: model_doc/instructblip
         title: InstructBLIP
-      - local: model_doc/kosmos-2
-        title: KOSMOS-2
       - local: model_doc/layoutlm
         title: LayoutLM
       - local: model_doc/layoutlmv2
@@ -703,8 +693,6 @@
         title: OneFormer
       - local: model_doc/owlvit
         title: OWL-ViT
-      - local: model_doc/owlv2
-        title: OWLv2
       - local: model_doc/perceiver
         title: Perceiver
       - local: model_doc/pix2struct

docs/source/en/glossary.md

@@ -112,12 +112,6 @@ A type of layer in a neural network where the input matrix is multiplied element
 
 ## D
 
-### DataParallel (DP)
-
-Parallelism technique for training on multiple GPUs where the same setup is replicated multiple times, with each instance 
-receiving a distinct data slice. The processing is done in parallel and all setups are synchronized at the end of each training step.
-Learn more about how DataParallel works [here](perf_train_gpu_many#dataparallel-vs-distributeddataparallel).
-
 ### decoder input IDs
 
 This input is specific to encoder-decoder models, and contains the input IDs that will be fed to the decoder. These
@@ -346,12 +340,6 @@ A pipeline in 🤗 Transformers is an abstraction referring to a series of steps
 
 For more details, see [Pipelines for inference](https://huggingface.co/docs/transformers/pipeline_tutorial).
 
-### PipelineParallel (PP)
-
-Parallelism technique in which the model is split up vertically (layer-level) across multiple GPUs, so that only one or 
-several layers of the model are placed on a single GPU. Each GPU processes in parallel different stages of the pipeline 
-and working on a small chunk of the batch. Learn more about how PipelineParallel works [here](perf_train_gpu_many#from-naive-model-parallelism-to-pipeline-parallelism).
-
 ### pixel values
 
 A tensor of the numerical representations of an image that is passed to a model. The pixel values have a shape of [`batch_size`, `num_channels`, `height`, `width`], and are generated from an image processor.
@@ -422,10 +410,6 @@ An example of a semi-supervised learning approach is "self-training", in which a
 Models that generate a new sequence from an input, like translation models, or summarization models (such as
 [Bart](model_doc/bart) or [T5](model_doc/t5)).
 
-### Sharded DDP
-
-Another name for the foundational [ZeRO](#zero-redundancy-optimizer--zero-) concept as used by various other implementations of ZeRO.
-
 ### stride
 
 In [convolution](#convolution) or [pooling](#pooling), the stride refers to the distance the kernel is moved over a matrix. A stride of 1 means the kernel is moved one pixel over at a time, and a stride of 2 means the kernel is moved two pixels over at a time.
@@ -436,14 +420,6 @@ A form of model training that directly uses labeled data to correct and instruct
 
 ## T
 
-### Tensor Parallelism (TP)
-
-Parallelism technique for training on multiple GPUs in which each tensor is split up into multiple chunks, so instead of 
-having the whole tensor reside on a single GPU, each shard of the tensor resides on its designated GPU. Shards gets 
-processed separately and in parallel on different GPUs and the results are synced at the end of the processing step. 
-This is what is sometimes called horizontal parallelism, as the splitting happens on horizontal level.
-Learn more about Tensor Parallelism [here](perf_train_gpu_many#tensor-parallelism).
-
 ### token
 
 A part of a sentence, usually a word, but can also be a subword (non-common words are often split in subwords) or a
@@ -513,12 +489,3 @@ Self-attention based deep learning model architecture.
 ### unsupervised learning
 
 A form of model training in which data provided to the model is not labeled. Unsupervised learning techniques leverage statistical information of the data distribution to find patterns useful for the task at hand.
-
-## Z
-
-### Zero Redundancy Optimizer (ZeRO)
-
-Parallelism technique which performs sharding of the tensors somewhat similar to [TensorParallel](#tensorparallel--tp-), 
-except the whole tensor gets reconstructed in time for a forward or backward computation, therefore the model doesn't need 
-to be modified. This method also supports various offloading techniques to compensate for limited GPU memory. 
-Learn more about ZeRO [here](perf_train_gpu_many#zero-data-parallelism).

docs/source/en/index.md

@@ -97,7 +97,7 @@ Flax), PyTorch, and/or TensorFlow.
 |              [Conditional DETR](model_doc/conditional_detr)              |       ✅        |         ❌         |      ❌      |
 |                      [ConvBERT](model_doc/convbert)                      |       ✅        |         ✅         |      ❌      |
 |                      [ConvNeXT](model_doc/convnext)                      |       ✅        |         ✅         |      ❌      |
-|                    [ConvNeXTV2](model_doc/convnextv2)                    |       ✅        |         ✅         |      ❌      |
+|                    [ConvNeXTV2](model_doc/convnextv2)                    |       ✅        |         ❌         |      ❌      |
 |                           [CPM](model_doc/cpm)                           |       ✅        |         ✅         |      ✅      |
 |                       [CPM-Ant](model_doc/cpmant)                        |       ✅        |         ❌         |      ❌      |
 |                          [CTRL](model_doc/ctrl)                          |       ✅        |         ✅         |      ❌      |
@@ -138,7 +138,6 @@ Flax), PyTorch, and/or TensorFlow.
 |                          [FNet](model_doc/fnet)                          |       ✅        |         ❌         |      ❌      |
 |                      [FocalNet](model_doc/focalnet)                      |       ✅        |         ❌         |      ❌      |
 |                  [Funnel Transformer](model_doc/funnel)                  |       ✅        |         ✅         |      ❌      |
-|                          [Fuyu](model_doc/fuyu)                          |       ✅        |         ❌         |      ❌      |
 |                           [GIT](model_doc/git)                           |       ✅        |         ❌         |      ❌      |
 |                          [GLPN](model_doc/glpn)                          |       ✅        |         ❌         |      ❌      |
 |                       [GPT Neo](model_doc/gpt_neo)                       |       ✅        |         ❌         |      ✅      |
@@ -158,7 +157,6 @@ Flax), PyTorch, and/or TensorFlow.
 |                      [Informer](model_doc/informer)                      |       ✅        |         ❌         |      ❌      |
 |                  [InstructBLIP](model_doc/instructblip)                  |       ✅        |         ❌         |      ❌      |
 |                       [Jukebox](model_doc/jukebox)                       |       ✅        |         ❌         |      ❌      |
-|                      [KOSMOS-2](model_doc/kosmos-2)                      |       ✅        |         ❌         |      ❌      |
 |                      [LayoutLM](model_doc/layoutlm)                      |       ✅        |         ✅         |      ❌      |
 |                    [LayoutLMv2](model_doc/layoutlmv2)                    |       ✅        |         ❌         |      ❌      |
 |                    [LayoutLMv3](model_doc/layoutlmv3)                    |       ✅        |         ✅         |      ❌      |
@@ -211,7 +209,6 @@ Flax), PyTorch, and/or TensorFlow.
 |                    [OpenLlama](model_doc/open-llama)                     |       ✅        |         ❌         |      ❌      |
 |                           [OPT](model_doc/opt)                           |       ✅        |         ✅         |      ✅      |
 |                       [OWL-ViT](model_doc/owlvit)                        |       ✅        |         ❌         |      ❌      |
-|                         [OWLv2](model_doc/owlv2)                         |       ✅        |         ❌         |      ❌      |
 |                       [Pegasus](model_doc/pegasus)                       |       ✅        |         ✅         |      ✅      |
 |                     [PEGASUS-X](model_doc/pegasus_x)                     |       ✅        |         ❌         |      ❌      |
 |                     [Perceiver](model_doc/perceiver)                     |       ✅        |         ❌         |      ❌      |
@@ -237,7 +234,6 @@ Flax), PyTorch, and/or TensorFlow.
 |                      [RoFormer](model_doc/roformer)                      |       ✅        |         ✅         |      ✅      |
 |                          [RWKV](model_doc/rwkv)                          |       ✅        |         ❌         |      ❌      |
 |                           [SAM](model_doc/sam)                           |       ✅        |         ✅         |      ❌      |
-|                  [SeamlessM4T](model_doc/seamless_m4t)                   |       ✅        |         ❌         |      ❌      |
 |                     [SegFormer](model_doc/segformer)                     |       ✅        |         ✅         |      ❌      |
 |                           [SEW](model_doc/sew)                           |       ✅        |         ❌         |      ❌      |
 |                         [SEW-D](model_doc/sew-d)                         |       ✅        |         ❌         |      ❌      |

docs/source/en/llm_tutorial.md

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

docs/source/en/llm_tutorial_optimization.md

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

docs/source/en/main_classes/callback.md

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

docs/source/en/main_classes/deepspeed.md

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

docs/source/en/main_classes/feature_extractor.md

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

docs/source/en/main_classes/logging.md

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

docs/source/en/main_classes/output.md

@@ -44,7 +44,6 @@ an optional `attentions` attribute. Here we have the `loss` since we passed alon
 
 When passing `output_hidden_states=True` you may expect the `outputs.hidden_states[-1]` to match `outputs.last_hidden_states` exactly.
 However, this is not always the case. Some models apply normalization or subsequent process to the last hidden state when it's returned.
-
 </Tip>
 
 

docs/source/en/main_classes/pipelines.md

@@ -481,12 +481,6 @@ Pipelines available for multimodal tasks include the following.
     - __call__
     - all
 
-### MaskGenerationPipeline
-
-[[autodoc]] MaskGenerationPipeline
-    - __call__
-    - all
-
 ### VisualQuestionAnsweringPipeline
 
 [[autodoc]] VisualQuestionAnsweringPipeline

docs/source/en/main_classes/quantization.md

@@ -16,102 +16,11 @@ rendered properly in your Markdown viewer.
 
 # Quantize 🤗 Transformers models
 
-## AWQ integration
-
-AWQ method has been introduced in the [*AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration* paper](https://arxiv.org/abs/2306.00978). With AWQ you can run models in 4-bit precision, while preserving its original quality (i.e. no performance degradation) with a superior throughput that other quantization methods presented below - reaching similar throughput as pure `float16` inference.
-
-We now support inference with any AWQ model, meaning anyone can load and use AWQ weights that are pushed on the Hub or saved locally. Note that using AWQ requires to have access to a NVIDIA GPU. CPU inference is not supported yet. 
-
-### Quantizing a model
-
-We advise users to look at different existing tools in the ecosystem to quantize their models with AWQ algorithm, such as:
-
-- [`llm-awq`](https://github.com/mit-han-lab/llm-awq) from MIT Han Lab
-- [`autoawq`](https://github.com/casper-hansen/AutoAWQ) from [`casper-hansen`](https://github.com/casper-hansen)
-- Intel neural compressor from Intel - through [`optimum-intel`](https://huggingface.co/docs/optimum/main/en/intel/optimization_inc)
-
-Many other tools might exist in the ecosystem, please feel free to open a PR to add them to the list.
-Currently the integration with 🤗 Transformers is only available for models that have been quantized using `autoawq` library and `llm-awq`. Most of the models quantized with `auto-awq` can be found under [`TheBloke`](https://huggingface.co/TheBloke) namespace of 🤗 Hub, and to quantize models with `llm-awq` please refer to the [`convert_to_hf.py`](https://github.com/mit-han-lab/llm-awq/blob/main/examples/convert_to_hf.py) script in the examples folder of [`llm-awq`](https://github.com/mit-han-lab/llm-awq/).
-
-### Load a quantized model
-
-You can load a quantized model from the Hub using the `from_pretrained` method. Make sure that the pushed weights are quantized, by checking that the attribute `quantization_config` is present in the model's configuration file (`configuration.json`). You can confirm that the model is quantized in the AWQ format by checking the field `quantization_config.quant_method` which should be set to `"awq"`. Note that loading the model will set other weights in `float16` by default for performance reasons. If you want to change that behavior, you can pass `torch_dtype` argument to `torch.float32` or `torch.bfloat16`. You can find in the sections below some example snippets and notebook.
-
-## Example usage
-
-First, you need to install [`autoawq`](https://github.com/casper-hansen/AutoAWQ) library
-
-```bash
-pip install autoawq
-```
-
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "TheBloke/zephyr-7B-alpha-AWQ"
-model = AutoModelForCausalLM.from_pretrained(model_id, device_map="cuda:0")
-```
-
-In case you first load your model on CPU, make sure to move it to your GPU device before using 
-
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "TheBloke/zephyr-7B-alpha-AWQ"
-model = AutoModelForCausalLM.from_pretrained(model_id).to("cuda:0")
-```
-
-### Combining AWQ and Flash Attention
-
-You can combine AWQ quantization with Flash Attention to get a model that is both quantized and faster. Simply load the model using `from_pretrained` and pass `use_flash_attention_2=True` argument.
-
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model = AutoModelForCausalLM.from_pretrained("TheBloke/zephyr-7B-alpha-AWQ", use_flash_attention_2=True, device_map="cuda:0")
-```
-
-### Benchmarks
-
-We performed some speed, throughput and latency benchmarks using [`optimum-benchmark`](https://github.com/huggingface/optimum-benchmark) library. 
-
-Note at that time of writing this documentation section, the available quantization methods were: `awq`, `gptq` and `bitsandbytes`.
-
-The benchmark was run on a NVIDIA-A100 instance and the model used was [`TheBloke/Mistral-7B-v0.1-AWQ`](https://huggingface.co/TheBloke/Mistral-7B-v0.1-AWQ) for the AWQ model, [`TheBloke/Mistral-7B-v0.1-GPTQ`](https://huggingface.co/TheBloke/Mistral-7B-v0.1-GPTQ) for the GPTQ model. We also benchmarked it against `bitsandbytes` quantization methods and native `float16` model. Some results are shown below:
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/quantization/forward_memory_plot.png">
-</div>
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/quantization/generate_memory_plot.png">
-</div>
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/quantization/generate_throughput_plot.png">
-</div>
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/quantization/forward_latency_plot.png">
-</div>
-
-You can find the full results together with packages versions in [this link](https://github.com/huggingface/optimum-benchmark/tree/main/examples/running-mistral).
-
-From the results it appears that AWQ quantization method is the fastest quantization method for inference, text generation and among the lowest peak memory for text generation. However, AWQ seems to have the largest forward latency per batch size. 
-
-### Google colab demo
-
-Check out how to use this integration throughout this [Google Colab demo](https://colab.research.google.com/drive/1HzZH89yAXJaZgwJDhQj9LqSBux932BvY)!
-
-### AwqConfig
-
-[[autodoc]] AwqConfig
-
 ## `AutoGPTQ` Integration
 
 🤗 Transformers has integrated `optimum` API to perform GPTQ quantization on language models. You can load and quantize your model in 8, 4, 3 or even 2 bits without a big drop of performance and faster inference speed! This is supported by most GPU hardwares.
 
-To learn more about the quantization model, check out: 
+To learn more about the the quantization model, check out: 
 - the [GPTQ](https://arxiv.org/pdf/2210.17323.pdf) paper
 - the `optimum` [guide](https://huggingface.co/docs/optimum/llm_quantization/usage_guides/quantization) on GPTQ quantization
 - the [`AutoGPTQ`](https://github.com/PanQiWei/AutoGPTQ) library used as the backend
@@ -139,7 +48,6 @@ Note that GPTQ integration supports for now only text models and you may encount
 GPTQ is a quantization method that requires weights calibration before using the quantized models. If you want to quantize transformers model from scratch, it might take some time before producing the quantized model (~5 min on a Google colab for `facebook/opt-350m` model). 
 
 Hence, there are two different scenarios where you want to use GPTQ-quantized models. The first use case would be to load models that has been already quantized by other users that are available on the Hub, the second use case would be to quantize your model from scratch and save it or push it on the Hub so that other users can also use it.
-
 #### GPTQ Configuration
 
 In order to load and quantize a model, you need to create a [`GPTQConfig`]. You need to pass the number of `bits`, a `dataset` in order to calibrate the quantization and the `tokenizer` of the model in order prepare the dataset.
@@ -151,7 +59,6 @@ gptq_config = GPTQConfig(bits=4, dataset = "c4", tokenizer=tokenizer)
 ```
 
 Note that you can pass your own dataset as a list of string. However, it is highly recommended to use the dataset from the GPTQ paper. 
-
 ```python
 dataset = ["auto-gptq is an easy-to-use model quantization library with user-friendly apis, based on GPTQ algorithm."]
 quantization = GPTQConfig(bits=4, dataset = dataset, tokenizer=tokenizer)
@@ -164,17 +71,14 @@ You can quantize a model by using `from_pretrained` and setting the `quantizatio
 ```python
 from transformers import AutoModelForCausalLM
 model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=gptq_config)
-
 ```
 Note that you will need a GPU to quantize a model. We will put the model in the cpu and move the modules back and forth to the gpu in order to quantize them.
 
 If you want to maximize your gpus usage while using cpu offload, you can set `device_map = "auto"`.
-
 ```python
 from transformers import AutoModelForCausalLM
 model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", quantization_config=gptq_config)
 ```
-
 Note that disk offload is not supported. Furthermore, if you are out of memory because of the dataset, you may have to pass `max_memory` in `from_pretained`. Checkout this [guide](https://huggingface.co/docs/accelerate/usage_guides/big_modeling#designing-a-device-map) to learn more about `device_map` and `max_memory`.
 
 <Tip warning={true}>
@@ -191,14 +95,12 @@ tokenizer.push_to_hub("opt-125m-gptq")
 ```
 
 If you want to save your quantized model on your local machine, you can also do it with `save_pretrained`: 
-
 ```python
 quantized_model.save_pretrained("opt-125m-gptq")
 tokenizer.save_pretrained("opt-125m-gptq")
 ```
 
-Note that if you have quantized your model with a `device_map`, make sure to move the entire model to one of your gpus or the `cpu` before saving it.
-
+Note that if you have quantized your model with a `device_map`, make sure to move the entire model to one of your gpus or the `cpu` before saving it. 
 ```python
 quantized_model.to("cpu")
 quantized_model.save_pretrained("opt-125m-gptq")
@@ -215,7 +117,6 @@ model = AutoModelForCausalLM.from_pretrained("{your_username}/opt-125m-gptq")
 ```
 
 If you want to load a model faster and without allocating more memory than needed, the `device_map` argument also works with quantized model. Make sure that you have `accelerate` library installed.
-
 ```python
 from transformers import AutoModelForCausalLM
 model = AutoModelForCausalLM.from_pretrained("{your_username}/opt-125m-gptq", device_map="auto")
@@ -223,25 +124,16 @@ model = AutoModelForCausalLM.from_pretrained("{your_username}/opt-125m-gptq", de
 
 ### Exllama kernels for faster inference
 
-For 4-bit model, you can use the exllama kernels in order to a faster inference speed. It is activated by default. You can change that behavior by passing `use_exllama` in [`GPTQConfig`]. This will overwrite the quantization config stored in the config. Note that you will only be able to overwrite the attributes related to the kernels. Furthermore, you need to have the entire model on gpus if you want to use exllama kernels. Also, you can perform CPU inference using Auto-GPTQ for Auto-GPTQ version > 0.4.2 by passing `device_map` = "cpu". For CPU inference, you have to pass `use_exllama = False` in the `GPTQConfig.`
+For 4-bit model, you can use the exllama kernels in order to a faster inference speed. It is activated by default. You can change that behavior by passing `disable_exllama` in [`GPTQConfig`]. This will overwrite the quantization config stored in the config. Note that you will only be able to overwrite the attributes related to the kernels. Furthermore, you need to have the entire model on gpus if you want to use exllama kernels. 
 
 ```py
 import torch
-gptq_config = GPTQConfig(bits=4)
-model = AutoModelForCausalLM.from_pretrained("{your_username}/opt-125m-gptq", device_map="auto", quantization_config=gptq_config)
-```
-
-With the release of the exllamav2 kernels, you can get faster inference speed compared to the exllama kernels. You just need to pass `exllama_config={"version": 2}` in [`GPTQConfig`]:
-
-```py
-import torch
-gptq_config = GPTQConfig(bits=4, exllama_config={"version":2})
+gptq_config = GPTQConfig(bits=4, disable_exllama=False)
 model = AutoModelForCausalLM.from_pretrained("{your_username}/opt-125m-gptq", device_map="auto", quantization_config = gptq_config)
 ```
 
 Note that only 4-bit models are supported for now. Furthermore, it is recommended to deactivate the exllama kernels if you are finetuning a quantized model with peft. 
 
-You can find the benchmark of these kernels [here](https://github.com/huggingface/optimum/tree/main/tests/benchmark#gptq-benchmark)
 #### Fine-tune a quantized model 
 
 With the official support of adapters in the Hugging Face ecosystem, you can fine-tune models that have been quantized with GPTQ. 
@@ -444,7 +336,6 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 
 model = AutoModelForCausalLM.from_pretrained("{your_username}/bloom-560m-8bit", device_map="auto")
 ```
-
 Note that in this case, you don't need to specify the arguments `load_in_8bit=True`, but you need to make sure that `bitsandbytes` and `accelerate` are installed.
 Note also that `device_map` is optional but setting `device_map = 'auto'` is prefered for inference as it will dispatch efficiently the model on the available ressources.
 
@@ -465,7 +356,6 @@ quantization_config = BitsAndBytesConfig(llm_int8_enable_fp32_cpu_offload=True)
 ```
 
 Let's say you want to load `bigscience/bloom-1b7` model, and you have just enough GPU RAM to fit the entire model except the `lm_head`. Therefore write a custom device_map as follows:
-
 ```python
 device_map = {
     "transformer.word_embeddings": 0,

docs/source/en/main_classes/trainer.md

@@ -18,12 +18,6 @@ rendered properly in your Markdown viewer.
 
 The [`Trainer`] class provides an API for feature-complete training in PyTorch for most standard use cases. It's used in most of the [example scripts](https://github.com/huggingface/transformers/tree/main/examples).
 
-<Tip>
-
-If you're looking to fine-tune a language model like Llama-2 or Mistral on a text dataset using autoregressive techniques, consider using [`trl`](https://github.com/huggingface/trl)'s [`~trl.SFTTrainer`]. The [`~trl.SFTTrainer`] wraps the [`Trainer`] and is specially optimized for this particular task and supports sequence packing, LoRA, quantization, and DeepSpeed for efficient scaling to any model size. On the other hand, the [`Trainer`] is a more versatile option, suitable for a broader spectrum of tasks.
-
-</Tip>
-
 Before instantiating your [`Trainer`], create a [`TrainingArguments`] to access all the points of customization during training.
 
 The API supports distributed training on multiple GPUs/TPUs, mixed precision through [NVIDIA Apex](https://github.com/NVIDIA/apex) and Native AMP for PyTorch.
@@ -210,7 +204,6 @@ python -m torch.distributed.launch --nproc_per_node=2  trainer-program.py ...
 ```
 
 if you have either [`accelerate`](https://github.com/huggingface/accelerate) or [`deepspeed`](https://github.com/microsoft/DeepSpeed) installed you can also accomplish the same by using one of:
-
 ```bash
 accelerate launch --num_processes 2 trainer-program.py ...
 ```
@@ -247,7 +240,6 @@ CUDA_VISIBLE_DEVICES=2,0 python -m torch.distributed.launch trainer-program.py .
 Here your physical GPUs 0 and 2 are mapped to `cuda:1` and `cuda:0` correspondingly.
 
 The above examples were all for `DistributedDataParallel` use pattern, but the same method works for [`DataParallel`](https://pytorch.org/docs/stable/generated/torch.nn.DataParallel.html) as well:
-
 ```bash
 CUDA_VISIBLE_DEVICES=2,0 python trainer-program.py ...
 ```
@@ -740,27 +732,3 @@ Sections that were moved:
 | <a href="./deepspeed#deepspeed-grad-clip">Gradient Clipping</a><a id="gradient-clipping"></a>
 | <a href="./deepspeed#deepspeed-weight-extraction">Getting The Model Weights Out</a><a id="getting-the-model-weights-out"></a>
 ]
-
-## Boost your fine-tuning performances using NEFTune
-
-
-NEFTune is a technique to boost the performance of chat models and was introduced by the paper “NEFTune: Noisy Embeddings Improve Instruction Finetuning” from Jain et al. it consists of adding noise to the embedding vectors during training. According to the abstract of the paper:
-
-> Standard finetuning of LLaMA-2-7B using Alpaca achieves 29.79% on AlpacaEval, which rises to 64.69% using noisy embeddings. NEFTune also improves over strong baselines on modern instruction datasets. Models trained with Evol-Instruct see a 10% improvement, with ShareGPT an 8% improvement, and with OpenPlatypus an 8% improvement. Even powerful models further refined with RLHF such as LLaMA-2-Chat benefit from additional training with NEFTune.
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/trl-internal-testing/example-images/resolve/main/images/neft-screenshot.png">
-</div>
-
-To use it in `Trainer` simply pass `neftune_noise_alpha` when creating your `TrainingArguments` instance. Note that to avoid any surprising behaviour, NEFTune is disabled after training to retrieve back the original behaviour of the embedding layer.
-
-```python
-from transformers import Trainer, TrainingArguments
-
-args = TrainingArguments(..., neftune_noise_alpha=0.1)
-trainer = Trainer(..., args=args)
-
-...
-
-trainer.train()
-```

docs/source/en/model_doc/bark.md

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

docs/source/en/model_doc/clip.md

@@ -83,23 +83,9 @@ This model was contributed by [valhalla](https://huggingface.co/valhalla). The o
 
 A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with CLIP.
 
-- [Fine tuning CLIP with Remote Sensing (Satellite) images and captions](https://huggingface.co/blog/fine-tune-clip-rsicd), a blog post about how to fine-tune CLIP with [RSICD dataset](https://github.com/201528014227051/RSICD_optimal) and comparison of performance changes due to data augmentation.
-- This [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/contrastive-image-text) shows how to train a CLIP-like vision-text dual encoder model using a pre-trained vision and text encoder using [COCO dataset](https://cocodataset.org/#home).
-
-<PipelineTag pipeline="image-to-text"/>
-
-- A [notebook](https://colab.research.google.com/drive/1tuoAC5F4sC7qid56Z0ap-stR3rwdk0ZV?usp=sharing) on how to use a pretrained CLIP for inference with beam search for image captioning. 🌎
-
-**Image retrieval**
-
-- A [notebook](https://colab.research.google.com/drive/1bLVwVKpAndpEDHqjzxVPr_9nGrSbuOQd?usp=sharing) on image retrieval using pretrained CLIP and computing MRR(Mean Reciprocal Rank) score. 🌎
-- A [notebook](https://colab.research.google.com/github/deep-diver/image_search_with_natural_language/blob/main/notebooks/Image_Search_CLIP.ipynb) on image retrieval and showing the similarity score. 🌎
-- A [notebook](https://colab.research.google.com/drive/1xO-wC_m_GNzgjIBQ4a4znvQkvDoZJvH4?usp=sharing) on how to map images and texts to the same vector space using Multilingual CLIP. 🌎 
-- A [notebook](https://colab.research.google.com/github/vivien000/clip-demo/blob/master/clip.ipynb#scrollTo=uzdFhRGqiWkR) on how to run CLIP on semantic image search using [Unsplash](https://unsplash.com) and [TMBD](https://www.themoviedb.org/) datasets. 🌎
-
-**Explainability**
-
-- A [notebook](https://colab.research.google.com/github/hila-chefer/Transformer-MM-Explainability/blob/main/CLIP_explainability.ipynb) on how to visualize similarity between input token and image segment. 🌎
+- A blog post on [How to fine-tune CLIP on 10,000 image-text pairs](https://huggingface.co/blog/fine-tune-clip-rsicd).
+- CLIP is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/contrastive-image-text).
+- A [notebook](https://colab.research.google.com/drive/1zip3zmrbuKerAfC1d2uS1mqQS-QykXnl?usp=sharing) on how to fine-tune the CLIP model with Korean multimodal dataset. 🌎🇰🇷
 
 If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we will review it.
 The resource should ideally demonstrate something new instead of duplicating an existing resource.

docs/source/en/model_doc/convnextv2.md

@@ -58,15 +58,4 @@ If you're interested in submitting a resource to be included here, please feel f
 ## ConvNextV2ForImageClassification
 
 [[autodoc]] ConvNextV2ForImageClassification
-    - forward
-
-## TFConvNextV2Model
-
-[[autodoc]] TFConvNextV2Model
-    - call
-
-
-## TFConvNextV2ForImageClassification
-
-[[autodoc]] TFConvNextV2ForImageClassification
-    - call
+    - forward

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

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

docs/source/en/model_doc/fuyu.md

@@ -1,115 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Fuyu
-
-## Overview
-
-The Fuyu model was created by [ADEPT](https://www.adept.ai/blog/fuyu-8b), and authored by Rohan Bavishi, Erich Elsen, Curtis Hawthorne, Maxwell Nye, Augustus Odena, Arushi Somani, Sağnak Taşırlar. 
-
-The authors introduced Fuyu-8B, a decoder-only multimodal model based on the classic transformers architecture, with query and key normalization. A linear encoder is added to create multimodal embeddings from image inputs. 
-
-By treating image tokens like text tokens and using a special image-newline character, the model knows when an image line ends. Image positional embeddings are removed. This avoids the need for different training phases for various image resolutions. With 8 billion parameters and licensed under CC-BY-NC, Fuyu-8B is notable for its ability to handle both text and images, its impressive context size of 16K, and its overall performance.
-
-<Tip warning={true}>
-
-The `Fuyu` models were trained using `bfloat16`, but the original inference uses `float16` The checkpoints uploaded on the hub use `torch_dtype = 'float16'` which will be
-used by the `AutoModel` API to cast the checkpoints from `torch.float32` to `torch.float16`. 
-
-The `dtype` of the online weights is mostly irrelevant, unless you are using `torch_dtype="auto"` when initializing a model using `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`. The reason is that the model will first be downloaded ( using the `dtype` of the checkpoints online) then it will be cast to the default `dtype` of `torch` (becomes `torch.float32`). Users should specify the `torch_dtype` they want, and if they don't it will be `torch.float32`.
-
-Finetuning the model in `float16` is not recommended and known to produce `nan`, as such the model should be fine-tuned in `bfloat16`.
-
-</Tip>
-
-
-Tips:
-
-- To convert the model, you need to clone the original repository using `git clone https://github.com/persimmon-ai-labs/adept-inference`, then get the checkpoints:
-
-```bash
-git clone https://github.com/persimmon-ai-labs/adept-inference
-wget path/to/fuyu-8b-model-weights.tar
-tar -xvf fuyu-8b-model-weights.tar
-python src/transformers/models/fuyu/convert_fuyu_weights_to_hf.py  --input_dir /path/to/downloaded/fuyu/weights/ --output_dir /output/path \
-    --pt_model_path /path/to/fuyu_8b_release/iter_0001251/mp_rank_00/model_optim_rng.pt
-    --ada_lib_path /path/to/adept-inference
-```
-
-For the chat model:
-```bash
-wget https://axtkn4xl5cip.objectstorage.us-phoenix-1.oci.customer-oci.com/n/axtkn4xl5cip/b/adept-public-data/o/8b_chat_model_release.tar
-tar -xvf 8b_base_model_release.tar
-```
-Then, model can be loaded via:
-
-```py 
-from transformers import FuyuConfig, FuyuForCausalLM
-model_config = FuyuConfig()
-model = FuyuForCausalLM(model_config).from_pretrained('/output/path')
-```
-
-Inputs need to be passed through a specific Processor to have the correct formats.
-A processor requires an image_processor and a tokenizer. Hence, inputs can be loaded via:
-
-```py
-from PIL import Image
-from transformers import AutoTokenizer
-from transformers.models.fuyu.processing_fuyu import FuyuProcessor
-from transformers.models.fuyu.image_processing_fuyu import FuyuImageProcessor
-
-
-tokenizer = AutoTokenizer.from_pretrained('adept-hf-collab/fuyu-8b')
-image_processor = FuyuImageProcessor()
-
-
-processor = FuyuProcessor(image_processor=image_processor, tokenizer=tokenizer)
-text_prompt = "Generate a coco-style caption.\\n"
-
-bus_image_url = "https://huggingface.co/datasets/hf-internal-testing/fixtures-captioning/resolve/main/bus.png"
-bus_image_pil = Image.open(io.BytesIO(requests.get(bus_image_url).content))
-inputs_to_model = processor(text=text_prompt, images=image_pil)
-
-
-```
-
-This model was contributed by [Molbap](https://huggingface.co/Molbap).
-The original code can be found [here](https://github.com/persimmon-ai-labs/adept-inference).
-
-- Fuyu uses a `sentencepiece` based tokenizer, with a `Unigram` model. It supports bytefallback, which is only available in `tokenizers==0.14.0` for the fast tokenizer.
-The `LlamaTokenizer` is used as it is a standard wrapper around sentencepiece. 
-
-- The authors suggest to use the following prompt for image captioning: `f"Generate a coco-style caption.\\n"`
-
-
-## FuyuConfig
-
-[[autodoc]] FuyuConfig
-
-## FuyuForCausalLM
-
-[[autodoc]] FuyuForCausalLM
-    - forward
-
-## FuyuImageProcessor
-
-[[autodoc]] FuyuImageProcessor
-    - __call__
-
-## FuyuProcessor
-
-[[autodoc]] FuyuProcessor
-    - __call__

docs/source/en/model_doc/gpt_bigcode.md

@@ -42,45 +42,6 @@ The main differences compared to GPT2.
 
 You can read more about the optimizations in the [original pull request](https://github.com/huggingface/transformers/pull/22575)
 
-## Combining Starcoder and Flash Attention 2
-
-First, make sure to install the latest version of Flash Attention 2 to include the sliding window attention feature.
-
-```bash
-pip install -U flash-attn --no-build-isolation
-```
-
-Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of flash-attn repository. Make also sure to load your model in half-precision (e.g. `torch.float16``)
-
-To load and run a model using Flash Attention 2, refer to the snippet below:
-
-```python
->>> import torch
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
->>> device = "cuda" # the device to load the model onto
-
->>> model = AutoModelForCausalLM.from_pretrained("bigcode/gpt_bigcode-santacoder", torch_dtype=torch.float16, use_flash_attention_2=True)
->>> tokenizer = AutoTokenizer.from_pretrained("bigcode/gpt_bigcode-santacoder")
-
->>> prompt = "def hello_world():"
-
->>> model_inputs = tokenizer([prompt], return_tensors="pt").to(device)
->>> model.to(device)
-
->>> generated_ids = model.generate(**model_inputs, max_new_tokens=30, do_sample=False)
->>> tokenizer.batch_decode(generated_ids)[0]
-'def hello_world():\n    print("hello world")\n\nif __name__ == "__main__":\n    print("hello world")\n<|endoftext|>'
-```
-
-### Expected speedups
-
-Below is a expected speedup diagram that compares pure inference time between the native implementation in transformers using `bigcode/starcoder` checkpoint and the Flash Attention 2 version of the model using two different sequence lengths.
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/starcoder-speedup.png">
-</div>
-
-
 ## GPTBigCodeConfig
 
 [[autodoc]] GPTBigCodeConfig

docs/source/en/model_doc/jukebox.md

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

docs/source/en/model_doc/kosmos-2.md

@@ -1,98 +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.
-
--->
-
-# KOSMOS-2
-
-## Overview
-
-The KOSMOS-2 model was proposed in [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://arxiv.org/abs/2306.14824) by Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, Furu Wei.
-
-KOSMOS-2 is a Transformer-based causal language model and is trained using the next-word prediction task on a web-scale
-dataset of grounded image-text pairs [GRIT](https://huggingface.co/datasets/zzliang/GRIT). The spatial coordinates of
-the bounding boxes in the dataset are converted to a sequence of location tokens, which are appended to their respective
-entity text spans (for example, `a snowman` followed by `<patch_index_0044><patch_index_0863>`). The data format is
-similar to “hyperlinks” that connect the object regions in an image to their text span in the corresponding caption.
-
-The abstract from the paper is the following:
-
-*We introduce Kosmos-2, a Multimodal Large Language Model (MLLM), enabling new capabilities of perceiving object descriptions (e.g., bounding boxes) and grounding text to the visual world. Specifically, we represent refer expressions as links in Markdown, i.e., ``[text span](bounding boxes)'', where object descriptions are sequences of location tokens. Together with multimodal corpora, we construct large-scale data of grounded image-text pairs (called GrIT) to train the model. In addition to the existing capabilities of MLLMs (e.g., perceiving general modalities, following instructions, and performing in-context learning), Kosmos-2 integrates the grounding capability into downstream applications. We evaluate Kosmos-2 on a wide range of tasks, including (i) multimodal grounding, such as referring expression comprehension, and phrase grounding, (ii) multimodal referring, such as referring expression generation, (iii) perception-language tasks, and (iv) language understanding and generation. This work lays out the foundation for the development of Embodiment AI and sheds light on the big convergence of language, multimodal perception, action, and world modeling, which is a key step toward artificial general intelligence. Code and pretrained models are available at https://aka.ms/kosmos-2.*
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/kosmos_2_overview.jpg"
-alt="drawing" width="600"/>
-
-<small> Overview of tasks that KOSMOS-2 can handle. Taken from the <a href="https://arxiv.org/abs/2306.14824">original paper</a>. </small>
-
-## Example
-
-```python
->>> from PIL import Image
->>> import requests
->>> from transformers import AutoProcessor, Kosmos2ForConditionalGeneration
-
->>> model = Kosmos2ForConditionalGeneration.from_pretrained("microsoft/kosmos-2-patch14-224")
->>> processor = AutoProcessor.from_pretrained("microsoft/kosmos-2-patch14-224")
-
->>> url = "https://huggingface.co/microsoft/kosmos-2-patch14-224/resolve/main/snowman.jpg"
->>> image = Image.open(requests.get(url, stream=True).raw)
-
->>> prompt = "<grounding> An image of"
-
->>> inputs = processor(text=prompt, images=image, return_tensors="pt")
-
->>> generated_ids = model.generate(
-...     pixel_values=inputs["pixel_values"],
-...     input_ids=inputs["input_ids"],
-...     attention_mask=inputs["attention_mask"],
-...     image_embeds=None,
-...     image_embeds_position_mask=inputs["image_embeds_position_mask"],
-...     use_cache=True,
-...     max_new_tokens=64,
-... )
->>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
->>> processed_text = processor.post_process_generation(generated_text, cleanup_and_extract=False)
->>> processed_text
-'<grounding> An image of<phrase> a snowman</phrase><object><patch_index_0044><patch_index_0863></object> warming himself by<phrase> a fire</phrase><object><patch_index_0005><patch_index_0911></object>.'
-
->>> caption, entities = processor.post_process_generation(generated_text)
->>> caption
-'An image of a snowman warming himself by a fire.'
-
->>> entities
-[('a snowman', (12, 21), [(0.390625, 0.046875, 0.984375, 0.828125)]), ('a fire', (41, 47), [(0.171875, 0.015625, 0.484375, 0.890625)])]
-```
-
-This model was contributed by [Yih-Dar SHIEH](https://huggingface.co/ydshieh). The original code can be found [here](https://github.com/microsoft/unilm/tree/master/kosmos-2).
-
-## Kosmos2Config
-
-[[autodoc]] Kosmos2Config
-
-## Kosmos2ImageProcessor
-
-## Kosmos2Processor
-
-[[autodoc]] Kosmos2Processor
-    - __call__
-
-## Kosmos2Model
-
-[[autodoc]] Kosmos2Model
-    - forward
-
-## Kosmos2ForConditionalGeneration
-
-[[autodoc]] Kosmos2ForConditionalGeneration
-    - forward

docs/source/en/model_doc/llama2.md

@@ -28,12 +28,12 @@ Checkout all Llama2 models [here](https://huggingface.co/models?search=llama2)
 
 <Tip warning={true}>
 
-The `Llama2` models were trained using `bfloat16`, but the original inference uses `float16`. The checkpoints uploaded on the Hub use `torch_dtype = 'float16'`, which will be
+The `Llama2` models were trained using `bfloat16`, but the original inference uses `float16. The checkpoints uploaded on the hub use `torch_dtype = 'float16'` which will be
 used by the `AutoModel` API to cast the checkpoints from `torch.float32` to `torch.float16`. 
 
-The `dtype` of the online weights is mostly irrelevant unless you are using `torch_dtype="auto"` when initializing a model using `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`. The reason is that the model will first be downloaded ( using the `dtype` of the checkpoints online), then it will be casted to the default `dtype` of `torch` (becomes `torch.float32`), and finally, if there is a `torch_dtype` provided in the config, it will be used. 
+The `dtype` of the online weights is mostly irrelevant, unless you are using `torch_dtype="auto"` when initializing a model using `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`. The reason is that the model will first be downloaded ( using the `dtype` of the checkpoints online) then it will be casted to the default `dtype` of `torch` (becomes `torch.float32`) and finally, if there is a `torch_dtype` provided in the config, it will be used. 
 
-Training the model in `float16` is not recommended and is known to produce `nan`; as such, the model should be trained in `bfloat16`.
+Training the model in `float16` is not recommended and known to produce `nan`, as such the model should be trained in `bfloat16`.
 
 </Tip>
 

docs/source/en/model_doc/mistral.md

@@ -55,7 +55,7 @@ These ready-to-use checkpoints can be downloaded and used via the HuggingFace Hu
 
 >>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
 >>> tokenizer.batch_decode(generated_ids)[0]
-"The expected output"
+"The expected outupt"
 ```
 
 Raw weights for `Mistral-7B-v0.1` and `Mistral-7B-Instruct-v0.1` can be downloaded from:
@@ -76,7 +76,7 @@ python src/transformers/models/mistral/convert_mistral_weights_to_hf.py \
 You can then load the converted model from the `output/path`:
 
 ```python
-from transformers import MistralForCausalLM, LlamaTokenizer
+from transformers import MistralForCausalLM, LlamaTokenzier
 
 tokenizer = LlamaTokenizer.from_pretrained("/output/path")
 model = MistralForCausalLM.from_pretrained("/output/path")
@@ -109,7 +109,7 @@ To load and run a model using Flash Attention 2, refer to the snippet below:
 
 >>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
 >>> tokenizer.batch_decode(generated_ids)[0]
-"The expected output"
+"The expected outupt"
 ```
 
 ### Expected speedups

docs/source/en/model_doc/mra.md

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

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

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

docs/source/en/model_doc/open-llama.md

@@ -33,13 +33,15 @@ This model differs from the [OpenLLaMA models](https://huggingface.co/models?sea
 
 ## Overview
 
-The Open-Llama model was proposed in the open source Open-Llama project by community developer s-JoL.
+The Open-Llama model was proposed in [Open-Llama project](https://github.com/s-JoL/Open-Llama) by community developer s-JoL.
 
 The model is mainly based on LLaMA with some modifications, incorporating memory-efficient attention from Xformers, stable embedding from Bloom, and shared input-output embedding from PaLM.
 And the model is pre-trained on both Chinese and English, which gives it better performance on Chinese language tasks.
 
 This model was contributed by [s-JoL](https://huggingface.co/s-JoL).
-The original code was released on GitHub by [s-JoL](https://github.com/s-JoL), but is now removed.
+The original code can be found [Open-Llama](https://github.com/s-JoL/Open-Llama).
+Checkpoint and usage can be found at [s-JoL/Open-Llama-V1](https://huggingface.co/s-JoL/Open-Llama-V1).
+
 
 ## OpenLlamaConfig
 

docs/source/en/model_doc/owlv2.md

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

docs/source/en/model_doc/owlvit.md

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

docs/source/en/model_doc/seamless_m4t.md

@@ -1,218 +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.
--->
-
-# SeamlessM4T
-
-## Overview
-
-The SeamlessM4T model was proposed in [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) by the Seamless Communication team from Meta AI.
-
-SeamlessM4T is a collection of models designed to provide high quality translation, allowing people from different linguistic communities to communicate effortlessly through speech and text.
-
-SeamlessM4T enables multiple tasks without relying on separate models:
-
-- Speech-to-speech translation (S2ST)
-- Speech-to-text translation (S2TT)
-- Text-to-speech translation (T2ST)
-- Text-to-text translation (T2TT)
-- Automatic speech recognition (ASR)
-
-[`SeamlessM4TModel`] can perform all the above tasks, but each task also has its own dedicated sub-model.
-
-The abstract from the paper is the following:
-
-*What does it take to create the Babel Fish, a tool that can help individuals translate speech between any two languages? While recent breakthroughs in text-based models have pushed machine translation coverage beyond 200 languages, unified speech-to-speech translation models have yet to achieve similar strides. More specifically, conventional speech-to-speech translation systems rely on cascaded systems that perform translation progressively, putting high-performing unified systems out of reach. To address these gaps, we introduce SeamlessM4T, a single model that supports speech-to-speech translation, speech-to-text translation, text-to-speech translation, text-to-text translation, and automatic speech recognition for up to 100 languages. To build this, we used 1 million hours of open speech audio data to learn self-supervised speech representations with w2v-BERT 2.0. Subsequently, we created a multimodal corpus of automatically aligned speech translations. Filtered and combined with human-labeled and pseudo-labeled data, we developed the first multilingual system capable of translating from and into English for both speech and text. On FLEURS, SeamlessM4T sets a new standard for translations into multiple target languages, achieving an improvement of 20% BLEU over the previous SOTA in direct speech-to-text translation. Compared to strong cascaded models, SeamlessM4T improves the quality of into-English translation by 1.3 BLEU points in speech-to-text and by 2.6 ASR-BLEU points in speech-to-speech. Tested for robustness, our system performs better against background noises and speaker variations in speech-to-text tasks compared to the current SOTA model. Critically, we evaluated SeamlessM4T on gender bias and added toxicity to assess translation safety. Finally, all contributions in this work are open-sourced and accessible at https://github.com/facebookresearch/seamless_communication*
-
-## Usage
-
-First, load the processor and a checkpoint of the model:
-
-```python
->>> from transformers import AutoProcessor, SeamlessM4TModel
-
->>> processor = AutoProcessor.from_pretrained("facebook/hf-seamless-m4t-medium")
->>> model = SeamlessM4TModel.from_pretrained("facebook/hf-seamless-m4t-medium")
-```
-
-You can seamlessly use this model on text or on audio, to generated either translated text or translated audio.
-
-Here is how to use the processor to process text and audio:
-
-```python
->>> # let's load an audio sample from an Arabic speech corpus
->>> from datasets import load_dataset
->>> dataset = load_dataset("arabic_speech_corpus", split="test", streaming=True)
->>> audio_sample = next(iter(dataset))["audio"]
-
->>> # now, process it
->>> audio_inputs = processor(audios=audio_sample["array"], return_tensors="pt")
-
->>> # now, process some English test as well
->>> text_inputs = processor(text = "Hello, my dog is cute", src_lang="eng", return_tensors="pt")
-```
-
-
-### Speech
-
-[`SeamlessM4TModel`] can *seamlessly* generate text or speech with few or no changes. Let's target Russian voice translation:
-
-```python
->>> audio_array_from_text = model.generate(**text_inputs, tgt_lang="rus")[0].cpu().numpy().squeeze()
->>> audio_array_from_audio = model.generate(**audio_inputs, tgt_lang="rus")[0].cpu().numpy().squeeze()
-```
-
-With basically the same code, I've translated English text and Arabic speech to Russian speech samples.
-
-### Text
-
-Similarly, you can generate translated text from audio files or from text with the same model. You only have to pass `generate_speech=False` to [`SeamlessM4TModel.generate`].
-This time, let's translate to French.
-
-```python 
->>> # from audio
->>> output_tokens = model.generate(**audio_inputs, tgt_lang="fra", generate_speech=False)
->>> translated_text_from_audio = processor.decode(output_tokens[0].tolist()[0], skip_special_tokens=True)
-
->>> # from text
->>> output_tokens = model.generate(**text_inputs, tgt_lang="fra", generate_speech=False)
->>> translated_text_from_text = processor.decode(output_tokens[0].tolist()[0], skip_special_tokens=True)
-```
-
-### Tips
-
-
-#### 1. Use dedicated models
-
-[`SeamlessM4TModel`] is transformers top level model to generate speech and text, but you can also use dedicated models that perform the task without additional components, thus reducing the memory footprint.
-For example, you can replace the audio-to-audio generation snippet with the model dedicated to the S2ST task, the rest is exactly the same code: 
-
-```python
->>> from transformers import SeamlessM4TForSpeechToSpeech
->>> model = SeamlessM4TForSpeechToSpeech.from_pretrained("facebook/hf-seamless-m4t-medium")
-```
-
-Or you can replace the text-to-text generation snippet with the model dedicated to the T2TT task, you only have to remove `generate_speech=False`.
-
-```python
->>> from transformers import SeamlessM4TForTextToText
->>> model = SeamlessM4TForTextToText.from_pretrained("facebook/hf-seamless-m4t-medium")
-```
-
-Feel free to try out [`SeamlessM4TForSpeechToText`] and [`SeamlessM4TForTextToSpeech`] as well.
-
-#### 2. Change the speaker identity
-
-You have the possibility to change the speaker used for speech synthesis with the `spkr_id` argument. Some `spkr_id` works better than other for some languages!
-
-#### 3. Change the generation strategy
-
-You can use different [generation strategies](./generation_strategies) for speech and text generation, e.g `.generate(input_ids=input_ids, text_num_beams=4, speech_do_sample=True)` which will successively perform beam-search decoding on the text model, and multinomial sampling on the speech model.
-
-#### 4. Generate speech and text at the same time
-
-Use `return_intermediate_token_ids=True` with [`SeamlessM4TModel`] to return both speech and text !
-
-## Model architecture
-
-
-SeamlessM4T features a versatile architecture that smoothly handles the sequential generation of text and speech. This setup comprises two sequence-to-sequence (seq2seq) models. The first model translates the input modality into translated text, while the second model generates speech tokens, known as "unit tokens," from the translated text.
-
-Each modality has its own dedicated encoder with a unique architecture. Additionally, for speech output, a vocoder inspired by the [HiFi-GAN](https://arxiv.org/abs/2010.05646) architecture is placed on top of the second seq2seq model.
-
-Here's how the generation process works:
-
-- Input text or speech is processed through its specific encoder.
-- A decoder creates text tokens in the desired language.
-- If speech generation is required, the second seq2seq model, following a standard encoder-decoder structure, generates unit tokens.
-- These unit tokens are then passed through the final vocoder to produce the actual speech.
-
-
-This model was contributed by [ylacombe](https://huggingface.co/ylacombe). The original code can be found [here](https://github.com/facebookresearch/seamless_communication).
-
-## SeamlessM4TModel
-
-[[autodoc]] SeamlessM4TModel
-    - generate
-
-
-## SeamlessM4TForTextToSpeech
-
-[[autodoc]] SeamlessM4TForTextToSpeech
-    - generate
-
-
-## SeamlessM4TForSpeechToSpeech
-
-[[autodoc]] SeamlessM4TForSpeechToSpeech
-    - generate
-
-
-## SeamlessM4TForTextToText
-
-[[autodoc]] transformers.SeamlessM4TForTextToText
-    - forward
-    - generate
-
-## SeamlessM4TForSpeechToText
-
-[[autodoc]] transformers.SeamlessM4TForSpeechToText
-    - forward
-    - generate
-
-## SeamlessM4TConfig
-
-[[autodoc]] SeamlessM4TConfig
-
-
-## SeamlessM4TTokenizer
-
-[[autodoc]] SeamlessM4TTokenizer
-    - __call__
-    - build_inputs_with_special_tokens
-    - get_special_tokens_mask
-    - create_token_type_ids_from_sequences
-    - save_vocabulary
-
-
-## SeamlessM4TTokenizerFast
-
-[[autodoc]] SeamlessM4TTokenizerFast
-    - __call__
-
-## SeamlessM4TFeatureExtractor
-
-[[autodoc]] SeamlessM4TFeatureExtractor
-    - __call__
-
-## SeamlessM4TProcessor
-
-[[autodoc]] SeamlessM4TProcessor
-    - __call__
-
-## SeamlessM4TCodeHifiGan
-
-[[autodoc]] SeamlessM4TCodeHifiGan
-
-
-## SeamlessM4THifiGan
-
-[[autodoc]] SeamlessM4THifiGan
-
-## SeamlessM4TTextToUnitModel
-
-[[autodoc]] SeamlessM4TTextToUnitModel
-
-## SeamlessM4TTextToUnitForConditionalGeneration
-
-[[autodoc]] SeamlessM4TTextToUnitForConditionalGeneration
-
-

docs/source/en/model_doc/whisper.md

@@ -48,8 +48,6 @@ The original code can be found [here](https://github.com/openai/whisper).
     - get_special_tokens_mask
     - create_token_type_ids_from_sequences
     - save_vocabulary
-    - batch_decode
-    - decode
 
 ## WhisperTokenizerFast
 
@@ -59,8 +57,6 @@ The original code can be found [here](https://github.com/openai/whisper).
     - get_special_tokens_mask
     - create_token_type_ids_from_sequences
     - save_vocabulary
-    - batch_decode
-    - decode
 
 ## WhisperFeatureExtractor
 
@@ -86,12 +82,6 @@ The original code can be found [here](https://github.com/openai/whisper).
 
 [[autodoc]] WhisperForConditionalGeneration
     - forward
-    - generate
-
-## WhisperForCausalLM
-
-[[autodoc]] WhisperForCausalLM
-    - forward
 
 ## WhisperForAudioClassification
 

docs/source/en/perf_infer_cpu.md

@@ -13,80 +13,46 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# CPU inference
+# Efficient Inference on CPU
 
-With some optimizations, it is possible to efficiently run large model inference on a CPU. One of these optimization techniques involves compiling the PyTorch code into an intermediate format for high-performance environments like C++. The other technique fuses multiple operations into one kernel to reduce the overhead of running each operation separately.
+This guide focuses on inferencing large models efficiently on CPU.
 
-You'll learn how to use [BetterTransformer](https://pytorch.org/blog/a-better-transformer-for-fast-transformer-encoder-inference/) for faster inference, and how to convert your PyTorch code to [TorchScript](https://pytorch.org/tutorials/beginner/Intro_to_TorchScript_tutorial.html). If you're using an Intel CPU, you can also use [graph optimizations](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/tutorials/features.html#graph-optimization) from [Intel Extension for PyTorch](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/index.html) to boost inference speed even more. Finally, learn how to use 🤗 Optimum to accelerate inference with ONNX Runtime or OpenVINO (if you're using an Intel CPU).
+## `BetterTransformer` for faster inference
 
-## BetterTransformer
+We have recently integrated `BetterTransformer` for faster inference on CPU for text, image and audio models. Check the documentation about this integration [here](https://huggingface.co/docs/optimum/bettertransformer/overview) for more details.
 
-BetterTransformer accelerates inference with its fastpath (native PyTorch specialized implementation of Transformer functions) execution. The two optimizations in the fastpath execution are:
+## PyTorch JIT-mode (TorchScript)
+TorchScript is a way to create serializable and optimizable models from PyTorch code. Any TorchScript program can be saved from a Python process and loaded in a process where there is no Python dependency.
+Comparing to default eager mode, jit mode in PyTorch normally yields better performance for model inference from optimization methodologies like operator fusion.
 
-1. fusion, which combines multiple sequential operations into a single "kernel" to reduce the number of computation steps
-2. skipping the inherent sparsity of padding tokens to avoid unnecessary computation with nested tensors
+For a gentle introduction to TorchScript, see the Introduction to [PyTorch TorchScript tutorial](https://pytorch.org/tutorials/beginner/Intro_to_TorchScript_tutorial.html#tracing-modules).
 
-BetterTransformer also converts all attention operations to use the more memory-efficient [scaled dot product attention](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention).
+### IPEX Graph Optimization with JIT-mode
+Intel® Extension for PyTorch provides further optimizations in jit mode for Transformers series models. It is highly recommended for users to take advantage of Intel® Extension for PyTorch with jit mode. Some frequently used operator patterns from Transformers models are already supported in Intel® Extension for PyTorch with jit mode fusions. Those fusion patterns like Multi-head-attention fusion, Concat Linear, Linear+Add, Linear+Gelu, Add+LayerNorm fusion and etc. are enabled and perform well. The benefit of the fusion is delivered to users in a transparent fashion. According to the analysis, ~70% of most popular NLP tasks in question-answering, text-classification, and token-classification can get performance benefits with these fusion patterns for both Float32 precision and BFloat16 Mixed precision.
 
-<Tip>
+Check more detailed information for [IPEX Graph Optimization](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/tutorials/features/graph_optimization.html).
 
-BetterTransformer is not supported for all models. Check this [list](https://huggingface.co/docs/optimum/bettertransformer/overview#supported-models) to see if a model supports BetterTransformer.
+#### IPEX installation:
 
-</Tip>
+IPEX release is following PyTorch, check the approaches for [IPEX installation](https://intel.github.io/intel-extension-for-pytorch/).
 
-Before you start, make sure you have 🤗 Optimum [installed](https://huggingface.co/docs/optimum/installation).
-
-Enable BetterTransformer with the [`PreTrainedModel.to_bettertransformer`] method:
-
-```py
-from transformers import AutoModelForCausalLM
-
-model = AutoModelForCausalLM.from_pretrained("bigcode/starcoder")
-model.to_bettertransformer()
-```
-
-## TorchScript
-
-TorchScript is an intermediate PyTorch model representation that can be run in production environments where performance is important. You can train a model in PyTorch and then export it to TorchScript to free the model from Python performance constraints. PyTorch [traces](https://pytorch.org/docs/stable/generated/torch.jit.trace.html) a model to return a [`ScriptFunction`] that is optimized with just-in-time compilation (JIT). Compared to the default eager mode, JIT mode in PyTorch typically yields better performance for inference using optimization techniques like operator fusion.
-
-For a gentle introduction to TorchScript, see the [Introduction to PyTorch TorchScript](https://pytorch.org/tutorials/beginner/Intro_to_TorchScript_tutorial.html) tutorial.
-
-With the [`Trainer`] class, you can enable JIT mode for CPU inference by setting the `--jit_mode_eval` flag:
-
-```bash
-python run_qa.py \
---model_name_or_path csarron/bert-base-uncased-squad-v1 \
---dataset_name squad \
---do_eval \
---max_seq_length 384 \
---doc_stride 128 \
---output_dir /tmp/ \
---no_cuda \
---jit_mode_eval
-```
+### Usage of JIT-mode
+To enable JIT-mode in Trainer for evaluaion or prediction, users should add `jit_mode_eval` in Trainer command arguments.
 
 <Tip warning={true}>
 
-For PyTorch >= 1.14.0, JIT-mode could benefit any model for prediction and evaluaion since the dict input is supported in `jit.trace`.
+for PyTorch >= 1.14.0. JIT-mode could benefit any models for prediction and evaluaion since dict input is supported in jit.trace
 
-For PyTorch < 1.14.0, JIT-mode could benefit a model if its forward parameter order matches the tuple input order in `jit.trace`, such as a question-answering model. If the forward parameter order does not match the tuple input order in `jit.trace`, like a text classification model, `jit.trace` will fail and we are capturing this with the exception here to make it fallback. Logging is used to notify users.
+for PyTorch < 1.14.0. JIT-mode could benefit models whose forward parameter order matches the tuple input order in jit.trace, like question-answering model
+In the case where the forward parameter order does not match the tuple input order in jit.trace, like text-classification models, jit.trace will fail and we are capturing this with the exception here to make it fallback. Logging is used to notify users.
 
 </Tip>
 
-## IPEX graph optimization
+Take an example of the use cases on [Transformers question-answering](https://github.com/huggingface/transformers/tree/main/examples/pytorch/question-answering)
 
-Intel® Extension for PyTorch (IPEX) provides further optimizations in JIT mode for Intel CPUs, and we recommend combining it with TorchScript for even faster performance. The IPEX [graph optimization](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/tutorials/features/graph_optimization.html) fuses operations like Multi-head attention, Concat Linear, Linear + Add, Linear + Gelu, Add + LayerNorm, and more.
 
-To take advantage of these graph optimizations, make sure you have IPEX [installed](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/tutorials/installation.html):
-
-```bash
-pip install intel_extension_for_pytorch
-```
-
-Set the `--use_ipex` and `--jit_mode_eval` flags in the [`Trainer`] class to enable JIT mode with the graph optimizations:
-
-```bash
-python run_qa.py \
+- Inference using jit mode on CPU:
+<pre>python run_qa.py \
 --model_name_or_path csarron/bert-base-uncased-squad-v1 \
 --dataset_name squad \
 --do_eval \
@@ -94,34 +60,16 @@ python run_qa.py \
 --doc_stride 128 \
 --output_dir /tmp/ \
 --no_cuda \
---use_ipex \
---jit_mode_eval
-```
+<b>--jit_mode_eval </b></pre> 
 
-## 🤗 Optimum
-
-<Tip>
-
-Learn more details about using ORT with 🤗 Optimum in the [Optimum Inference with ONNX Runtime](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/models) guide. This section only provides a brief and simple example.
-
-</Tip>
-
-ONNX Runtime (ORT) is a model accelerator that runs inference on CPUs by default. ORT is supported by 🤗 Optimum which can be used in 🤗 Transformers, without making too many changes to your code. You only need to replace the 🤗 Transformers `AutoClass` with its equivalent [`~optimum.onnxruntime.ORTModel`] for the task you're solving, and load a checkpoint in the ONNX format.
-
-For example, if you're running inference on a question answering task, load the [optimum/roberta-base-squad2](https://huggingface.co/optimum/roberta-base-squad2) checkpoint which contains a `model.onnx` file:
-
-```py
-from transformers import AutoTokenizer, pipeline
-from optimum.onnxruntime import ORTModelForQuestionAnswering
-
-model = ORTModelForQuestionAnswering.from_pretrained("optimum/roberta-base-squad2")
-tokenizer = AutoTokenizer.from_pretrained("deepset/roberta-base-squad2")
-
-onnx_qa = pipeline("question-answering", model=model, tokenizer=tokenizer)
-
-question = "What's my name?"
-context = "My name is Philipp and I live in Nuremberg."
-pred = onnx_qa(question, context)
-```
-
-If you have an Intel CPU, take a look at 🤗 [Optimum Intel](https://huggingface.co/docs/optimum/intel/index) which supports a variety of compression techniques (quantization, pruning, knowledge distillation) and tools for converting models to the [OpenVINO](https://huggingface.co/docs/optimum/intel/inference) format for higher performance inference.
+- Inference with IPEX using jit mode on CPU:
+<pre>python run_qa.py \
+--model_name_or_path csarron/bert-base-uncased-squad-v1 \
+--dataset_name squad \
+--do_eval \
+--max_seq_length 384 \
+--doc_stride 128 \
+--output_dir /tmp/ \
+--no_cuda \
+<b>--use_ipex \</b>
+<b>--jit_mode_eval</b></pre> 

docs/source/en/perf_infer_gpu_many.md

@@ -0,0 +1,124 @@
+<!--Copyright 2022 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Efficient Inference on a Multiple GPUs
+
+This document contains information on how to efficiently infer on a multiple GPUs. 
+<Tip>
+
+Note: A multi GPU setup can use the majority of the strategies described in the [single GPU section](./perf_infer_gpu_one). You must be aware of simple techniques, though, that can be used for a better usage.
+
+</Tip>
+
+## Flash Attention 2
+
+Flash Attention 2 integration also works in a multi-GPU setup, check out the appropriate section in the [single GPU section](./perf_infer_gpu_one#Flash-Attention-2)
+
+## BetterTransformer
+
+[BetterTransformer](https://huggingface.co/docs/optimum/bettertransformer/overview) converts 🤗 Transformers models to use the PyTorch-native fastpath execution, which calls optimized kernels like Flash Attention under the hood.  
+
+BetterTransformer is also supported for faster inference on single and multi-GPU for text, image, and audio models.
+
+<Tip>
+
+Flash Attention can only be used for models using fp16 or bf16 dtype. Make sure to cast your model to the appropriate dtype before using BetterTransformer.
+  
+</Tip>
+
+### Decoder models
+
+For text models, especially decoder-based models (GPT, T5, Llama, etc.), the BetterTransformer API converts all attention operations to use the [`torch.nn.functional.scaled_dot_product_attention` operator](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) (SDPA) that is only available in PyTorch 2.0 and onwards. 
+
+To convert a model to BetterTransformer:
+
+```python
+from transformers import AutoModelForCausalLM
+
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-350m")
+# convert the model to BetterTransformer
+model.to_bettertransformer()
+
+# Use it for training or inference
+```
+
+SDPA can also call [Flash Attention](https://arxiv.org/abs/2205.14135) kernels under the hood. To enable Flash Attention or to check that it is available in a given setting (hardware, problem size), use [`torch.backends.cuda.sdp_kernel`](https://pytorch.org/docs/master/backends.html#torch.backends.cuda.sdp_kernel) as a context manager:
+
+
+```diff
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+tokenizer = AutoTokenizer.from_pretrained("facebook/opt-350m")
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-350m").to("cuda")
+# convert the model to BetterTransformer
+model.to_bettertransformer()
+
+input_text = "Hello my dog is cute and"
+inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
+
++ with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
+    outputs = model.generate(**inputs)
+
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```
+
+If you see a bug with a traceback saying 
+
+```bash
+RuntimeError: No available kernel.  Aborting execution.
+```
+
+try using the PyTorch nightly version, which may have a broader coverage for Flash Attention:
+
+```bash
+pip3 install -U --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu118
+```
+
+Have a look at this [blog post](https://pytorch.org/blog/out-of-the-box-acceleration/) to learn more about what is possible with the BetterTransformer + SDPA API.
+
+### Encoder models
+
+For encoder models during inference, BetterTransformer dispatches the forward call of encoder layers to an equivalent of [`torch.nn.TransformerEncoderLayer`](https://pytorch.org/docs/stable/generated/torch.nn.TransformerEncoderLayer.html) that will execute the fastpath implementation of the encoder layers.
+
+Because `torch.nn.TransformerEncoderLayer` fastpath does not support training, it is dispatched to `torch.nn.functional.scaled_dot_product_attention` instead, which does not leverage nested tensors but can use Flash Attention or Memory-Efficient Attention fused kernels.
+
+More details about BetterTransformer performance can be found in this [blog post](https://medium.com/pytorch/bettertransformer-out-of-the-box-performance-for-huggingface-transformers-3fbe27d50ab2), and you can learn more about BetterTransformer for encoder models in this [blog](https://pytorch.org/blog/a-better-transformer-for-fast-transformer-encoder-inference/).
+
+
+## Advanced usage: mixing FP4 (or Int8) and BetterTransformer
+
+You can combine the different methods described above to get the best performance for your model. For example, you can use BetterTransformer with FP4 mixed-precision inference + flash attention:
+
+```py
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.float16
+)
+
+tokenizer = AutoTokenizer.from_pretrained("facebook/opt-350m")
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-350m", quantization_config=quantization_config)
+
+input_text = "Hello my dog is cute and"
+inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
+    outputs = model.generate(**inputs)
+
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```

docs/source/en/perf_infer_gpu_one.md

@@ -13,38 +13,39 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# GPU inference
+# Efficient Inference on a Single GPU
 
-GPUs are the standard choice of hardware for machine learning, unlike CPUs, because they are optimized for memory bandwidth and parallelism. To keep up with the larger sizes of modern models or to run these large models on existing and older hardware, there are several optimizations you can use to speed up GPU inference. In this guide, you'll learn how to use FlashAttention-2 (a more memory-efficient attention mechanism), BetterTransformer (a PyTorch native fastpath execution), and bitsandbytes to quantize your model to a lower precision. Finally, learn how to use 🤗 Optimum to accelerate inference with ONNX Runtime on Nvidia GPUs.
+In addition to this guide, relevant information can be found as well in [the guide for training on a single GPU](perf_train_gpu_one) and [the guide for inference on CPUs](perf_infer_cpu).
+
+## Flash Attention 2
 
 <Tip>
 
-The majority of the optimizations described here also apply to multi-GPU setups!
+Note that this feature is experimental and might considerably change in future versions. For instance, the Flash Attention 2 API might migrate to `BetterTransformer` API in the near future.
 
 </Tip>
 
-## FlashAttention-2
+Flash Attention 2 can considerably speed up transformer-based models' training and inference speed. Flash Attention 2 has been introduced in the [official Flash Attention repository](https://github.com/Dao-AILab/flash-attention) by Tri Dao et al. The scientific paper on Flash Attention can be found [here](https://arxiv.org/abs/2205.14135).
+
+Make sure to follow the installation guide on the repository mentioned above to properly install Flash Attention 2. Once that package is installed, you can benefit from this feature.
+
+We natively support Flash Attention 2 for the following models:
+
+- Llama
+- Mistral
+- Falcon
+
+You can request to add Flash Attention 2 support for more models by opening an issue on GitHub, and even open a Pull Request to integrate the changes. The supported models can be used for inference and training, including training with padding tokens - *which is currently not supported for `BetterTransformer` API below.*
 
 <Tip>
 
-FlashAttention-2 is experimental and may change considerably in future versions.
-
+Flash Attention 2 can only be used when the models' dtype is  `fp16` or `bf16` and runs only on NVIDIA-GPU devices. Make sure to cast your model to the appropriate dtype and load them on a supported device before using that feature.
+  
 </Tip>
 
-[FlashAttention-2](https://huggingface.co/papers/2205.14135) is a faster and more efficient implementation of the standard attention mechanism that can significantly speedup inference by:
+### Quick usage
 
-1. additionally parallelizing the attention computation over sequence length
-2. partitioning the work between GPU threads to reduce communication and shared memory reads/writes between them
-
-FlashAttention-2 supports inference with Llama, Mistral, and Falcon models. You can request to add FlashAttention-2 support for another model by opening a GitHub Issue or Pull Request.
-
-Before you begin, make sure you have FlashAttention-2 installed (see the [installation](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#installation-and-features) guide for more details about prerequisites):
-
-```bash
-pip install flash-attn --no-build-isolation
-```
-
-To enable FlashAttention-2, add the `use_flash_attention_2` parameter to [`~AutoModelForCausalLM.from_pretrained`]:
+To enable Flash Attention 2 in your model, add `use_flash_attention_2` in the `from_pretrained` arguments:
 
 ```python
 import torch
@@ -60,29 +61,74 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 ```
 
-<Tip>
+And use it for generation or fine-tuning.
 
-FlashAttention-2 can only be used when the model's dtype is  `fp16` or `bf16`, and it only runs on Nvidia GPUs. Make sure to cast your model to the appropriate dtype and load them on a supported device before using FlashAttention-2.
-  
-</Tip>
+### Expected speedups
 
-FlashAttention-2 can be combined with other optimization techniques like quantization to further speedup inference. For example, you can combine FlashAttention-2 with 8-bit or 4-bit quantization:
+You can benefit from considerable speedups for fine-tuning and inference, especially for long sequences. However, since Flash Attention does not support computing attention scores with padding tokens under the hood, we must manually pad / unpad the attention scores for batched inference when the sequence contains padding tokens. This leads to a significant slowdown for batched generations with padding tokens. 
 
-```py
+To overcome this, one should use Flash Attention without padding tokens in the sequence for training (e.g., by packing a dataset, i.e., concatenating sequences until reaching the maximum sequence length. An example is provided [here](https://github.com/huggingface/transformers/blob/main/examples/pytorch/language-modeling/run_clm.py#L516).
+
+Below is the expected speedup you can get for a simple forward pass on [tiiuae/falcon-7b](https://hf.co/tiiuae/falcon-7b) with a sequence length of 4096 and various batch sizes, without padding tokens:
+
+<div style="text-align: center">
+<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/falcon-7b-inference-large-seqlen.png">
+</div>
+
+Below is the expected speedup you can get for a simple forward pass on [`meta-llama/Llama-7b-hf`](https://hf.co/meta-llama/Llama-7b-hf) with a sequence length of 4096 and various batch sizes, without padding tokens:
+
+<div style="text-align: center">
+<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/llama-7b-inference-large-seqlen.png">
+</div>
+
+For sequences with padding tokens (training with padding tokens or generating with padding tokens), we need to unpad / pad the input sequences to compute correctly the attention scores. For relatively small sequence length, on pure forward pass, this creates an overhead leading to a small speedup (below 30% of the input has been filled with padding tokens). 
+
+<div style="text-align: center">
+<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/llama-2-small-seqlen-padding.png">
+</div>
+
+But for large sequence length you can benefit from interesting speedup for pure inference (also training)
+
+Note that Flash Attention makes the attention computation more memory efficient, meaning you can train with much larger sequence lengths without facing CUDA OOM issues. It can lead up to memory reduction up to 20 for large sequence length. Check out [the official flash attention repository](https://github.com/Dao-AILab/flash-attention) for more details.
+
+<div style="text-align: center">
+<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/llama-2-large-seqlen-padding.png">
+</div>
+
+
+### Advanced usage
+
+You can combine this feature with many exisiting feature for model optimization. Check out few examples below:
+
+### Combining Flash Attention 2 and 8-bit models
+
+You can combine this feature together with 8-bit quantization:
+
+```python
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, LlamaForCausalLM
 
 model_id = "tiiuae/falcon-7b"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 
-# load in 8bit
 model = AutoModelForCausalLM.from_pretrained(
     model_id, 
     load_in_8bit=True,
     use_flash_attention_2=True,
 )
+```
+
+### Combining Flash Attention 2 and 4-bit models
+
+You can combine this feature together with 4-bit quantization:
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, LlamaForCausalLM
+
+model_id = "tiiuae/falcon-7b"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
 
-# load in 4bit
 model = AutoModelForCausalLM.from_pretrained(
     model_id, 
     load_in_4bit=True,
@@ -90,77 +136,85 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 ```
 
-### Expected speedups
+### Combining Flash Attention 2 and PEFT
 
-You can benefit from considerable speedups for inference, especially for inputs with long sequences. However, since FlashAttention-2 does not support computing attention scores with padding tokens, you must manually pad/unpad the attention scores for batched inference when the sequence contains padding tokens. This leads to a significant slowdown for batched generations with padding tokens.
+You can combine this feature together with PEFT for training adapters using Flash Attention 2 under the hood:
 
-To overcome this, you should use FlashAttention-2 without padding tokens in the sequence during training (by packing a dataset or [concatenating sequences](https://github.com/huggingface/transformers/blob/main/examples/pytorch/language-modeling/run_clm.py#L516) until reaching the maximum sequence length).
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, LlamaForCausalLM
+from peft import LoraConfig
 
-For a single forward pass on [tiiuae/falcon-7b](https://hf.co/tiiuae/falcon-7b) with a sequence length of 4096 and various batch sizes without padding tokens, the expected speedup is:
+model_id = "tiiuae/falcon-7b"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
 
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/falcon-7b-inference-large-seqlen.png">
-</div>
+model = AutoModelForCausalLM.from_pretrained(
+    model_id, 
+    load_in_4bit=True,
+    use_flash_attention_2=True,
+)
 
-For a single forward pass on [meta-llama/Llama-7b-hf](https://hf.co/meta-llama/Llama-7b-hf) with a sequence length of 4096 and various batch sizes without padding tokens, the expected speedup is:
+lora_config = LoraConfig(
+    r=8,
+    task_type="CAUSAL_LM"
+)
 
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/llama-7b-inference-large-seqlen.png">
-</div>
+model.add_adapter(lora_config)
 
-For sequences with padding tokens (generating with padding tokens), you need to unpad/pad the input sequences to correctly compute the attention scores. With a relatively small sequence length, a single forward pass creates overhead leading to a small speedup (in the example below, 30% of the input is filled with padding tokens):
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/llama-2-small-seqlen-padding.png">
-</div>
-
-But for larger sequence lengths, you can expect even more speedup benefits:
-
-<Tip>
-
-FlashAttention is more memory efficient, meaning you can train on much larger sequence lengths without running into out-of-memory issues. You can potentially reduce memory usage up to 20x for larger sequence lengths. Take a look at the [flash-attention](https://github.com/Dao-AILab/flash-attention) repository for more details.
-
-</Tip>
-
-<div style="text-align: center">
-<img src="https://huggingface.co/datasets/ybelkada/documentation-images/resolve/main/llama-2-large-seqlen-padding.png">
-</div>
+... # train your model
+```
 
 ## BetterTransformer
 
+[BetterTransformer](https://huggingface.co/docs/optimum/bettertransformer/overview) converts 🤗 Transformers models to use the PyTorch-native fastpath execution, which calls optimized kernels like Flash Attention under the hood.  
+
+BetterTransformer is also supported for faster inference on single and multi-GPU for text, image, and audio models.
+
 <Tip>
 
-Check out our benchmarks with BetterTransformer and scaled dot product attention in the [Out of the box acceleration and memory savings of 🤗 decoder models with PyTorch 2.0](https://pytorch.org/blog/out-of-the-box-acceleration/) and learn more about the fastpath execution in the [BetterTransformer](https://medium.com/pytorch/bettertransformer-out-of-the-box-performance-for-huggingface-transformers-3fbe27d50ab2) blog post.
-
+Flash Attention can only be used for models using fp16 or bf16 dtype. Make sure to cast your model to the appropriate dtype before using BetterTransformer.
+  
 </Tip>
 
-BetterTransformer accelerates inference with its fastpath (native PyTorch specialized implementation of Transformer functions) execution. The two optimizations in the fastpath execution are:
+### Encoder models
 
-1. fusion, which combines multiple sequential operations into a single "kernel" to reduce the number of computation steps
-2. skipping the inherent sparsity of padding tokens to avoid unnecessary computation with nested tensors
+PyTorch-native [`nn.MultiHeadAttention`](https://pytorch.org/blog/a-better-transformer-for-fast-transformer-encoder-inference/) attention fastpath, called BetterTransformer, can be used with Transformers through the integration in the [🤗 Optimum library](https://huggingface.co/docs/optimum/bettertransformer/overview).
 
-BetterTransformer also converts all attention operations to use the more memory-efficient [scaled dot product attention (SDPA)](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention), and it calls optimized kernels like [FlashAttention](https://huggingface.co/papers/2205.14135) under the hood.
+PyTorch's attention fastpath allows to speed up inference through kernel fusions and the use of [nested tensors](https://pytorch.org/docs/stable/nested.html). Detailed benchmarks can be found in [this blog post](https://medium.com/pytorch/bettertransformer-out-of-the-box-performance-for-huggingface-transformers-3fbe27d50ab2).
 
-Before you start, make sure you have 🤗 Optimum [installed](https://huggingface.co/docs/optimum/installation).
-
-Then you can enable BetterTransformer with the [`PreTrainedModel.to_bettertransformer`] method:
+After installing the [`optimum`](https://github.com/huggingface/optimum) package, to use Better Transformer during inference, the relevant internal modules are replaced by calling [`~PreTrainedModel.to_bettertransformer`]:
 
 ```python
 model = model.to_bettertransformer()
 ```
 
-You can return the original Transformers model with the [`~PreTrainedModel.reverse_bettertransformer`] method. You should use this before saving your model to use the canonical Transformers modeling:
+The method [`~PreTrainedModel.reverse_bettertransformer`] allows to go back to the original modeling, which should be used before saving the model in order to use the canonical transformers modeling:
 
-```py
+```python
 model = model.reverse_bettertransformer()
 model.save_pretrained("saved_model")
 ```
 
-### FlashAttention
+Have a look at this [blog post](https://medium.com/pytorch/bettertransformer-out-of-the-box-performance-for-huggingface-transformers-3fbe27d50ab2) to learn more about what is possible to do with `BetterTransformer` API for encoder models.
 
-SDPA can also call FlashAttention kernels under the hood. FlashAttention can only be used for models using the `fp16` or `bf16` dtype, so make sure to cast your model to the appropriate dtype before using it.
+### Decoder models
+
+For text models, especially decoder-based models (GPT, T5, Llama, etc.), the BetterTransformer API converts all attention operations to use the [`torch.nn.functional.scaled_dot_product_attention` operator](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) (SDPA) that is only available in PyTorch 2.0 and onwards. 
+
+To convert a model to BetterTransformer:
+
+```python
+from transformers import AutoModelForCausalLM
+
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-350m")
+# convert the model to BetterTransformer
+model.to_bettertransformer()
+
+# Use it for training or inference
+```
+
+SDPA can also call [Flash Attention](https://arxiv.org/abs/2205.14135) kernels under the hood. To enable Flash Attention or to check that it is available in a given setting (hardware, problem size), use [`torch.backends.cuda.sdp_kernel`](https://pytorch.org/docs/master/backends.html#torch.backends.cuda.sdp_kernel) as a context manager:
 
-To enable FlashAttention or to check whether it is available in a given setting (hardware, problem size), use [`torch.backends.cuda.sdp_kernel`](https://pytorch.org/docs/master/backends.html#torch.backends.cuda.sdp_kernel) as a context manager:
 
 ```diff
 import torch
@@ -180,32 +234,47 @@ inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 
-If you see a bug with the traceback below, try using nightly version of PyTorch which may have broader coverage for FlashAttention:
+If you see a bug with a traceback saying 
 
 ```bash
-RuntimeError: No available kernel. Aborting execution.
+RuntimeError: No available kernel.  Aborting execution.
+```
 
-# install PyTorch nightly
+try using the PyTorch nightly version, which may have a broader coverage for Flash Attention:
+
+```bash
 pip3 install -U --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu118
 ```
 
-## bitsandbytes
+Or make sure your model is correctly casted in float16 or bfloat16
 
-bitsandbytes is a quantization library that includes support for 4-bit and 8-bit quantization. Quantization reduces your model size compared to its native full precision version, making it easier to fit large models onto GPUs with limited memory.
 
-Make sure you have bitsnbytes and 🤗 Accelerate installed:
+Have a look at [this detailed blogpost](https://pytorch.org/blog/out-of-the-box-acceleration/) to read more about what is possible to do with `BetterTransformer` + SDPA API.
 
-```bash
-# these versions support 8-bit and 4-bit
-pip install bitsandbytes>=0.39.0 accelerate>=0.20.0
+## `bitsandbytes` integration for FP4 mixed-precision inference
 
-# install Transformers
-pip install transformers
-```
+You can install `bitsandbytes` and benefit from easy model compression on GPUs. Using FP4 quantization you can expect to reduce up to 8x the model size compared to its native full precision version. Check out below how to get started.
 
-### 4-bit
+<Tip>
 
-To load a model in 4-bit for inference, use the `load_in_4bit` parameter. The `device_map` parameter is optional, but we recommend setting it to `"auto"` to allow 🤗 Accelerate to automatically and efficiently allocate the model given the available resources in the environment.
+Note that this feature can also be used in a multi GPU setup.
+
+</Tip>
+
+### Requirements [[requirements-for-fp4-mixedprecision-inference]]
+
+- Latest `bitsandbytes` library
+`pip install bitsandbytes>=0.39.0`
+
+- Install latest `accelerate` from source
+`pip install git+https://github.com/huggingface/accelerate.git`
+
+- Install latest `transformers` from source
+`pip install git+https://github.com/huggingface/transformers.git`
+
+### Running FP4 models - single GPU setup - Quickstart
+
+You can quickly run a FP4 model on a single GPU by running the following code:
 
 ```py
 from transformers import AutoModelForCausalLM
@@ -213,8 +282,16 @@ from transformers import AutoModelForCausalLM
 model_name = "bigscience/bloom-2b5"
 model_4bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_4bit=True)
 ```
+Note that `device_map` is optional but setting `device_map = 'auto'` is prefered for inference as it will dispatch efficiently the model on the available ressources.
 
-To load a model in 4-bit for inference with multiple GPUs, you can control how much GPU RAM you want to allocate to each GPU. For example, to distribute 600MB of memory to the first GPU and 1GB of memory to the second GPU:
+### Running FP4 models - multi GPU setup
+
+The way to load your mixed 4-bit model in multiple GPUs is as follows (same command as single GPU setup):
+```py
+model_name = "bigscience/bloom-2b5"
+model_4bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_4bit=True)
+```
+But you can control the GPU RAM you want to allocate on each GPU using `accelerate`. Use the `max_memory` argument as follows:
 
 ```py
 max_memory_mapping = {0: "600MB", 1: "1GB"}
@@ -223,16 +300,44 @@ model_4bit = AutoModelForCausalLM.from_pretrained(
     model_name, device_map="auto", load_in_4bit=True, max_memory=max_memory_mapping
 )
 ```
+In this example, the first GPU will use 600MB of memory and the second 1GB.
 
-### 8-bit
+### Advanced usage
+
+For more advanced usage of this method, please have a look at the [quantization](main_classes/quantization) documentation page.
+
+## `bitsandbytes` integration for Int8 mixed-precision matrix decomposition
 
 <Tip>
 
-If you're curious and interested in learning more about the concepts underlying 8-bit quantization, read the [Gentle Introduction to 8-bit Matrix Multiplication for transformers at scale using Hugging Face Transformers, Accelerate and bitsandbytes](https://huggingface.co/blog/hf-bitsandbytes-integration) blog post.
+Note that this feature can also be used in a multi GPU setup.
 
 </Tip>
 
-To load a model in 8-bit for inference, use the `load_in_8bit` parameter. The `device_map` parameter is optional, but we recommend setting it to `"auto"` to allow 🤗 Accelerate to automatically and efficiently allocate the model given the available resources in the environment:
+From the paper [`LLM.int8() : 8-bit Matrix Multiplication for Transformers at Scale`](https://arxiv.org/abs/2208.07339), we support Hugging Face integration for all models in the Hub with a few lines of code.
+The method reduces `nn.Linear` size by 2 for `float16` and `bfloat16` weights and by 4 for `float32` weights, with close to no impact to the quality by operating on the outliers in half-precision.
+
+![HFxbitsandbytes.png](https://cdn-uploads.huggingface.co/production/uploads/1659861207959-62441d1d9fdefb55a0b7d12c.png)
+
+Int8 mixed-precision matrix decomposition works by separating a matrix multiplication into two streams: (1) a systematic feature outlier stream matrix multiplied in fp16 (0.01%), (2) a regular stream of int8 matrix multiplication (99.9%). With this method, int8 inference with no predictive degradation is possible for very large models.
+For more details regarding the method, check out the [paper](https://arxiv.org/abs/2208.07339) or our [blogpost about the integration](https://huggingface.co/blog/hf-bitsandbytes-integration).
+
+![MixedInt8.gif](https://cdn-uploads.huggingface.co/production/uploads/1660567469965-62441d1d9fdefb55a0b7d12c.gif)
+
+Note, that you would require a GPU to run mixed-8bit models as the kernels have been compiled for GPUs only. Make sure that you have enough GPU memory to store the quarter (or half if your model weights are in half precision) of the model before using this feature.
+Below are some notes to help you use this module, or follow the demos on [Google colab](#colab-demos).
+
+### Requirements [[requirements-for-int8-mixedprecision-matrix-decomposition]]
+
+- If you have `bitsandbytes<0.37.0`, make sure you run on NVIDIA GPUs that support 8-bit tensor cores (Turing, Ampere or newer architectures - e.g. T4, RTX20s RTX30s, A40-A100). For `bitsandbytes>=0.37.0`, all GPUs should be supported.
+- Install the correct version of `bitsandbytes` by running:
+`pip install bitsandbytes>=0.31.5`
+- Install `accelerate`
+`pip install accelerate>=0.12.0`
+
+### Running mixed-Int8 models - single GPU setup
+
+After installing the required libraries, the way to load your mixed 8-bit model is as follows:
 
 ```py
 from transformers import AutoModelForCausalLM
@@ -241,7 +346,12 @@ model_name = "bigscience/bloom-2b5"
 model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
 ```
 
-If you're loading a model in 8-bit for text generation, you should use the [`~transformers.GenerationMixin.generate`] method instead of the [`Pipeline`] function which is not optimized for 8-bit models and will be slower. Some sampling strategies, like nucleus sampling, are also not supported by the [`Pipeline`] for 8-bit models. You should also place all inputs on the same device as the model:
+For text generation, we recommend:
+
+* using the model's `generate()` method instead of the `pipeline()` function. Although inference is possible with the `pipeline()` function, it is not optimized for mixed-8bit models, and will be slower than using the `generate()` method. Moreover, some sampling strategies are like nucleaus sampling are not supported by the `pipeline()` function for mixed-8bit models.
+* placing all inputs on the same device as the model.
+
+Here is a simple example:
 
 ```py
 from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -256,7 +366,15 @@ generated_ids = model.generate(**inputs)
 outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
 ```
 
-To load a model in 4-bit for inference with multiple GPUs, you can control how much GPU RAM you want to allocate to each GPU. For example, to distribute 1GB of memory to the first GPU and 2GB of memory to the second GPU:
+
+### Running mixed-int8 models - multi GPU setup
+
+The way to load your mixed 8-bit model in multiple GPUs is as follows (same command as single GPU setup):
+```py
+model_name = "bigscience/bloom-2b5"
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+```
+But you can control the GPU RAM you want to allocate on each GPU using `accelerate`. Use the `max_memory` argument as follows:
 
 ```py
 max_memory_mapping = {0: "1GB", 1: "2GB"}
@@ -265,56 +383,27 @@ model_8bit = AutoModelForCausalLM.from_pretrained(
     model_name, device_map="auto", load_in_8bit=True, max_memory=max_memory_mapping
 )
 ```
+In this example, the first GPU will use 1GB of memory and the second 2GB.
 
-<Tip>
+### Colab demos
 
-Feel free to try running a 11 billion parameter [T5 model](https://colab.research.google.com/drive/1YORPWx4okIHXnjW7MSAidXN29mPVNT7F?usp=sharing) or the 3 billion parameter [BLOOM model](https://colab.research.google.com/drive/1qOjXfQIAULfKvZqwCen8-MoWKGdSatZ4?usp=sharing) for inference on Google Colab's free tier GPUs!
+With this method you can infer on models that were not possible to infer on a Google Colab before.
+Check out the demo for running T5-11b (42GB in fp32)! Using 8-bit quantization on Google Colab:
 
-</Tip>
+[![Open In Colab: T5-11b demo](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1YORPWx4okIHXnjW7MSAidXN29mPVNT7F?usp=sharing)
 
-## 🤗 Optimum
+Or this demo for BLOOM-3B:
 
-<Tip>
+[![Open In Colab: BLOOM-3b demo](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1qOjXfQIAULfKvZqwCen8-MoWKGdSatZ4?usp=sharing)
 
-Learn more details about using ORT with 🤗 Optimum in the [Accelerated inference on NVIDIA GPUs](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#accelerated-inference-on-nvidia-gpus) guide. This section only provides a brief and simple example.
+## Advanced usage: mixing FP4 (or Int8) and BetterTransformer
 
-</Tip>
-
-ONNX Runtime (ORT) is a model accelerator that supports accelerated inference on Nvidia GPUs. ORT uses optimization techniques like fusing common operations into a single node and constant folding to reduce the number of computations performed and speedup inference. ORT also places the most computationally intensive operations on the GPU and the rest on the CPU to intelligently distribute the workload between the two devices.
-
-ORT is supported by 🤗 Optimum which can be used in 🤗 Transformers. You'll need to use an [`~optimum.onnxruntime.ORTModel`] for the task you're solving, and specify the `provider` parameter which can be set to either [`CUDAExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#cudaexecutionprovider) or [`TensorrtExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#tensorrtexecutionprovider). If you want to load a model that was not yet exported to ONNX, you can set `export=True` to convert your model on-the-fly to the ONNX format :
-
-```py
-from optimum.onnxruntime import ORTModelForSequenceClassification
-
-ort_model = ORTModelForSequenceClassification.from_pretrained(
-  "distilbert-base-uncased-finetuned-sst-2-english",
-  export=True,
-  provider="CUDAExecutionProvider",
-)
-```
-
-Now you're free to use the model for inference:
-
-```py
-from optimum.pipelines import pipeline
-from transformers import AutoTokenizer
-
-tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased-finetuned-sst-2-english")
-
-pipeline = pipeline(task="text-classification", model=ort_model, tokenizer=tokenizer, device="cuda:0")
-result = pipeline("Both the music and visual were astounding, not to mention the actors performance.")
-```
-
-## Combine optimizations
-
-It is often possible to combine several of the optimization techniques described above to get the best inference performance possible for your model. For example, you can load a model in 4-bit, and then enable BetterTransformer with FlashAttention:
+You can combine the different methods described above to get the best performance for your model. For example, you can use BetterTransformer with FP4 mixed-precision inference + flash attention:
 
 ```py
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
-# load model in 4-bit
 quantization_config = BitsAndBytesConfig(
     load_in_4bit=True,
     bnb_4bit_compute_dtype=torch.float16
@@ -323,13 +412,9 @@ quantization_config = BitsAndBytesConfig(
 tokenizer = AutoTokenizer.from_pretrained("facebook/opt-350m")
 model = AutoModelForCausalLM.from_pretrained("facebook/opt-350m", quantization_config=quantization_config)
 
-# enable BetterTransformer
-model = model.to_bettertransformer()
-
 input_text = "Hello my dog is cute and"
 inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
 
-# enable FlashAttention
 with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
     outputs = model.generate(**inputs)
 

docs/source/en/perf_infer_special.md

@@ -1,4 +1,4 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
+<!--Copyright 2022 The HuggingFace Team. All rights reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
 the License. You may obtain a copy of the License at
@@ -15,4 +15,4 @@ rendered properly in your Markdown viewer.
 
 # Inference on Specialized Hardware
 
-こちらのドキュメントは、専用のハードウェアでの推論方法についての情報がまもなく提供されます。その間に、CPUでの推論に関するガイドをご覧いただけます。[the guide for inference on CPUs](perf_infer_cpu).
+This document will be completed soon with information on how to infer on specialized hardware. In the meantime you can check out [the guide for inference on CPUs](perf_infer_cpu).

docs/source/en/perf_train_gpu_many.md

@@ -15,154 +15,143 @@ rendered properly in your Markdown viewer.
 
 # Efficient Training on Multiple GPUs
 
-If training a model on a single GPU is too slow or if the model's weights do not fit in a single GPU's memory, transitioning 
-to a multi-GPU setup may be a viable option. Prior to making this transition, thoroughly explore all the strategies covered 
-in the [Methods and tools for efficient training on a single GPU](perf_train_gpu_one) as they are universally applicable 
-to model training on any number of GPUs. Once you have employed those strategies and found them insufficient for your 
-case on a single GPU, consider moving to multiple GPUs.
-
-Transitioning from a single GPU to multiple GPUs requires the introduction of some form of parallelism, as the workload 
-must be distributed across the resources. Multiple techniques can be employed to achieve parallelism, such as data 
-parallelism, tensor parallelism, and pipeline parallelism. It's important to note that there isn't a one-size-fits-all 
-solution, and the optimal settings depend on the specific hardware configuration you are using. 
-
-This guide offers an in-depth overview of individual types of parallelism, as well as guidance on ways to combine   
-techniques and choosing an appropriate approach. For step-by-step tutorials on distributed training, please refer to
-the [🤗 Accelerate documentation](https://huggingface.co/docs/accelerate/index). 
+When training on a single GPU is too slow or the model weights don't fit in a single GPUs memory we use a multi-GPU setup. Switching from a single GPU to multiple requires some form of parallelism as the work needs to be distributed. There are several techniques to achieve parallism such as data, tensor, or pipeline parallism. However, there is no one solution to fit them all and which settings works best depends on the hardware you are running on. While the main concepts most likely will apply to any other framework, this article is focused on PyTorch-based implementations.
 
 <Tip>
 
-While the main concepts discussed in this guide are likely applicable across frameworks, here we focus on 
-PyTorch-based implementations.
+ Note: Most of the strategies introduced in the [single GPU section](perf_train_gpu_one) (such as mixed precision training or gradient accumulation) are generic and apply to training models in general so make sure to have a look at it before diving into the following sections such as multi-GPU or CPU training.
 
 </Tip>
 
-Before diving deeper into the specifics of each technique, let's go over the rough decision process when training 
-large models on a large infrastructure.
+We will first discuss in depth various 1D parallelism techniques and their pros and cons and then look at how they can be combined into 2D and 3D parallelism to enable an even faster training and to support even bigger models. Various other powerful alternative approaches will be presented.
 
-## Scalability strategy
+## Concepts
 
-Begin by estimating how much vRAM is required to train your model. For models hosted on the 🤗 Hub, use our 
-[Model Memory Calculator](https://huggingface.co/spaces/hf-accelerate/model-memory-usage), which gives you 
-accurate calculations within a few percent margin.  
+The following is the brief description of the main concepts that will be described later in depth in this document.
 
-**Parallelization strategy for a single Node / multi-GPU setup**
+1. **DataParallel (DP)** - the same setup is replicated multiple times, and each being fed a slice of the data. The processing is done in parallel and all setups are synchronized at the end of each training step.
+2. **TensorParallel (TP)** - each tensor is split up into multiple chunks, so instead of having the whole tensor reside on a single gpu, each shard of the tensor resides on its designated gpu. During processing each shard gets processed separately and in parallel on different GPUs and the results are synced at the end of the step. This is what one may call horizontal parallelism, as the splitting happens on horizontal level.
+3. **PipelineParallel (PP)** - the model is split up vertically (layer-level) across multiple GPUs, so that only one or several layers of the model are places on a single gpu. Each gpu processes in parallel different stages of the pipeline and working on a small chunk of the batch.
+4. **Zero Redundancy Optimizer (ZeRO)** - Also performs sharding of the tensors somewhat similar to TP, except the whole tensor gets reconstructed in time for a forward or backward computation, therefore the model doesn't need to be modified. It also supports various offloading techniques to compensate for limited GPU memory.
+5. **Sharded DDP** - is another name for the foundational ZeRO concept as used by various other implementations of ZeRO.
 
-When training a model on a single node with multiple GPUs, your choice of parallelization strategy can significantly 
-impact performance. Here's a breakdown of your options:
+Before diving deeper into the specifics of each concept we first have a look at the rough decision process when training large models on a large infrastructure.
 
-**Case 1: Your model fits onto a single GPU**
+## Scalability Strategy
 
-If your model can comfortably fit onto a single GPU, you have two primary options:
+**⇨ Single Node / Multi-GPU**
+* Model fits onto a single GPU:
 
-1. DDP - Distributed DataParallel
-2. ZeRO - depending on the situation and configuration used, this method may or may not be faster, however, it's worth experimenting with it.
+    1. DDP - Distributed DP
+    2. ZeRO - may or may not be faster depending on the situation and configuration used
 
-**Case 2: Your model doesn't fit onto a single GPU:**
+* Model doesn't fit onto a single GPU:
 
-If your model is too large for a single GPU, you have several alternatives to consider:
+    1. PP
+    2. ZeRO
+    3. TP
 
-1. PipelineParallel (PP)
-2. ZeRO
-3. TensorParallel (TP)
+    With very fast intra-node connectivity of NVLINK or NVSwitch all three should be mostly on par, without these PP will be faster than TP or ZeRO. The degree of TP may also make a difference. Best to experiment to find the winner on your particular setup.
 
-With very fast inter-node connectivity (e.g., NVLINK or NVSwitch) all three strategies (PP, ZeRO, TP) should result in 
-similar performance. However, without these, PP will be faster than TP or ZeRO. The degree of TP may also 
-make a difference. It's best to experiment with your specific setup to determine the most suitable strategy.
+    TP is almost always used within a single node. That is TP size <= gpus per node.
 
-TP is almost always used within a single node. That is TP size <= GPUs per node.
+* Largest Layer not fitting into a single GPU:
 
-**Case 3: Largest layer of your model does not fit onto a single GPU**
+    1. If not using ZeRO - must use TP, as PP alone won't be able to fit.
+    2. With ZeRO see the same entry for "Single GPU" above
 
-1. If you are not using ZeRO, you have to use TensorParallel (TP), because PipelineParallel (PP) alone won't be sufficient to accommodate the large layer.
-2. If you are using ZeRO, additionally adopt techniques from the [Methods and tools for efficient training on a single GPU](perf_train_gpu_one).
 
-**Parallelization strategy for a multi-Node / multi-GPU setup**
+**⇨ Multi-Node / Multi-GPU**
 
-* When you have fast inter-node connectivity (e.g., NVLINK or NVSwitch) consider using one of these options:
+* When you have fast inter-node connectivity:
 
     1. ZeRO - as it requires close to no modifications to the model
-    2. A combination of PipelineParallel(PP) with TensorParallel(TP) and DataParallel(DP) - this approach will result in fewer communications, but requires significant changes to the model
+    2. PP+TP+DP - less communications, but requires massive changes to the model
 
-* When you have slow inter-node connectivity and still low on GPU memory:
+* when you have slow inter-node connectivity and still low on GPU memory:
+
+    1. DP+PP+TP+ZeRO-1
 
-    1. Employ a combination of DataParallel(DP) with PipelineParallel(PP), TensorParallel(TP), and ZeRO.
 
-In the following sections of this guide we dig deeper into how these different parallelism methods work.
 
 ## Data Parallelism
 
-Even with only 2 GPUs, you can readily leverage the accelerated training capabilities offered by PyTorch's built-in features, 
-such as `DataParallel` (DP) and `DistributedDataParallel` (DDP). Note that 
-[PyTorch documentation](https://pytorch.org/docs/master/generated/torch.nn.DataParallel.html) recommends to prefer 
-`DistributedDataParallel` (DDP) over `DataParallel` (DP) for multi-GPU training as it works for all models.
-Let's take a look at how these two methods work and what makes them different.
+Most users with just 2 GPUs already enjoy the increased training speed up thanks to `DataParallel` (DP) and `DistributedDataParallel` (DDP) that are almost trivial to use. This is a built-in feature of Pytorch. Note that in general it is advised to use DDP as it is better maintained and works for all models while DP might fail for some models. [PyTorch documentation](https://pytorch.org/docs/master/generated/torch.nn.DataParallel.html) itself recommends the use of DDP.
 
-### DataParallel vs DistributedDataParallel
+### DP vs DDP
 
-To understand the key differences in inter-GPU communication overhead between the two methods, let's review the processes per batch:
+`DistributedDataParallel` (DDP) is typically faster than `DataParallel` (DP), but it is not always the case:
+* while DP is python threads-based, DDP is multiprocess-based - and as such it has no python threads limitations, such as GIL
+* on the other hand a slow inter-connectivity between the GPU cards could lead to an actual slower outcome with DDP
+
+Here are the main differences in the inter-GPU communication overhead between the two modes:
 
 [DDP](https://pytorch.org/docs/master/notes/ddp.html):
 
-- At the start time the main process replicates the model once from GPU 0 to the rest of GPUs
+- At the start time the main process replicates the model once from gpu 0 to the rest of gpus
 - Then for each batch:
-   1. Each GPU directly consumes its mini-batch of data.
-   2. During `backward`, once the local gradients are ready, they are averaged across all processes.
+   1. each gpu consumes each own mini-batch of data directly
+   2. during `backward`, once the local gradients are ready, they are then averaged across all processes
 
 [DP](https://pytorch.org/docs/master/generated/torch.nn.DataParallel.html):
 
 For each batch:
-   1. GPU 0 reads the batch of data and then sends a mini-batch to each GPU.
-   2. The up-to-date model is replicated from GPU 0 to each GPU. 
-   3. `forward` is executed, and output from each GPU is sent to GPU 0 to compute the loss.
-   4. The loss is distributed from GPU 0 to all GPUs, and `backward` is run. 
-   5. Gradients from each GPU are sent to GPU 0 and averaged. 
+   1. gpu 0 reads the batch of data and then sends a mini-batch to each gpu
+   2. replicates the up-to-date model from gpu 0 to each gpu
+   3. runs `forward` and sends output from each gpu to gpu 0, computes loss
+   4. scatters loss from gpu 0 to all gpus, runs `backward`
+   5. sends gradients from each gpu to gpu 0 and averages those
 
-Key differences include:
-1. DDP performs only a single communication per batch - sending gradients, while DP performs five different data exchanges per batch.
-DDP copies data using [torch.distributed](https://pytorch.org/docs/master/distributed.html), while DP copies data within 
-the process via Python threads (which introduces limitations associated with GIL). As a result, **`DistributedDataParallel` (DDP) is generally faster than `DataParallel` (DP)** unless you have slow GPU card inter-connectivity.
-2. Under DP, GPU 0 performs significantly more work than other GPUs, resulting in GPU under-utilization. 
-3. DDP supports distributed training across multiple machines, whereas DP does not.
+The only communication DDP performs per batch is sending gradients, whereas DP does 5 different data exchanges per batch.
 
-This is not an exhaustive list of differences between DP and DDP, however, other nuances are out of scope of this guide.
-You can get a deeper understanding of these methods by reading this [article](https://www.telesens.co/2019/04/04/distributed-data-parallel-training-using-pytorch-on-aws/).
+DP copies data within the process via python threads, whereas DDP copies data via [torch.distributed](https://pytorch.org/docs/master/distributed.html).
 
-Let's illustrate the differences between DP and DDP with an experiment. We'll benchmark the differences between DP and 
-DDP with an added context of NVLink presence:  
+Under DP gpu 0 performs a lot more work than the rest of the gpus, thus resulting in under-utilization of gpus.
 
-* Hardware: 2x TITAN RTX 24GB each + NVlink with 2 NVLinks (`NV2` in `nvidia-smi topo -m`).
-* Software: `pytorch-1.8-to-be` + `cuda-11.0` / `transformers==4.3.0.dev0`.
+You can use DDP across multiple machines, but this is not the case with DP.
 
-To disable the NVLink feature on one of the benchmarks, we use `NCCL_P2P_DISABLE=1`. 
+There are other differences between DP and DDP but they aren't relevant to this discussion.
 
-Here is the benchmarking code and outputs:
+If you want to go really deep into understanding these 2 modes, this [article](https://www.telesens.co/2019/04/04/distributed-data-parallel-training-using-pytorch-on-aws/) is highly recommended, as it has great diagrams, includes multiple benchmarks and profiler outputs on various hardware, explains all the nuances that you may need to know.
 
-**DP**
+Let's look at an actual benchmark:
+
+| Type   | NVlink | Time |
+| :----- | -----  | ---: |
+| 2:DP   | Y      | 110s |
+| 2:DDP  | Y      | 101s |
+| 2:DDP  | N      | 131s |
+
+
+Analysis:
+
+Here DP is ~10% slower than DDP w/ NVlink, but ~15% faster than DDP w/o NVlink
+
+The real difference will depend on how much data each GPU needs to sync with the others - the more there is to sync, the more a slow link will slow down the total runtime.
+
+Here is the full benchmark code and outputs:
+
+`NCCL_P2P_DISABLE=1` was used to disable the NVLink feature on the corresponding benchmark.
 
 ```
+
+# DP
 rm -r /tmp/test-clm; CUDA_VISIBLE_DEVICES=0,1 \
 python examples/pytorch/language-modeling/run_clm.py \
 --model_name_or_path gpt2 --dataset_name wikitext --dataset_config_name wikitext-2-raw-v1 \
 --do_train --output_dir /tmp/test-clm --per_device_train_batch_size 4 --max_steps 200
 
 {'train_runtime': 110.5948, 'train_samples_per_second': 1.808, 'epoch': 0.69}
-```
 
-**DDP w/ NVlink**
-
-```
+# DDP w/ NVlink
 rm -r /tmp/test-clm; CUDA_VISIBLE_DEVICES=0,1 \
 python -m torch.distributed.launch --nproc_per_node 2 examples/pytorch/language-modeling/run_clm.py \
 --model_name_or_path gpt2 --dataset_name wikitext --dataset_config_name wikitext-2-raw-v1 \
 --do_train --output_dir /tmp/test-clm --per_device_train_batch_size 4 --max_steps 200
 
 {'train_runtime': 101.9003, 'train_samples_per_second': 1.963, 'epoch': 0.69}
-```
 
-**DDP w/o NVlink**
-
-```
+# DDP w/o NVlink
 rm -r /tmp/test-clm; NCCL_P2P_DISABLE=1 CUDA_VISIBLE_DEVICES=0,1 \
 python -m torch.distributed.launch --nproc_per_node 2 examples/pytorch/language-modeling/run_clm.py \
 --model_name_or_path gpt2 --dataset_name wikitext --dataset_config_name wikitext-2-raw-v1 \
@@ -171,34 +160,17 @@ python -m torch.distributed.launch --nproc_per_node 2 examples/pytorch/language-
 {'train_runtime': 131.4367, 'train_samples_per_second': 1.522, 'epoch': 0.69}
 ```
 
-Here are the same benchmarking results gathered in a table for convenience:
-
-| Type   | NVlink | Time |
-| :----- | -----  | ---: |
-| 2:DP   | Y      | 110s |
-| 2:DDP  | Y      | 101s |
-| 2:DDP  | N      | 131s |
-
-As you can see, in this case DP is ~10% slower than DDP with NVlink, but ~15% faster than DDP without NVlink.
-The real difference will depend on how much data each GPU needs to sync with the others - the more there is to sync, 
-the more a slow link will impede the overall runtime.
+Hardware: 2x TITAN RTX 24GB each + NVlink with 2 NVLinks (`NV2` in `nvidia-smi topo -m`)
+Software: `pytorch-1.8-to-be` + `cuda-11.0` / `transformers==4.3.0.dev0`
 
 ## ZeRO Data Parallelism
 
-ZeRO-powered data parallelism (ZeRO-DP) is illustrated in the following diagram from this [blog post](https://www.microsoft.com/en-us/research/blog/zero-deepspeed-new-system-optimizations-enable-training-models-with-over-100-billion-parameters/).
+ZeRO-powered data parallelism (ZeRO-DP) is described on the following diagram from this [blog post](https://www.microsoft.com/en-us/research/blog/zero-deepspeed-new-system-optimizations-enable-training-models-with-over-100-billion-parameters/)
+![DeepSpeed-Image-1](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-zero.png)
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-zero.png" alt="DeepSpeed-Image-1"/>
- </div>
-
-While it may appear complex, it is a very similar concept to `DataParallel` (DP). The difference is that instead of 
-replicating the full model parameters, gradients and optimizer states, each GPU stores only a slice of it. Then, at 
-run-time when the full layer parameters are needed just for the given layer, all GPUs synchronize to give each other 
-parts that they miss.
-
-To illustrate this idea, consider a simple model with 3 layers (La, Lb, and Lc), where each layer has 3 parameters. 
-Layer La, for example, has weights a0, a1 and a2:
+It can be difficult to wrap one's head around it, but in reality the concept is quite simple. This is just the usual `DataParallel` (DP), except, instead of replicating the full model params, gradients and optimizer states, each GPU stores only a slice of it.  And then at run-time when the full layer params are needed just for the given layer, all GPUs synchronize to give each other parts that they miss - this is it.
 
+Consider this simple model with 3 layers, where each layer has 3 params:
 ```
 La | Lb | Lc
 ---|----|---
@@ -206,8 +178,9 @@ a0 | b0 | c0
 a1 | b1 | c1
 a2 | b2 | c2
 ```
+Layer La has weights a0, a1 and a2.
 
-If we have 3 GPUs, ZeRO-DP splits the model onto 3 GPUs like so:
+If we have 3 GPUs, the Sharded DDP (= Zero-DP) splits the model onto 3 GPUs like so:
 
 ```
 GPU0:
@@ -226,213 +199,165 @@ La | Lb | Lc
 a2 | b2 | c2
 ```
 
-In a way, this is the same horizontal slicing as tensor parallelism, as opposed to Vertical 
-slicing, where one puts whole layer-groups on different GPUs. Now let's see how this works: 
-
-Each of these GPUs will get the usual mini-batch as it works in DP:
+In a way this is the same horizontal slicing, as tensor parallelism, if you imagine the typical DNN diagram. Vertical slicing is where one puts whole layer-groups on different GPUs. But it's just the starting point.
 
+Now each of these GPUs will get the usual mini-batch as it works in DP:
 ```
 x0 => GPU0
 x1 => GPU1
 x2 => GPU2
 ```
 
-The inputs are passed without modifications as if they would be processed by the original model.
+The inputs are unmodified - they think they are going to be processed by the normal model.
 
-First, the inputs get to the layer `La`. What happens at this point?
+First, the inputs hit the layer La.
 
-On GPU0: the x0 mini-batch requires the a0, a1, a2 parameters to do its forward path through the layer, but the GPU0 has only a0. 
-It will get a1 from GPU1 and a2 from GPU2, bringing all the pieces of the model together.
+Let's focus just on GPU0: x0 needs a0, a1, a2 params to do its forward path, but GPU0 has only a0 - it gets sent a1 from GPU1 and a2 from GPU2, bringing all pieces of the model together.
 
-In parallel, GPU1 gets another mini-batch - x1. GPU1 has the a1 parameter, but needs a0 and a2, so it gets those from GPU0 and GPU2.
-Same happens to GPU2 that gets the mini-batch x2. It gets a0 and a1 from GPU0 and GPU1.
+In parallel, GPU1 gets mini-batch x1 and it only has a1, but needs a0 and a2 params, so it gets those from GPU0 and GPU2.
 
-This way each of the 3 GPUs gets the full tensors reconstructed and makes a forward pass with its own mini-batch.
-As soon as the calculation is done, the data that is no longer needed gets dropped - it's only used during the calculation. 
-The reconstruction is done efficiently via a pre-fetch.
+Same happens to GPU2 that gets input x2. It gets a0 and a1 from GPU0 and GPU1, and with its a2 it reconstructs the full tensor.
 
-Then the whole process is repeated for layer Lb, then Lc forward-wise, and then backward Lc -> Lb -> La.
+All 3 GPUs get the full tensors reconstructed and a forward happens.
 
-<Tip>
+As soon as the calculation is done, the data that is no longer needed gets dropped - it's only used during the calculation. The reconstruction is done efficiently via a pre-fetch.
 
-This mechanism is similar to an efficient group backpacking strategy: person A carries the tent, person B carries the stove,
-and person C carries the axe. Each night they all share what they have with others and get from others what they don't have, 
-and in the morning they pack up their allocated type of gear and continue on their way. This is what ZeRO DP/Sharded DDP is.
-Compare this strategy to the simple one where each person has to carry their own tent, stove and axe (similar to 
-DataParallel (DP and DDP) in PyTorch), which would be far more inefficient. 
+And the whole process is repeated for layer Lb, then Lc forward-wise, and then backward Lc -> Lb -> La.
 
-</Tip>
+To me this sounds like an efficient group backpacking weight distribution strategy:
+
+1. person A carries the tent
+2. person B carries the stove
+3. person C carries the axe
+
+Now each night they all share what they have with others and get from others what they don't have, and in the morning they pack up their allocated type of gear and continue on their way. This is Sharded DDP / Zero DP.
+
+Compare this strategy to the simple one where each person has to carry their own tent, stove and axe, which would be far more inefficient. This is DataParallel (DP and DDP) in Pytorch.
 
 While reading the literature on this topic you may encounter the following synonyms: Sharded, Partitioned.
-If you pay close attention the way ZeRO partitions the model's weights - it looks very similar to tensor parallelism 
-which will be discussed later. This is because it partitions/shards each layer's weights, unlike vertical model parallelism 
-which is discussed next.
+
+If you pay close attention the way ZeRO partitions the model's weights - it looks very similar to tensor parallelism which will be discussed later. This is because it partitions/shards each layer's weights, unlike vertical model parallelism which is discussed next.
 
 Implementations:
 
 - [DeepSpeed](https://www.deepspeed.ai/features/#the-zero-redundancy-optimizer) ZeRO-DP stages 1+2+3
-- [`Accelerate` integration](https://huggingface.co/docs/accelerate/en/usage_guides/deepspeed) 
 - [`transformers` integration](main_classes/trainer#trainer-integrations)
 
-## From Naive Model Parallelism to Pipeline Parallelism
+## Naive Model Parallelism (Vertical) and Pipeline Parallelism
 
-To explain Pipeline parallelism, we'll first look into Naive Model Parallelism (MP), also known as Vertical MP. This approach
-involves distributing groups of model layers across multiple GPUs by assigning specific layers to specific GPUs with `.to()`. 
-As data flows through these layers, it is moved to the same GPU as the layer, while the other layers remain untouched.
+Naive Model Parallelism (MP) is where one spreads groups of model layers across multiple GPUs. The mechanism is relatively simple - switch the desired layers `.to()` the desired devices and now whenever the data goes in and out those layers switch the data to the same device as the layer and leave the rest unmodified.
 
-We refer to this Model parallelism as "Vertical" because of how models are typically visualized. For example, the 
-following diagram shows an 8-layer model split vertically into two slices, placing layers 0-3 onto 
-GPU0 and 4-7 to GPU1:
+We refer to it as Vertical MP, because if you remember how most models are drawn, we slice the layers vertically. For example, if the following diagram shows an 8-layer model:
 
 ```
 ===================  ===================
 |  0 | 1 | 2 | 3  |  |  4 | 5 | 6 | 7  |
 ===================  ===================
-        GPU0                 GPU1
+        gpu0                 gpu1
 ```
+we just sliced it in 2 vertically, placing layers 0-3 onto GPU0 and 4-7 to GPU1.
 
-In this example, when data moves from layer 0 to 3, it's no different from regular forward pass. However, passing data 
-from layer 3 to 4 requires moving it from GPU0 to GPU1, introducing a communication overhead. If the participating 
-GPUs are on the same compute node (e.g. same physical machine) this copying is fast, but if the GPUs are distributed 
-across different compute nodes (e.g. multiple machines), the communication overhead could be substantially greater.
+Now while data travels from layer 0 to 1, 1 to 2 and 2 to 3 this is just the normal model. But when data needs to pass from layer 3 to layer 4 it needs to travel from GPU0 to GPU1 which introduces a communication overhead. If the participating GPUs are on the same compute node (e.g. same physical machine) this copying is pretty fast, but if the GPUs are located on different compute nodes (e.g. multiple machines) the communication overhead could be significantly larger.
 
-Following that, layers 4 to 7 work as they would in the original model. Upon completion of the 7th layer, there is often 
-a need to send the data back to layer 0 where the labels are (or alternatively send the labels to the last layer). Now the loss can be 
-computed and the optimizer can do its work.
+Then layers 4 to 5 to 6 to 7 are as a normal model would have and when the 7th layer completes we often need to send the data back to layer 0 where the labels are (or alternatively send the labels to the last layer). Now the loss can be computed and the optimizer can do its work.
 
-Naive Model Parallelism comes several shortcomings:
-- **All but one GPU are idle at any given moment**: if 4 GPUs are used, it's nearly identical to quadrupling the amount of memory of a single GPU, and ignoring the rest of the hardware. 
-- **Overhead in data transfer between devices**:  E.g. 4x 6GB cards will be able to accommodate the same size as 1x 24GB card using naive MP, but a single 24GB card will complete the training faster, because it doesn't have the data copying overhead. But, say, if you have 40GB cards and need to fit a 45GB model you can with 4x 40GB cards (but barely because of the gradient and optimizer states)
-- **Copying shared embeddings**: Shared embeddings may need to get copied back and forth between GPUs.
+Problems:
+- the main deficiency and why this one is called "naive" MP, is that all but one GPU is idle at any given moment. So if 4 GPUs are used, it's almost identical to quadrupling the amount of memory of a single GPU, and ignoring the rest of the hardware. Plus there is the overhead of copying the data between devices. So 4x 6GB cards will be able to accommodate the same size as 1x 24GB card using naive MP, except the latter will complete the training faster, since it doesn't have the data copying overhead. But, say, if you have 40GB cards and need to fit a 45GB model you can with 4x 40GB cards (but barely because of the gradient and optimizer states)
+- shared embeddings may need to get copied back and forth between GPUs.
 
-Now that you are familiar with how the naive approach to model parallelism works and its shortcomings, let's look at Pipeline Parallelism (PP).
-PP is almost identical to a naive MP, but it solves the GPU idling problem by chunking the incoming batch into micro-batches 
-and artificially creating a pipeline, which allows different GPUs to concurrently participate in the computation process.
+Pipeline Parallelism (PP) is almost identical to a naive MP, but it solves the GPU idling problem, by chunking the incoming batch into micro-batches and artificially creating a pipeline, which allows different GPUs to concurrently participate in the computation process.
 
-The following illustration from the [GPipe paper](https://ai.googleblog.com/2019/03/introducing-gpipe-open-source-library.html) 
-shows the naive MP on the top, and PP on the bottom:
+The following illustration from the [GPipe paper](https://ai.googleblog.com/2019/03/introducing-gpipe-open-source-library.html) shows the naive MP on the top, and PP on the bottom:
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-gpipe-bubble.png" alt="MP vs PP"/>
-</div>
+![mp-pp](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-gpipe-bubble.png)
 
-At the bottom of the diagram, you can observe that the Pipeline Parallelism (PP) approach minimizes the number of idle 
-GPU zones, referred to as 'bubbles'. Both parts of the diagram show a parallelism level of degree 4, meaning that 4 GPUs 
-are involved in the pipeline. You can see that there's a forward path of 4 pipe stages (F0, F1, F2 and F3) followed by 
-a backward path in reverse order (B3, B2, B1, and B0).
+It's easy to see from the bottom diagram how PP has less dead zones, where GPUs are idle. The idle parts are referred to as the "bubble".
 
-PP introduces a new hyperparameter to tune - `chunks`, which determines how many data chunks are sent in a sequence 
-through the same pipe stage. For example, in the bottom diagram you can see `chunks=4`. GPU0 performs the same 
-forward path on chunk 0, 1, 2 and 3 (F0,0, F0,1, F0,2, F0,3) and then it waits for other GPUs to do complete their work. 
-Only when the other GPUs begin to complete their work, GPU0 starts to work again doing the backward path for chunks 
-3, 2, 1 and 0 (B0,3, B0,2, B0,1, B0,0).
+Both parts of the diagram show a parallelism that is of degree 4. That is 4 GPUs are participating in the pipeline. So there is the forward path of 4 pipe stages F0, F1, F2 and F3 and then the return reverse order backward path of B3, B2, B1 and B0.
 
-Note that this is the same concept as gradient accumulation steps. PyTorch uses `chunks`, while DeepSpeed refers 
-to the same hyperparameter as gradient accumulation steps.
+PP introduces a new hyper-parameter to tune and it's `chunks` which defines how many chunks of data are sent in a sequence through the same pipe stage. For example, in the bottom diagram you can see that `chunks=4`. GPU0 performs the same forward path on chunk 0, 1, 2 and 3 (F0,0, F0,1, F0,2, F0,3) and then it waits for other GPUs to do their work and only when their work is starting to be complete, GPU0 starts to work again doing the backward path for chunks 3, 2, 1 and 0 (B0,3, B0,2, B0,1, B0,0).
 
-Because of the chunks, PP introduces the notion of micro-batches (MBS). DP splits the global data batch size into 
-mini-batches, so if you have a DP degree of 4, a global batch size of 1024 gets split up into 4 mini-batches of 
-256 each (1024/4). And if the number of `chunks` (or GAS) is 32 we end up with a micro-batch size of 8 (256/32). Each 
-Pipeline stage works with a single micro-batch at a time. To calculate the global batch size of the DP + PP setup, 
-use the formula: `mbs * chunks * dp_degree` (`8 * 32 * 4 = 1024`).
-With `chunks=1` you end up with the naive MP, which is inefficient. With a large `chunks` value you end up with 
-tiny micro-batch sizes which is also inefficient. For this reason, we encourage to experiment with the `chunks` value to 
-find the one that leads to the most efficient GPUs utilization.
+Note that conceptually this is the same concept as gradient accumulation steps (GAS). Pytorch uses `chunks`, whereas DeepSpeed refers to the same hyper-parameter as GAS.
 
-You may notice a bubble of "dead" time on the diagram that can't be parallelized because the last `forward` stage 
-has to wait for `backward` to complete the pipeline. The purpose of finding the best value for `chunks` is to enable a high 
-concurrent GPU utilization across all participating GPUs which translates to minimizing the size of the bubble.
+Because of the chunks, PP introduces the concept of micro-batches (MBS). DP splits the global data batch size into mini-batches, so if you have a DP degree of 4, a global batch size of 1024 gets split up into 4 mini-batches of 256 each (1024/4). And if the number of `chunks` (or GAS) is 32 we end up with a micro-batch size of 8 (256/32). Each Pipeline stage works with a single micro-batch at a time.
 
-Pipeline API solutions have been implemented in:
+To calculate the global batch size of the DP + PP setup we then do: `mbs*chunks*dp_degree` (`8*32*4=1024`).
+
+Let's go back to the diagram.
+
+With `chunks=1` you end up with the naive MP, which is very inefficient. With a very large `chunks` value you end up with tiny micro-batch sizes which could be not every efficient either. So one has to experiment to find the value that leads to the highest efficient utilization of the gpus.
+
+While the diagram shows that there is a bubble of "dead" time that can't be parallelized because the last `forward` stage has to wait for `backward` to complete the pipeline, the purpose of finding the best value for `chunks` is to enable a high concurrent GPU utilization across all participating GPUs which translates to minimizing the size of the bubble.
+
+There are 2 groups of solutions - the traditional Pipeline API and the more modern solutions that make things much easier for the end user.
+
+Traditional Pipeline API solutions:
 - PyTorch
 - DeepSpeed
 - Megatron-LM
 
-These come with some shortcomings:
-- They have to modify the model quite heavily, because Pipeline requires one to rewrite the normal flow of modules into a `nn.Sequential` sequence of the same, which may require changes to the design of the model.
-- Currently the Pipeline API is very restricted. If you had a bunch of Python variables being passed in the very first stage of the Pipeline, you will have to find a way around it. Currently, the pipeline interface requires either a single Tensor or a tuple of Tensors as the only input and output. These tensors must have a batch size as the very first dimension, since pipeline is going to chunk the mini batch into micro-batches. Possible improvements are being discussed here https://github.com/pytorch/pytorch/pull/50693
-- Conditional control flow at the level of pipe stages is not possible - e.g., Encoder-Decoder models like T5 require special workarounds to handle a conditional encoder stage.
-- They have to arrange each layer so that the output of one layer becomes an input to the other layer.
-
-More recent solutions include:
+Modern solutions:
 - Varuna
 - Sagemaker
 
-We have not experimented with Varuna and SageMaker but their papers report that they have overcome the list of problems 
-mentioned above and that they require smaller changes to the user's model.
+Problems with traditional Pipeline API solutions:
+- have to modify the model quite heavily, because Pipeline requires one to rewrite the normal flow of modules into a `nn.Sequential` sequence of the same, which may require changes to the design of the model.
+- currently the Pipeline API is very restricted. If you had a bunch of python variables being passed in the very first stage of the Pipeline, you will have to find a way around it. Currently, the pipeline interface requires either a single Tensor or a tuple of Tensors as the only input and output. These tensors must have a batch size as the very first dimension, since pipeline is going to chunk the mini batch into micro-batches. Possible improvements are being discussed here https://github.com/pytorch/pytorch/pull/50693
+- conditional control flow at the level of pipe stages is not possible - e.g., Encoder-Decoder models like T5 require special workarounds to handle a conditional encoder stage.
+- have to arrange each layer so that the output of one model becomes an input to the other model.
+
+We are yet to experiment with Varuna and SageMaker but their papers report that they have overcome the list of problems mentioned above and that they require much smaller changes to the user's model.
 
 Implementations:
-- [PyTorch](https://pytorch.org/docs/stable/pipeline.html) (initial support in pytorch-1.8, and progressively getting improved in 1.9 and more so in 1.10). Some [examples](https://github.com/pytorch/pytorch/blob/master/benchmarks/distributed/pipeline/pipe.py)
+- [Pytorch](https://pytorch.org/docs/stable/pipeline.html) (initial support in pytorch-1.8, and progressively getting improved in 1.9 and more so in 1.10). Some [examples](https://github.com/pytorch/pytorch/blob/master/benchmarks/distributed/pipeline/pipe.py)
 - [DeepSpeed](https://www.deepspeed.ai/tutorials/pipeline/)
 - [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) has an internal implementation - no API.
 - [Varuna](https://github.com/microsoft/varuna)
 - [SageMaker](https://arxiv.org/abs/2111.05972) - this is a proprietary solution that can only be used on AWS.
 - [OSLO](https://github.com/tunib-ai/oslo) - this is implemented based on the Hugging Face Transformers.
 
-🤗 Transformers status: as of this writing none of the models supports full-PP. GPT2 and T5 models have naive MP support. 
-The main obstacle is being unable to convert the models to `nn.Sequential` and have all the inputs to be Tensors. This 
-is because currently the models include many features that make the conversion very complicated, and will need to be removed to accomplish that.
-
-DeepSpeed and Megatron-LM integrations are available in [🤗 Accelerate](https://huggingface.co/docs/accelerate/main/en/usage_guides/deepspeed)
+🤗 Transformers status: as of this writing none of the models supports full-PP. GPT2 and T5 models have naive MP support. The main obstacle is being unable to convert the models to `nn.Sequential` and have all the inputs to be Tensors. This is because currently the models include many features that make the conversion very complicated, and will need to be removed to accomplish that.
 
 Other approaches:
 
 DeepSpeed, Varuna and SageMaker use the concept of an [Interleaved Pipeline](https://docs.aws.amazon.com/sagemaker/latest/dg/model-parallel-core-features.html)
+![interleaved-pipeline-execution](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-sagemaker-interleaved-pipeline.png)
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-sagemaker-interleaved-pipeline.png" alt="Interleaved pipeline execution"/>
-</div>
+Here the bubble (idle time) is further minimized by prioritizing backward passes.
 
-Here the bubble (idle time) is further minimized by prioritizing backward passes. Varuna further attempts to improve the 
-schedule by using simulations to discover the most efficient scheduling.
+Varuna further tries to improve the schedule by using simulations to discover the most efficient scheduling.
 
-OSLO has pipeline parallelism implementation based on the Transformers without `nn.Sequential` conversion.
+OSLO has pipeline parallelism implementation based on the Transformers without `nn.Sequential` converting.
 
 ## Tensor Parallelism
 
-In Tensor Parallelism, each GPU processes a slice of a tensor and only aggregates the full tensor for operations requiring it.
-To describe this method, this section of the guide relies on the concepts and diagrams from the [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) 
-paper: [Efficient Large-Scale Language Model Training on GPU Clusters](https://arxiv.org/abs/2104.04473).
+In Tensor Parallelism each GPU processes only a slice of a tensor and only aggregates the full tensor for operations that require the whole thing.
+
+In this section we use concepts and diagrams from the [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) paper: [Efficient Large-Scale Language Model Training on GPU Clusters](https://arxiv.org/abs/2104.04473).
 
 The main building block of any transformer is a fully connected `nn.Linear` followed by a nonlinear activation `GeLU`.
-The dot dot-product part of it, following the Megatron's paper notation, can be written as `Y = GeLU(XA)`, where `X` is 
-an input vector, `Y` is the output vector, and `A` is the weight matrix.
 
-If we look at the computation in matrix form, you can see how the matrix multiplication can be split between multiple GPUs:
+Following the Megatron's paper notation, we can write the dot-product part of it as `Y = GeLU(XA)`, where `X` and `Y` are the input and output vectors, and `A` is the weight matrix.
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-parallel_gemm.png" alt="Parallel GEMM"/>
-</div>
+If we look at the computation in matrix form, it's easy to see how the matrix multiplication can be split between multiple GPUs:
+![Parallel GEMM](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-parallel_gemm.png)
 
-If we split the weight matrix `A` column-wise across `N` GPUs and perform matrix multiplications `XA_1` through `XA_n` in parallel, 
-then we will end up with `N` output vectors `Y_1, Y_2, ..., Y_n` which can be fed into `GeLU` independently:
+If we split the weight matrix `A` column-wise across `N` GPUs and perform matrix multiplications `XA_1` through `XA_n` in parallel, then we will end up with `N` output vectors `Y_1, Y_2, ..., Y_n` which can be fed into `GeLU` independently:
+![independent GeLU](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-independent-gelu.png)
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-independent-gelu.png" alt="Independent GeLU"/>
-</div>
+Using this principle, we can update an MLP of arbitrary depth, without the need for any synchronization between GPUs until the very end, where we need to reconstruct the output vector from shards. The Megatron-LM paper authors provide a helpful illustration for that:
+![parallel shard processing](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-parallel_shard_processing.png)
 
-Using this principle, we can update a multi-layer perceptron of arbitrary depth, without the need for any synchronization 
-between GPUs until the very end, where we need to reconstruct the output vector from shards. The Megatron-LM paper authors 
-provide a helpful illustration for that:
+Parallelizing the multi-headed attention layers is even simpler, since they are already inherently parallel, due to having multiple independent heads!
+![parallel self-attention](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-parallel_self_attention.png)
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-parallel_shard_processing.png" alt="Parallel shard processing"/>
-</div>
-
-Parallelizing the multi-headed attention layers is even simpler, since they are already inherently parallel, due to having 
-multiple independent heads!
-
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-tp-parallel_self_attention.png" alt="Parallel self-attention"/>
-</div>
-
-Special considerations: TP requires very fast network, and therefore it's not advisable to do TP across more than one node. 
-Practically, if a node has 4 GPUs, the highest TP degree is therefore 4. If you need a TP degree of 8, you need to use
-nodes that have at least 8 GPUs.
+Special considerations: TP requires very fast network, and therefore it's not advisable to do TP across more than one node. Practically, if a node has 4 GPUs, the highest TP degree is therefore 4. If you need a TP degree of 8, you need to use nodes that have at least 8 GPUs.
 
 This section is based on the original much more [detailed TP overview](https://github.com/huggingface/transformers/issues/10321#issuecomment-783543530).
 by [@anton-l](https://github.com/anton-l).
 
+SageMaker combines TP with DP for a more efficient processing.
+
 Alternative names:
 - DeepSpeed calls it [tensor slicing](https://www.deepspeed.ai/features/#model-parallelism)
 
@@ -442,27 +367,18 @@ Implementations:
 - [SageMaker](https://arxiv.org/abs/2111.05972) - this is a proprietary solution that can only be used on AWS.
 - [OSLO](https://github.com/tunib-ai/oslo) has the tensor parallelism implementation based on the Transformers.
 
-SageMaker combines TP with DP for a more efficient processing.
-
 🤗 Transformers status:
 - core: not yet implemented in the core
 - but if you want inference [parallelformers](https://github.com/tunib-ai/parallelformers) provides this support for most of our models. So until this is implemented in the core you can use theirs. And hopefully training mode will be supported too.
 - Deepspeed-Inference also supports our BERT, GPT-2, and GPT-Neo models in their super-fast CUDA-kernel-based inference mode, see more [here](https://www.deepspeed.ai/tutorials/inference-tutorial/)
 
-🤗 Accelerate integrates with [TP from Megatron-LM](https://huggingface.co/docs/accelerate/v0.23.0/en/usage_guides/megatron_lm).
+## DP+PP
 
-## Data Parallelism + Pipeline Parallelism
+The following diagram from the DeepSpeed [pipeline tutorial](https://www.deepspeed.ai/tutorials/pipeline/) demonstrates how one combines DP with PP.
 
-The following diagram from the DeepSpeed [pipeline tutorial](https://www.deepspeed.ai/tutorials/pipeline/) demonstrates 
-how one can combine DP with PP.
+![dp-pp-2d](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-zero-dp-pp.png)
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-zero-dp-pp.png" alt="DP + PP-2d"/>
-</div>
-
-Here it's important to see how DP rank 0 doesn't see GPU2 and DP rank 1 doesn't see GPU3. To DP there is just GPUs 0 
-and 1 where it feeds data as if there were just 2 GPUs. GPU0 "secretly" offloads some of its load to GPU2 using PP. 
-And GPU1 does the same by enlisting GPU3 to its aid.
+Here it's important to see how DP rank 0 doesn't see GPU2 and DP rank 1 doesn't see GPU3. To DP there is just GPUs 0 and 1 where it feeds data as if there were just 2 GPUs. GPU0 "secretly" offloads some of its load to GPU2 using PP. And GPU1 does the same by enlisting GPU3 to its aid.
 
 Since each dimension requires at least 2 GPUs, here you'd need at least 4 GPUs.
 
@@ -475,13 +391,11 @@ Implementations:
 
 🤗 Transformers status: not yet implemented
 
-## Data Parallelism + Pipeline Parallelism + Tensor Parallelism
+## DP+PP+TP
 
 To get an even more efficient training a 3D parallelism is used where PP is combined with TP and DP. This can be seen in the following diagram.
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-deepspeed-3d.png" alt="dp-pp-tp-3d"/>
-</div>
+![dp-pp-tp-3d](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-deepspeed-3d.png)
 
 This diagram is from a blog post [3D parallelism: Scaling to trillion-parameter models](https://www.microsoft.com/en-us/research/blog/deepspeed-extreme-scale-model-training-for-everyone/), which is a good read as well.
 
@@ -496,22 +410,15 @@ Implementations:
 
 🤗 Transformers status: not yet implemented, since we have no PP and TP.
 
-## ZeRO Data Parallelism + Pipeline Parallelism + Tensor Parallelism
+## ZeRO DP+PP+TP
 
-One of the main features of DeepSpeed is ZeRO, which is a super-scalable extension of DP. It has already been 
-discussed in [ZeRO Data Parallelism](#zero-data-parallelism). Normally it's a standalone feature that doesn't require PP or TP. 
-But it can be combined with PP and TP.
+One of the main features of DeepSpeed is ZeRO, which is a super-scalable extension of DP. It has already been discussed in [ZeRO Data Parallelism](#zero-data-parallelism). Normally it's a standalone feature that doesn't require PP or TP. But it can be combined with PP and TP.
 
 When ZeRO-DP is combined with PP (and optionally TP) it typically enables only ZeRO stage 1 (optimizer sharding).
 
-While it's theoretically possible to use ZeRO stage 2 (gradient sharding) with Pipeline Parallelism, it will have negative 
-performance impacts. There would need to be an additional reduce-scatter collective for every micro-batch to aggregate 
-the gradients before sharding, which adds a potentially significant communication overhead. By nature of Pipeline Parallelism, 
-small micro-batches are used and instead the focus is on trying to balance arithmetic intensity (micro-batch size) with
-minimizing the Pipeline bubble (number of micro-batches). Therefore those communication costs are going to impact the performance.
+While it's theoretically possible to use ZeRO stage 2 (gradient sharding) with Pipeline Parallelism, it will have bad performance impacts. There would need to be an additional reduce-scatter collective for every micro-batch to aggregate the gradients before sharding, which adds a potentially significant communication overhead. By nature of Pipeline Parallelism, small micro-batches are used and instead the focus is on trying to balance arithmetic intensity (micro-batch size) with minimizing the Pipeline bubble (number of micro-batches). Therefore those communication costs are going to hurt.
 
-In addition, there are already fewer layers than normal due to PP and so the memory savings won't be huge. PP already 
-reduces gradient size by ``1/PP``, and so gradient sharding savings on top of that are less significant than pure DP.
+In addition, There are already fewer layers than normal due to PP and so the memory savings won't be huge. PP already reduces gradient size by ``1/PP``, and so gradient sharding savings on top of that are less significant than pure DP.
 
 ZeRO stage 3 is not a good choice either for the same reason - more inter-node communications required.
 
@@ -548,9 +455,7 @@ Let's take 10 batches of sequence length 512. If we parallelize them by sample d
 
 * Operator
 
-If we perform layer normalization, we compute std first and mean second, and then we can normalize data. 
-Operator parallelism allows computing std and mean in parallel. So if we parallelize them by operator dimension into 2 
-devices (cuda:0, cuda:1), first we copy input data into both devices, and cuda:0 computes std, cuda:1 computes mean at the same time.
+If we perform layer normalization, we compute std first and mean second, and then we can normalize data. Operator parallelism allows computing std and mean in parallel. So if we parallelize them by operator dimension into 2 devices (cuda:0, cuda:1), first we copy input data into both devices, and cuda:0 computes std, cuda:1 computes mean at the same time.
 
 * Attribute
 
@@ -560,20 +465,66 @@ We have 10 batches of 512 length. If we parallelize them by attribute dimension
 
 It is similar with tensor model parallelism or naive layer-wise model parallelism.
 
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-flexflow.jpeg" alt="flex-flow-soap"/>
-</div>
+![flex-flow-soap](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/parallelism-flexflow.jpeg)
 
-The significance of this framework is that it takes resources like (1) GPU/TPU/CPU vs. (2) RAM/DRAM vs. (3) 
-fast-intra-connect/slow-inter-connect and it automatically optimizes all these algorithmically deciding which 
-parallelisation to use where.
+The significance of this framework is that it takes resources like (1) GPU/TPU/CPU vs. (2) RAM/DRAM vs. (3) fast-intra-connect/slow-inter-connect and it automatically optimizes all these  algorithmically deciding which parallelisation to use where.
 
-One very important aspect is that FlexFlow is designed for optimizing DNN parallelizations for models with static and 
-fixed workloads, since models with dynamic behavior may prefer different parallelization strategies across iterations.
+One very important aspect is that FlexFlow is designed for optimizing DNN parallelizations for models with static and fixed workloads, since models with dynamic behavior may prefer different parallelization strategies across iterations.
 
-So the promise is very attractive - it runs a 30min simulation on the cluster of choice and it comes up with the best 
-strategy to utilise this specific environment. If you add/remove/replace any parts it'll run and re-optimize the plan 
-for that. And then you can train. A different setup will have its own custom optimization.
+So the promise is very attractive - it runs a 30min simulation on the cluster of choice and it comes up with the best strategy to utilise this specific environment. If you add/remove/replace any parts it'll run and re-optimize the plan for that. And then you can train. A different setup will have its own custom optimization.
 
-🤗 Transformers status: Transformers models are FX-trace-able via [transformers.utils.fx](https://github.com/huggingface/transformers/blob/master/src/transformers/utils/fx.py), 
-which is a prerequisite for FlexFlow, however, changes are required on the FlexFlow side to make it work with Transformers models.
+🤗 Transformers status: not yet integrated. We already have our models FX-trace-able via [transformers.utils.fx](https://github.com/huggingface/transformers/blob/master/src/transformers/utils/fx.py), which is a prerequisite for FlexFlow, so someone needs to figure out what needs to be done to make FlexFlow work with our models.
+
+
+## Which Strategy To Use When
+
+Here is a very rough outline at which parallelism strategy to use when. The first on each list is typically faster.
+
+**⇨ Single GPU**
+
+* Model fits onto a single GPU:
+
+    1. Normal use
+
+* Model doesn't fit onto a single GPU:
+
+    1. ZeRO + Offload CPU and optionally NVMe
+    2. as above plus Memory Centric Tiling (see below for details) if the largest layer can't fit into a single GPU
+
+* Largest Layer not fitting into a single GPU:
+
+1. ZeRO - Enable [Memory Centric Tiling](https://deepspeed.readthedocs.io/en/latest/zero3.html#memory-centric-tiling) (MCT). It allows you to run arbitrarily large layers by automatically splitting them and executing them sequentially. MCT reduces the number of parameters that are live on a GPU, but it does not affect the activation memory. As this need is very rare as of this writing a manual override of `torch.nn.Linear` needs to be done by the user.
+
+**⇨ Single Node / Multi-GPU**
+
+* Model fits onto a single GPU:
+
+    1. DDP - Distributed DP
+    2. ZeRO - may or may not be faster depending on the situation and configuration used
+
+* Model doesn't fit onto a single GPU:
+
+    1. PP
+    2. ZeRO
+    3. TP
+
+    With very fast intra-node connectivity of NVLINK or NVSwitch all three should be mostly on par, without these PP will be faster than TP or ZeRO. The degree of TP may also make a difference. Best to experiment to find the winner on your particular setup.
+
+    TP is almost always used within a single node. That is TP size <= gpus per node.
+
+* Largest Layer not fitting into a single GPU:
+
+    1. If not using ZeRO - must use TP, as PP alone won't be able to fit.
+    2. With ZeRO see the same entry for "Single GPU" above
+
+
+**⇨ Multi-Node / Multi-GPU**
+
+* When you have fast inter-node connectivity:
+
+    1. ZeRO - as it requires close to no modifications to the model
+    2. PP+TP+DP - less communications, but requires massive changes to the model
+
+* when you have slow inter-node connectivity and still low on GPU memory:
+
+    1. DP+PP+TP+ZeRO-1

docs/source/en/performance.md

@@ -53,7 +53,7 @@ sections we go through the steps to run inference on CPU and single/multi-GPU se
 
 * [Inference on a single CPU](perf_infer_cpu)
 * [Inference on a single GPU](perf_infer_gpu_one)
-* [Multi-GPU inference](perf_infer_gpu_one)
+* [Multi-GPU inference](perf_infer_gpu_many)
 * [XLA Integration for TensorFlow Models](tf_xla)
 
 

docs/source/en/preprocessing.md

@@ -306,7 +306,7 @@ Create a function to preprocess the dataset so the audio samples are the same le
 ...     return inputs
 ```
 
-Apply the `preprocess_function` to the first few examples in the dataset:
+Apply the `preprocess_function` to the the first few examples in the dataset:
 
 ```py
 >>> processed_dataset = preprocess_function(dataset[:5])
@@ -412,7 +412,8 @@ If you wish to normalize images as a part of the augmentation transformation, us
 and `image_processor.image_std` values.
 </Tip>
 
-3. Then use 🤗 Datasets[`~datasets.Dataset.set_transform`] to apply the transforms on the fly:
+3. Then use 🤗 Datasets [`set_transform`](https://huggingface.co/docs/datasets/process.html#format-transform) to apply the transforms on the fly:
+
 ```py
 >>> dataset.set_transform(transforms)
 ```

docs/source/en/tasks/idefics.md

@@ -276,7 +276,7 @@ We can instruct the model to classify the image into one of the categories that
 >>> inputs = processor(prompt, return_tensors="pt").to("cuda")
 >>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
 
->>> generated_ids = model.generate(**inputs, max_new_tokens=6, bad_words_ids=bad_words_ids)
+>>> generated_ids = model.generate(**inputs, max_new_tokens=4, bad_words_ids=bad_words_ids)
 >>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
 >>> print(generated_text[0])
 Instruction: Classify the following image into a single category from the following list: ['animals', 'vegetables', 'city landscape', 'cars', 'office'].
@@ -357,7 +357,7 @@ for a batch of examples by passing a list of prompts:
 ...     ],
 ... ]
 
->>> inputs = processor(prompts, return_tensors="pt").to("cuda")
+>>> inputs = processor(prompts, return_tensors="pt")
 >>> bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
 
 >>> generated_ids = model.generate(**inputs, max_new_tokens=10, bad_words_ids=bad_words_ids)

docs/source/en/tasks/image_to_image.md

@@ -1,132 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Image-to-Image Task Guide
-
-[[open-in-colab]]
-
-Image-to-Image task is the task where an application receives an image and outputs another image. This has various subtasks, including image enhancement (super resolution, low light enhancement, deraining and so on), image inpainting, and more. 
-
-This guide will show you how to:
-- Use an image-to-image pipeline for super resolution task,
-- Run image-to-image models for same task without a pipeline.
-
-Note that as of the time this guide is released, `image-to-image` pipeline only supports super resolution task.
-
-Let's begin by installing the necessary libraries.
-
-```bash
-pip install transformers
-```
-
-We can now initialize the pipeline with a [Swin2SR model](https://huggingface.co/caidas/swin2SR-lightweight-x2-64). We can then infer with the pipeline by calling it with an image. As of now, only [Swin2SR models](https://huggingface.co/models?sort=trending&search=swin2sr) are supported in this pipeline. 
-
-```python
-from transformers import pipeline
-
-device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-pipe = pipeline(task="image-to-image", model="caidas/swin2SR-lightweight-x2-64", device=device)
-```
-
-Now, let's load an image.
-
-```python
-from PIL import Image
-import requests
-
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/cat.jpg"
-image = Image.open(requests.get(url, stream=True).raw)
-
-print(image.size)
-```
-```bash
-# (532, 432)
-```
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/cat.jpg" alt="Photo of a cat"/>
-</div>
-
-We can now do inference with the pipeline. We will get an upscaled version of the cat image. 
-
-```python
-upscaled = pipe(image)
-print(upscaled.size)
-```
-```bash
-# (1072, 880)
-```
-
-If you wish to do inference yourself with no pipeline, you can use the `Swin2SRForImageSuperResolution` and `Swin2SRImageProcessor` classes of transformers. We will use the same model checkpoint for this. Let's initialize the model and the processor.
-
-```python
-from transformers import Swin2SRForImageSuperResolution, Swin2SRImageProcessor 
-
-model = Swin2SRForImageSuperResolution.from_pretrained("caidas/swin2SR-lightweight-x2-64").to(device)
-processor = Swin2SRImageProcessor("caidas/swin2SR-lightweight-x2-64")
-```
-
-`pipeline` abstracts away the preprocessing and postprocessing steps that we have to do ourselves, so let's preprocess the image. We will pass the image to the processor and then move the pixel values to GPU. 
-
-```python
-pixel_values = processor(image, return_tensors="pt").pixel_values
-print(pixel_values.shape)
-
-pixel_values = pixel_values.to(device)
-```
-
-We can now infer the image by passing pixel values to the model.
-
-```python
-import torch
-
-with torch.no_grad():
-  outputs = model(pixel_values)
-```
-Output is an object of type `ImageSuperResolutionOutput` that looks like below 👇 
-
-```
-(loss=None, reconstruction=tensor([[[[0.8270, 0.8269, 0.8275,  ..., 0.7463, 0.7446, 0.7453],
-          [0.8287, 0.8278, 0.8283,  ..., 0.7451, 0.7448, 0.7457],
-          [0.8280, 0.8273, 0.8269,  ..., 0.7447, 0.7446, 0.7452],
-          ...,
-          [0.5923, 0.5933, 0.5924,  ..., 0.0697, 0.0695, 0.0706],
-          [0.5926, 0.5932, 0.5926,  ..., 0.0673, 0.0687, 0.0705],
-          [0.5927, 0.5914, 0.5922,  ..., 0.0664, 0.0694, 0.0718]]]],
-       device='cuda:0'), hidden_states=None, attentions=None)
-```
-We need to get the `reconstruction` and post-process it for visualization. Let's see how it looks like.
-
-```python
-outputs.reconstruction.data.shape
-# torch.Size([1, 3, 880, 1072])
-```
-
-We need to squeeze the output and get rid of axis 0, clip the values, then convert it to be numpy float. Then we will arrange axes to have the shape [1072, 880], and finally, bring the output back to range [0, 255].
-
-```python
-import numpy as np
-
-# squeeze, take to CPU and clip the values
-output = outputs.reconstruction.data.squeeze().cpu().clamp_(0, 1).numpy()
-# rearrange the axes
-output = np.moveaxis(output, source=0, destination=-1)
-# bring values back to pixel values range
-output = (output * 255.0).round().astype(np.uint8)
-Image.fromarray(output)
-```
-<div class="flex justify-center">
-     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/cat_upscaled.png" alt="Upscaled photo of a cat"/>
-</div>

docs/source/en/tasks/knowledge_distillation_for_image_classification.md

@@ -1,186 +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.
-
--->
-# Knowledge Distillation for Computer Vision
-
-[[open-in-colab]]
-
-Knowledge distillation is a technique used to transfer knowledge from a larger, more complex model (teacher) to a smaller, simpler model (student). To distill knowledge from one model to another, we take a pre-trained teacher model trained on a certain task (image classification for this case) and randomly initialize a student model to be trained on image classification. Next, we train the student model to minimize the difference between it's outputs and the teacher's outputs, thus making it mimic the behavior. It was first introduced in [Distilling the Knowledge in a Neural Network by Hinton et al](https://arxiv.org/abs/1503.02531). In this guide, we will do task-specific knowledge distillation. We will use the [beans dataset](https://huggingface.co/datasets/beans) for this.
-
-This guide demonstrates how you can distill a [fine-tuned ViT model](https://huggingface.co/merve/vit-mobilenet-beans-224) (teacher model) to a [MobileNet](https://huggingface.co/google/mobilenet_v2_1.4_224) (student model) using the [Trainer API](https://huggingface.co/docs/transformers/en/main_classes/trainer#trainer) of 🤗 Transformers. 
-
-Let's install the libraries needed for distillation and evaluating the process. 
-
-```bash
-pip install transformers datasets accelerate tensorboard evaluate --upgrade
-```
-
-In this example, we are using the `merve/beans-vit-224` model as teacher model. It's an image classification model, based on `google/vit-base-patch16-224-in21k` fine-tuned on beans dataset. We will distill this model to a randomly initialized MobileNetV2.
-
-We will now load the dataset. 
-
-```python
-from datasets import load_dataset
-
-dataset = load_dataset("beans")
-```
-
-We can use an image processor from either of the models, as in this case they return the same output with same resolution. We will use the `map()` method of `dataset` to apply the preprocessing to every split of the dataset. 
-
-```python
-from transformers import AutoImageProcessor
-teacher_processor = AutoImageProcessor.from_pretrained("merve/beans-vit-224")
-
-def process(examples):
-    processed_inputs = teacher_processor(examples["image"])
-    return processed_inputs
-
-processed_datasets = dataset.map(process, batched=True)
-```
-
-Essentially, we want the student model (a randomly initialized MobileNet) to mimic the teacher model (fine-tuned vision transformer). To achieve this, we first get the logits output from the teacher and the student. Then, we divide each of them by the parameter `temperature` which controls the importance of each soft target. A parameter called `lambda` weighs the importance of the distillation loss. In this example, we will use `temperature=5` and `lambda=0.5`. We will use the Kullback-Leibler Divergence loss to compute the divergence between the student and teacher. Given two data P and Q, KL Divergence explains how much extra information we need to represent P using Q. If two are identical, their KL divergence is zero, as there's no other information needed to explain P from Q. Thus, in the context of knowledge distillation, KL divergence is useful.
-
-
-```python
-from transformers import TrainingArguments, Trainer
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-
-class ImageDistilTrainer(Trainer):
-    def __init__(self, *args, teacher_model=None, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.teacher = teacher_model
-        self.student = student_model
-        self.loss_function = nn.KLDivLoss(reduction="batchmean")
-        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-        self.teacher.to(device)
-        self.teacher.eval()
-        self.temperature = temperature
-        self.lambda_param = lambda_param
-
-    def compute_loss(self, student, inputs, return_outputs=False):
-        student_output = self.student(**inputs)
-
-        with torch.no_grad():
-          teacher_output = self.teacher(**inputs)
-
-        # Compute soft targets for teacher and student
-        soft_teacher = F.softmax(teacher_output.logits / self.temperature, dim=-1)
-        soft_student = F.log_softmax(student_output.logits / self.temperature, dim=-1)
-
-        # Compute the loss
-        distillation_loss = self.loss_function(soft_student, soft_teacher) * (self.temperature ** 2)
-
-        # Compute the true label loss
-        student_target_loss = student_output.loss
-
-        # Calculate final loss
-        loss = (1. - self.lambda_param) * student_target_loss + self.lambda_param * distillation_loss
-        return (loss, student_output) if return_outputs else loss
-```
-
-We will now login to Hugging Face Hub so we can push our model to the Hugging Face Hub through the `Trainer`. 
-
-```python
-from huggingface_hub import notebook_login
-
-notebook_login()
-```
-
-Let's set the `TrainingArguments`, the teacher model and the student model. 
-
-```python
-from transformers import AutoModelForImageClassification, MobileNetV2Config, MobileNetV2ForImageClassification
-
-training_args = TrainingArguments(
-    output_dir="my-awesome-model",
-    num_train_epochs=30,
-    fp16=True,
-    logging_dir=f"{repo_name}/logs",
-    logging_strategy="epoch",
-    evaluation_strategy="epoch",
-    save_strategy="epoch",
-    load_best_model_at_end=True,
-    metric_for_best_model="accuracy",
-    report_to="tensorboard",
-    push_to_hub=True,
-    hub_strategy="every_save",
-    hub_model_id=repo_name,
-    )
-
-num_labels = len(processed_datasets["train"].features["labels"].names)
-
-# initialize models
-teacher_model = AutoModelForImageClassification.from_pretrained(
-    "merve/beans-vit-224",
-    num_labels=num_labels,
-    ignore_mismatched_sizes=True
-)
-
-# training MobileNetV2 from scratch
-student_config = MobileNetV2Config()
-student_config.num_labels = num_labels
-student_model = MobileNetV2ForImageClassification(student_config)
-```
-
-We can use `compute_metrics` function to evaluate our model on the test set. This function will be used during the training process to compute the `accuracy` & `f1` of our model.
-
-```python
-import evaluate
-import numpy as np
-
-accuracy = evaluate.load("accuracy")
-
-def compute_metrics(eval_pred):
-    predictions, labels = eval_pred
-    acc = accuracy.compute(references=labels, predictions=np.argmax(predictions, axis=1))
-    return {"accuracy": acc["accuracy"]}
-```
-
-Let's initialize the `Trainer` with the training arguments we defined. We will also initialize our data collator.
-
-```python
-from transformers import DefaultDataCollator
-
-data_collator = DefaultDataCollator()
-trainer = ImageDistilTrainer(
-    student_model=student_model,
-    teacher_model=teacher_model,
-    training_args=training_args,
-    train_dataset=processed_datasets["train"],
-    eval_dataset=processed_datasets["validation"],
-    data_collator=data_collator,
-    tokenizer=teacher_extractor,
-    compute_metrics=compute_metrics,
-    temperature=5,
-    lambda_param=0.5
-)
-```
-
-We can now train our model.
-
-```python
-trainer.train()
-```
-
-We can evaluate the model on the test set.
-
-```python
-trainer.evaluate(processed_datasets["test"])
-```
-
-On test set, our model reaches 72 percent accuracy. To have a sanity check over efficiency of distillation, we also trained MobileNet on the beans dataset from scratch with the same hyperparameters and observed 63 percent accuracy on the test set. We invite the readers to try different pre-trained teacher models, student architectures, distillation parameters and report their findings. The training logs and checkpoints for distilled model can be found in [this repository](https://huggingface.co/merve/vit-mobilenet-beans-224), and MobileNetV2 trained from scratch can be found in this [repository](https://huggingface.co/merve/resnet-mobilenet-beans-5).

docs/source/en/tasks/language_modeling.md

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

docs/source/en/tasks/prompting.md

@@ -1,439 +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.
-
--->
-
-
-# LLM prompting guide
-
-[[open-in-colab]]
-
-Large Language Models such as Falcon, LLaMA, etc. are pretrained transformer models initially trained to predict the 
-next token given some input text. They typically have billions of parameters and have been trained on trillions of 
-tokens for an extended period of time. As a result, these models become quite powerful and versatile, and you can use 
-them to solve multiple NLP tasks out of the box by instructing the models with natural language prompts.
-
-Designing such prompts to ensure the optimal output is often called "prompt engineering". Prompt engineering is an 
-iterative process that requires a fair amount of experimentation. Natural languages are much more flexible and expressive 
-than programming languages, however, they can also introduce some ambiguity. At the same time, prompts in natural language 
-are quite sensitive to changes. Even minor modifications in prompts can lead to wildly different outputs.
-
-While there is no exact recipe for creating prompts to match all cases, researchers have worked out a number of best 
-practices that help to achieve optimal results more consistently. 
-
-This guide covers the prompt engineering best practices to help you craft better LLM prompts and solve various NLP tasks. 
-You'll learn:
-
-- [Basics of prompting](#basic-prompts)
-- [Best practices of LLM prompting](#best-practices-of-llm-prompting)
-- [Advanced prompting techniques: few-shot prompting and chain-of-thought](#advanced-prompting-techniques)
-- [When to fine-tune instead of prompting](#prompting-vs-fine-tuning)
-
-<Tip>
-
-Prompt engineering is only a part of the LLM output optimization process. Another essential component is choosing the 
-optimal text generation strategy. You can customize how your LLM selects each of the subsequent tokens when generating 
-the text without modifying any of the trainable parameters. By tweaking the text generation parameters, you can reduce 
-repetition in the generated text and make it more coherent and human-sounding. 
-Text generation strategies and parameters are out of scope for this guide, but you can learn more about these topics in 
-the following guides: 
- 
-* [Generation with LLMs](../llm_tutorial)
-* [Text generation strategies](../generation_strategies)
-
-</Tip>
-
-## Basics of prompting
-
-### Types of models 
-
-The majority of modern LLMs are decoder-only transformers. Some examples include: [LLaMA](../model_doc/llama), 
-[Llama2](../model_doc/llama2), [Falcon](../model_doc/falcon), [GPT2](../model_doc/gpt2). However, you may encounter
-encoder-decoder transformer LLMs as well, for instance, [Flan-T5](../model_doc/flan-t5) and [BART](../model_doc/bart).
-
-Encoder-decoder-style models are typically used in generative tasks where the output **heavily** relies on the input, for 
-example, in translation and summarization. The decoder-only models are used for all other types of generative tasks.
-
-When using a pipeline to generate text with an LLM, it's important to know what type of LLM you are using, because 
-they use different pipelines. 
-
-Run inference with decoder-only models with the `text-generation` pipeline:
-
-```python
->>> from transformers import pipeline
->>> import torch
-
->>> torch.manual_seed(0) # doctest: +IGNORE_RESULT
-
->>> generator = pipeline('text-generation', model = 'gpt2')
->>> prompt = "Hello, I'm a language model"
-
->>> generator(prompt, max_length = 30)
-[{'generated_text': "Hello, I'm a language model expert, so I'm a big believer in the concept that I know very well and then I try to look into"}]
-```
-
-To run inference with an encoder-decoder, use the `text2text-generation` pipeline:
-
-```python
->>> text2text_generator = pipeline("text2text-generation", model = 'google/flan-t5-base')
->>> prompt = "Translate from English to French: I'm very happy to see you"
-
->>> text2text_generator(prompt)
-[{'generated_text': 'Je suis très heureuse de vous rencontrer.'}]
-```
-
-### Base vs instruct/chat models
-
-Most of the recent LLM checkpoints available on 🤗 Hub come in two versions: base and instruct (or chat). For example, 
-[`tiiuae/falcon-7b`](https://huggingface.co/tiiuae/falcon-7b) and [`tiiuae/falcon-7b-instruct`](https://huggingface.co/tiiuae/falcon-7b-instruct).
-
-Base models are excellent at completing the text when given an initial prompt, however, they are not ideal for NLP tasks 
-where they need to follow instructions, or for conversational use. This is where the instruct (chat) versions come in. 
-These checkpoints are the result of further fine-tuning of the pre-trained base versions on instructions and conversational data. 
-This additional fine-tuning makes them a better choice for many NLP tasks.  
-
-Let's illustrate some simple prompts that you can use with [`tiiuae/falcon-7b-instruct`](https://huggingface.co/tiiuae/falcon-7b-instruct) 
-to solve some common NLP tasks.
-
-### NLP tasks 
-
-First, let's set up the environment: 
-
-```bash
-pip install -q transformers accelerate
-```
-
-Next, let's load the model with the appropriate pipeline (`"text-generation"`): 
-
-```python
->>> from transformers import pipeline, AutoTokenizer
->>> import torch
-
->>> torch.manual_seed(0) # doctest: +IGNORE_RESULT
->>> model = "tiiuae/falcon-7b-instruct"
-
->>> tokenizer = AutoTokenizer.from_pretrained(model)
->>> pipe = pipeline(
-...     "text-generation",
-...     model=model,
-...     tokenizer=tokenizer,
-...     torch_dtype=torch.bfloat16,
-...     device_map="auto",
-... )
-```
-
-<Tip>
-
-Note that Falcon models were trained using the `bfloat16` datatype, so we recommend you use the same. This requires a recent 
-version of CUDA and works best on modern cards.
-
-</Tip>
-
-Now that we have the model loaded via the pipeline, let's explore how you can use prompts to solve NLP tasks.
-
-#### Text classification
-
-One of the most common forms of text classification is sentiment analysis, which assigns a label like "positive", "negative", 
-or "neutral" to a sequence of text. Let's write a prompt that instructs the model to classify a given text (a movie review). 
-We'll start by giving the instruction, and then specifying the text to classify. Note that instead of leaving it at that, we're 
-also adding the beginning of the response - `"Sentiment: "`:
-
-```python
->>> torch.manual_seed(0) # doctest: +IGNORE_RESULT
->>> prompt = """Classify the text into neutral, negative or positive. 
-... Text: This movie is definitely one of my favorite movies of its kind. The interaction between respectable and morally strong characters is an ode to chivalry and the honor code amongst thieves and policemen.
-... Sentiment:
-... """
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=10,
-... )
-
->>> for seq in sequences:
-...     print(f"Result: {seq['generated_text']}")
-Result: Classify the text into neutral, negative or positive. 
-Text: This movie is definitely one of my favorite movies of its kind. The interaction between respectable and morally strong characters is an ode to chivalry and the honor code amongst thieves and policemen.
-Sentiment:
-Positive
-```
-
-As a result, the output contains a classification label from the list we have provided in the instructions, and it is a correct one!
-
-<Tip>
-
-You may notice that in addition to the prompt, we pass a `max_new_tokens` parameter. It controls the number of tokens the 
-model shall generate, and it is one of the many text generation parameters that you can learn about 
-in [Text generation strategies](../generation_strategies) guide.
-
-</Tip>
-
-#### Named Entity Recognition
-
-Named Entity Recognition (NER) is a task of finding named entities in a piece of text, such as a person, location, or organization.
-Let's modify the instructions in the prompt to make the LLM perform this task. Here, let's also set `return_full_text = False` 
-so that output doesn't contain the prompt:
-
-```python
->>> torch.manual_seed(1) # doctest: +IGNORE_RESULT
->>> prompt = """Return a list of named entities in the text.
-... Text: The Golden State Warriors are an American professional basketball team based in San Francisco.
-... Named entities:
-... """
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=15,
-...     return_full_text = False,    
-... )
-
->>> for seq in sequences:
-...     print(f"{seq['generated_text']}")
-- Golden State Warriors
-- San Francisco
-```
-
-As you can see, the model correctly identified two named entities from the given text.
-
-#### Translation
-
-Another task LLMs can perform is translation. You can choose to use encoder-decoder models for this task, however, here,
-for the simplicity of the examples, we'll keep using Falcon-7b-instruct, which does a decent job. Once again, here's how 
-you can write a basic prompt to instruct a model to translate a piece of text from English to Italian: 
-
-```python
->>> torch.manual_seed(2) # doctest: +IGNORE_RESULT
->>> prompt = """Translate the English text to Italian.
-... Text: Sometimes, I've believed as many as six impossible things before breakfast.
-... Translation:
-... """
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=20,
-...     do_sample=True,
-...     top_k=10,
-...     return_full_text = False,
-... )
-
->>> for seq in sequences:
-...     print(f"{seq['generated_text']}")
-A volte, ho creduto a sei impossibili cose prima di colazione.
-```
-
-Here we've added a `do_sample=True` and `top_k=10` to allow the model to be a bit more flexible when generating output.
-
-#### Text summarization
-
-Similar to the translation, text summarization is another generative task where the output **heavily** relies on the input, 
-and encoder-decoder models can be a better choice. However, decoder-style models can be used for this task as well.
-Previously, we have placed the instructions at the very beginning of the prompt. However, the very end of the prompt can 
-also be a suitable location for instructions. Typically, it's better to place the instruction on one of the extreme ends.  
-
-```python
->>> torch.manual_seed(3) # doctest: +IGNORE_RESULT
->>> prompt = """Permaculture is a design process mimicking the diversity, functionality and resilience of natural ecosystems. The principles and practices are drawn from traditional ecological knowledge of indigenous cultures combined with modern scientific understanding and technological innovations. Permaculture design provides a framework helping individuals and communities develop innovative, creative and effective strategies for meeting basic needs while preparing for and mitigating the projected impacts of climate change.
-... Write a summary of the above text.
-... Summary:
-... """
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=30,
-...     do_sample=True,
-...     top_k=10,
-...     return_full_text = False,
-... )
-
->>> for seq in sequences:
-...     print(f"{seq['generated_text']}")
-Permaculture is an ecological design mimicking natural ecosystems to meet basic needs and prepare for climate change. It is based on traditional knowledge and scientific understanding.
-```
-
-#### Question answering
-
-For question answering task we can structure the prompt into the following logical components: instructions, context, question, and 
-the leading word or phrase (`"Answer:"`) to nudge the model to start generating the answer:
-
-```python
->>> torch.manual_seed(4) # doctest: +IGNORE_RESULT
->>> prompt = """Answer the question using the context below.
-... Context: Gazpacho is a cold soup and drink made of raw, blended vegetables. Most gazpacho includes stale bread, tomato, cucumbers, onion, bell peppers, garlic, olive oil, wine vinegar, water, and salt. Northern recipes often include cumin and/or pimentón (smoked sweet paprika). Traditionally, gazpacho was made by pounding the vegetables in a mortar with a pestle; this more laborious method is still sometimes used as it helps keep the gazpacho cool and avoids the foam and silky consistency of smoothie versions made in blenders or food processors.
-... Question: What modern tool is used to make gazpacho?
-... Answer:
-... """
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=10,
-...     do_sample=True,
-...     top_k=10,
-...     return_full_text = False,
-... )
-
->>> for seq in sequences:
-...     print(f"Result: {seq['generated_text']}")
-Result: Modern tools are used, such as immersion blenders
-```
-
-#### Reasoning
-
-Reasoning is one of the most difficult tasks for LLMs, and achieving good results often requires applying advanced prompting techniques, like 
-[Chain-of-though](#chain-of-thought).
-
-Let's try if we can make a model reason about a simple arithmetics task with a basic prompt: 
-
-```python
->>> torch.manual_seed(5) # doctest: +IGNORE_RESULT
->>> prompt = """There are 5 groups of students in the class. Each group has 4 students. How many students are there in the class?"""
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=30,
-...     do_sample=True,
-...     top_k=10,
-...     return_full_text = False,
-... )
-
->>> for seq in sequences:
-...     print(f"Result: {seq['generated_text']}")
-Result: 
-There are a total of 5 groups, so there are 5 x 4=20 students in the class.
-```
-
-Correct! Let's increase the complexity a little and see if we can still get away with a basic prompt:
-
-```python
->>> torch.manual_seed(6) # doctest: +IGNORE_RESULT
->>> prompt = """I baked 15 muffins. I ate 2 muffins and gave 5 muffins to a neighbor. My partner then bought 6 more muffins and ate 2. How many muffins do we now have?"""
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=10,
-...     do_sample=True,
-...     top_k=10,
-...     return_full_text = False,
-... )
-
->>> for seq in sequences:
-...     print(f"Result: {seq['generated_text']}")
-Result: 
-The total number of muffins now is 21
-```
-
-This is a wrong answer, it should be 12. In this case, this can be due to the prompt being too basic, or due to the choice 
-of model, after all we've picked the smallest version of Falcon. Reasoning is difficult for models of all sizes, but larger 
-models are likely to perform better. 
-
-## Best practices of LLM prompting
-
-In this section of the guide we have compiled a list of best practices that tend to improve the prompt results:
-
-* When choosing the model to work with, the latest and most capable models are likely to perform better. 
-* Start with a simple and short prompt, and iterate from there.
-* Put the instructions at the beginning of the prompt, or at the very end. When working with large context, models apply various optimizations to prevent Attention complexity from scaling quadratically. This may make a model more attentive to the beginning or end of a prompt than the middle.
-* Clearly separate instructions from the text they apply to - more on this in the next section. 
-* Be specific and descriptive about the task and the desired outcome - its format, length, style, language, etc.
-* Avoid ambiguous descriptions and instructions.
-* Favor instructions that say "what to do" instead of those that say "what not to do".
-* "Lead" the output in the right direction by writing the first word (or even begin the first sentence for the model).
-* Use advanced techniques like [Few-shot prompting](#few-shot-prompting) and [Chain-of-thought](#chain-of-thought)
-* Test your prompts with different models to assess their robustness. 
-* Version and track the performance of your prompts. 
-
-## Advanced prompting techniques
-
-### Few-shot prompting
-
-The basic prompts in the sections above are the examples of "zero-shot" prompts, meaning, the model has been given 
-instructions and context, but no examples with solutions. LLMs that have been fine-tuned on instruction datasets, generally 
-perform well on such "zero-shot" tasks. However, you may find that your task has more complexity or nuance, and, perhaps, 
-you have some requirements for the output that the model doesn't catch on just from the instructions. In this case, you can 
-try the technique called few-shot prompting. 
-
-In few-shot prompting, we provide examples in the prompt giving the model more context to improve the performance. 
-The examples condition the model to generate the output following the patterns in the examples.
-
-Here's an example: 
-
-```python
->>> torch.manual_seed(0) # doctest: +IGNORE_RESULT
->>> prompt = """Text: The first human went into space and orbited the Earth on April 12, 1961.
-... Date: 04/12/1961
-... Text: The first-ever televised presidential debate in the United States took place on September 28, 1960, between presidential candidates John F. Kennedy and Richard Nixon. 
-... Date:"""
-
->>> sequences = pipe(
-...     prompt,
-...     max_new_tokens=8,
-...     do_sample=True,
-...     top_k=10,
-... )
-
->>> for seq in sequences:
-...     print(f"Result: {seq['generated_text']}")
-Result: Text: The first human went into space and orbited the Earth on April 12, 1961.
-Date: 04/12/1961
-Text: The first-ever televised presidential debate in the United States took place on September 28, 1960, between presidential candidates John F. Kennedy and Richard Nixon. 
-Date: 09/28/1960
-```
-
-In the above code snippet we used a single example to demonstrate the desired output to the model, so this can be called a 
-"one-shot" prompting. However, depending on the task complexity you may need to use more than one example. 
-
-Limitations of the few-shot prompting technique: 
-- While LLMs can pick up on the patterns in the examples, these technique doesn't work well on complex reasoning tasks
-- Few-shot prompting requires creating lengthy prompts. Prompts with large number of tokens can increase computation and latency. There's also a limit to the length of the prompts.  
-- Sometimes when given a number of examples, models can learn patterns that you didn't intend them to learn, e.g. that the third movie review is always negative.
-
-### Chain-of-thought
-
-Chain-of-thought (CoT) prompting is a technique that nudges a model to produce intermediate reasoning steps thus improving 
-the results on complex reasoning tasks. 
-
-There are two ways of steering a model to producing the reasoning steps:
-- few-shot prompting by illustrating examples with detailed answers to questions, showing the model how to work through a problem.
-- by instructing the model to reason by adding phrases like "Let's think step by step" or "Take a deep breath and work through the problem step by step."
-
-If we apply the CoT technique to the muffins example from the [reasoning section](#reasoning) and use a larger model, 
-such as (`tiiuae/falcon-180B-chat`) which you can play with in the [HuggingChat](https://huggingface.co/chat/), 
-we'll get a significant improvement on the reasoning result:
-
-```text
-Let's go through this step-by-step:
-1. You start with 15 muffins.
-2. You eat 2 muffins, leaving you with 13 muffins.
-3. You give 5 muffins to your neighbor, leaving you with 8 muffins.
-4. Your partner buys 6 more muffins, bringing the total number of muffins to 14.
-5. Your partner eats 2 muffins, leaving you with 12 muffins.
-Therefore, you now have 12 muffins.
-```
-
-## Prompting vs fine-tuning
-
-You can achieve great results by optimizing your prompts, however, you may still ponder whether fine-tuning a model 
-would work better for your case. Here are some scenarios when fine-tuning a smaller model may be a preferred option:
-
-- Your domain is wildly different from what LLMs were pre-trained on and extensive prompt optimization did not yield sufficient results. 
-- You need your model to work well in a low-resource language.
-- You need the model to be trained on sensitive data that is under strict regulations. 
-- You have to use a small model due to cost, privacy, infrastructure or other limitations. 
-
-In all of the above examples, you will need to make sure that you either already have or can easily obtain a large enough 
-domain-specific dataset at a reasonable cost to fine-tune a model. You will also need to have enough time and resources 
-to fine-tune a model.
-
-If the above examples are not the case for you, optimizing prompts can prove to be more beneficial.   
-
-

docs/source/en/tasks/summarization.md

@@ -35,7 +35,7 @@ The task illustrated in this tutorial is supported by the following model archit
 
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
 
-[BART](../model_doc/bart), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [Encoder decoder](../model_doc/encoder-decoder), [FairSeq Machine-Translation](../model_doc/fsmt), [GPTSAN-japanese](../model_doc/gptsan-japanese), [LED](../model_doc/led), [LongT5](../model_doc/longt5), [M2M100](../model_doc/m2m_100), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MT5](../model_doc/mt5), [MVP](../model_doc/mvp), [NLLB](../model_doc/nllb), [NLLB-MOE](../model_doc/nllb-moe), [Pegasus](../model_doc/pegasus), [PEGASUS-X](../model_doc/pegasus_x), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [SeamlessM4T](../model_doc/seamless_m4t), [SwitchTransformers](../model_doc/switch_transformers), [T5](../model_doc/t5), [UMT5](../model_doc/umt5), [XLM-ProphetNet](../model_doc/xlm-prophetnet)
+[BART](../model_doc/bart), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [Encoder decoder](../model_doc/encoder-decoder), [FairSeq Machine-Translation](../model_doc/fsmt), [GPTSAN-japanese](../model_doc/gptsan-japanese), [LED](../model_doc/led), [LongT5](../model_doc/longt5), [M2M100](../model_doc/m2m_100), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MT5](../model_doc/mt5), [MVP](../model_doc/mvp), [NLLB](../model_doc/nllb), [NLLB-MOE](../model_doc/nllb-moe), [Pegasus](../model_doc/pegasus), [PEGASUS-X](../model_doc/pegasus_x), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [SwitchTransformers](../model_doc/switch_transformers), [T5](../model_doc/t5), [UMT5](../model_doc/umt5), [XLM-ProphetNet](../model_doc/xlm-prophetnet)
 
 <!--End of the generated tip-->
 

docs/source/en/tasks/translation.md

@@ -32,7 +32,7 @@ The task illustrated in this tutorial is supported by the following model archit
 
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
 
-[BART](../model_doc/bart), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [Encoder decoder](../model_doc/encoder-decoder), [FairSeq Machine-Translation](../model_doc/fsmt), [GPTSAN-japanese](../model_doc/gptsan-japanese), [LED](../model_doc/led), [LongT5](../model_doc/longt5), [M2M100](../model_doc/m2m_100), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MT5](../model_doc/mt5), [MVP](../model_doc/mvp), [NLLB](../model_doc/nllb), [NLLB-MOE](../model_doc/nllb-moe), [Pegasus](../model_doc/pegasus), [PEGASUS-X](../model_doc/pegasus_x), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [SeamlessM4T](../model_doc/seamless_m4t), [SwitchTransformers](../model_doc/switch_transformers), [T5](../model_doc/t5), [UMT5](../model_doc/umt5), [XLM-ProphetNet](../model_doc/xlm-prophetnet)
+[BART](../model_doc/bart), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [Encoder decoder](../model_doc/encoder-decoder), [FairSeq Machine-Translation](../model_doc/fsmt), [GPTSAN-japanese](../model_doc/gptsan-japanese), [LED](../model_doc/led), [LongT5](../model_doc/longt5), [M2M100](../model_doc/m2m_100), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MT5](../model_doc/mt5), [MVP](../model_doc/mvp), [NLLB](../model_doc/nllb), [NLLB-MOE](../model_doc/nllb-moe), [Pegasus](../model_doc/pegasus), [PEGASUS-X](../model_doc/pegasus_x), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [SwitchTransformers](../model_doc/switch_transformers), [T5](../model_doc/t5), [UMT5](../model_doc/umt5), [XLM-ProphetNet](../model_doc/xlm-prophetnet)
 
 <!--End of the generated tip-->
 

docs/source/en/testing.md

@@ -57,6 +57,8 @@ RUN_SLOW=1 pytest examples/
 
 
 
+
+
 ### Choosing which tests to run
 
 This document goes into many details of how tests can be run. If after reading everything, you need even more details
@@ -182,7 +184,6 @@ pytest -k "test and ada" tests/test_optimization.py
 ### Run `accelerate` tests
 
 Sometimes you need to run `accelerate` tests on your models. For that you can just add `-m accelerate_tests` to your command, if let's say you want to run these tests on `OPT` run:
-
 ```bash
 RUN_SLOW=1 pytest -m accelerate_tests tests/models/opt/test_modeling_opt.py 
 ```
@@ -513,7 +514,6 @@ n_gpu = get_gpu_count()  # works with torch and tf
 ### Testing with a specific PyTorch backend or device
 
 To run the test suite on a specific torch device add `TRANSFORMERS_TEST_DEVICE="$device"` where `$device` is the target backend. For example, to test on CPU only:
-
 ```bash
 TRANSFORMERS_TEST_DEVICE="cpu" pytest tests/utils/test_logging.py
 ```
@@ -521,29 +521,9 @@ TRANSFORMERS_TEST_DEVICE="cpu" pytest tests/utils/test_logging.py
 This variable is useful for testing custom or less common PyTorch backends such as `mps`. It can also be used to achieve the same effect as `CUDA_VISIBLE_DEVICES` by targeting specific GPUs or testing in CPU-only mode.
 
 Certain devices will require an additional import after importing `torch` for the first time. This can be specified using the environment variable `TRANSFORMERS_TEST_BACKEND`:
-
 ```bash
 TRANSFORMERS_TEST_BACKEND="torch_npu" pytest tests/utils/test_logging.py
 ```
-Alternative backends may also require the replacement of device-specific functions. For example `torch.cuda.manual_seed` may need to be replaced with a device-specific seed setter like `torch.npu.manual_seed` to correctly set a random seed on the device. To specify a new backend with backend-specific device functions when running the test suite, create a Python device specification file in the format:
-
-```
-import torch
-import torch_npu
-# !! Further additional imports can be added here !!
-
-# Specify the device name (eg. 'cuda', 'cpu', 'npu')
-DEVICE_NAME = 'npu'
-
-# Specify device-specific backends to dispatch to.
-# If not specified, will fallback to 'default' in 'testing_utils.py`
-MANUAL_SEED_FN = torch.npu.manual_seed
-EMPTY_CACHE_FN = torch.npu.empty_cache
-DEVICE_COUNT_FN = torch.npu.device_count
-```
-This format also allows for specification of any additional imports required. To use this file to replace equivalent methods in the test suite, set the environment variable `TRANSFORMERS_TEST_DEVICE_SPEC` to the path of the spec file.
-
-Currently, only `MANUAL_SEED_FN`, `EMPTY_CACHE_FN` and `DEVICE_COUNT_FN` are supported for device-specific dispatch.
 
 
 ### Distributed training
@@ -899,8 +879,7 @@ or the `xfail` way:
 def test_feature_x():
 ```
 
-
-Here's how to skip a test based on internal checks within the test:
+- Here is how to skip a test based on some internal check inside the test:
 
 ```python
 def test_feature_x():

docs/source/hi/_toctree.yml

@@ -1,3 +0,0 @@
-- sections:
-  - local: pipeline_tutorial
-    title: पाइपलाइनों के साथ अनुमान चलाएँ

docs/source/hi/pipeline_tutorial.md

@@ -1,317 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# अनुमान के लिए पाइपलाइन
-
-[`pipeline`] किसी भी भाषा, कंप्यूटर दृष्टि, भाषण और मल्टीमॉडल कार्यों पर अनुमान लगाने के लिए [Hub] (https://huggingface.co/models) से किसी भी मॉडल का उपयोग करना आसान बनाता है। भले ही आपके पास किसी विशिष्ट तौर-तरीके का अनुभव न हो या आप मॉडलों के पीछे अंतर्निहित कोड से परिचित न हों, फिर भी आप [`pipeline`] के अनुमान के लिए उनका उपयोग कर सकते हैं! यह ट्यूटोरियल आपको ये सिखाएगा:
-
-* अनुमान के लिए [`pipeline`] का उपयोग करें।
-* एक विशिष्ट टोकननाइज़र या मॉडल का उपयोग करें।
-* ऑडियो, विज़न और मल्टीमॉडल कार्यों के लिए [`pipeline`] का उपयोग करें।
-
-<Tip>
-
-समर्थित कार्यों और उपलब्ध मापदंडों की पूरी सूची के लिए [`pipeline`] दस्तावेज़ पर एक नज़र डालें।
-
-</Tip>
-
-## पाइपलाइन का उपयोग
-
-जबकि प्रत्येक कार्य में एक संबद्ध [`pipeline`] होता है, सामान्य [`pipeline`] अमूर्त का उपयोग करना आसान होता है जिसमें शामिल होता है
-सभी कार्य-विशिष्ट पाइपलाइनें। [`pipeline`] स्वचालित रूप से एक डिफ़ॉल्ट मॉडल और सक्षम प्रीप्रोसेसिंग क्लास लोड करता है
-आपके कार्य के लिए अनुमान का. आइए स्वचालित वाक् पहचान (एएसआर) के लिए [`pipeline`] का उपयोग करने का उदाहरण लें, या
-वाक्-से-पाठ.
-
-
-1. एक [`pipeline`] बनाकर प्रारंभ करें और अनुमान कार्य निर्दिष्ट करें:
-
-```py
->>> from transformers import pipeline
-
->>> transcriber = pipeline(task="automatic-speech-recognition")
-```
-
-2. अपना इनपुट [`pipeline`] पर भेजें। वाक् पहचान के मामले में, यह एक ऑडियो इनपुट फ़ाइल है:
-
-```py
->>> transcriber("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac")
-{'text': 'I HAVE A DREAM BUT ONE DAY THIS NATION WILL RISE UP LIVE UP THE TRUE MEANING OF ITS TREES'}
-```
-
-क्या वह परिणाम नहीं जो आपके मन में था? कुछ [सबसे अधिक डाउनलोड किए गए स्वचालित वाक् पहचान मॉडल](https://huggingface.co/models?pipeline_tag=automatic-speech-recognition&sort=trending) देखें
-यह देखने के लिए हब पर जाएं कि क्या आपको बेहतर ट्रांस्क्रिप्शन मिल सकता है।
-
-आइए OpenAI से [व्हिस्पर लार्ज-v2](https://huggingface.co/openai/whisper-large) मॉडल आज़माएं। व्हिस्पर जारी किया गया
-Wav2Vec2 की तुलना में 2 साल बाद, और लगभग 10 गुना अधिक डेटा पर प्रशिक्षित किया गया था। इस प्रकार, यह अधिकांश डाउनस्ट्रीम पर Wav2Vec2 को मात देता है
-बेंचमार्क. इसमें विराम चिह्न और आवरण की भविष्यवाणी करने का अतिरिक्त लाभ भी है, जिनमें से कोई भी संभव नहीं है
-Wav2Vec2.
-
-आइए इसे यहां आज़माकर देखें कि यह कैसा प्रदर्शन करता है:
-
-```py
->>> transcriber = pipeline(model="openai/whisper-large-v2")
->>> transcriber("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac")
-{'text': ' I have a dream that one day this nation will rise up and live out the true meaning of its creed.'}
-```
-
-अब यह परिणाम अधिक सटीक दिखता है! Wav2Vec2 बनाम व्हिस्पर पर गहन तुलना के लिए, [ऑडियो ट्रांसफॉर्मर्स कोर्स] (https://huggingface.co/learn/audio-course/chapter5/asr_models) देखें।
-हम वास्तव में आपको विभिन्न भाषाओं में मॉडल, आपके क्षेत्र में विशेषीकृत मॉडल और बहुत कुछ के लिए हब की जांच करने के लिए प्रोत्साहित करते हैं।
-आप हब पर सीधे अपने ब्राउज़र से मॉडल परिणामों की जांच और तुलना कर सकते हैं कि यह फिट बैठता है या नहीं
-अन्य मामलों की तुलना में कोने के मामलों को बेहतर ढंग से संभालता है।
-और यदि आपको अपने उपयोग के मामले के लिए कोई मॉडल नहीं मिलता है, तो आप हमेशा अपना खुद का [प्रशिक्षण](training) शुरू कर सकते हैं!
-
-यदि आपके पास कई इनपुट हैं, तो आप अपने इनपुट को एक सूची के रूप में पास कर सकते हैं:
-
-```py
-transcriber(
-    [
-        "https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac",
-        "https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/1.flac",
-    ]
-)
-```
-
-पाइपलाइनें प्रयोग के लिए बहुत अच्छी हैं क्योंकि एक मॉडल से दूसरे मॉडल पर स्विच करना मामूली काम है; हालाँकि, प्रयोग की तुलना में बड़े कार्यभार के लिए उन्हें अनुकूलित करने के कुछ तरीके हैं। संपूर्ण डेटासेट पर पुनरावृत्ति करने या वेबसर्वर में पाइपलाइनों का उपयोग करने के बारे में निम्नलिखित मार्गदर्शिकाएँ देखें:
-दस्तावेज़ों में से:
-* [डेटासेट पर पाइपलाइनों का उपयोग करना](#using-pipelines-on-a-dataset)
-* [वेबसर्वर के लिए पाइपलाइनों का उपयोग करना](./pipeline_webserver)
-
-## प्राचल
-
-[`pipeline`] कई मापदंडों का समर्थन करता है; कुछ कार्य विशिष्ट हैं, और कुछ सभी पाइपलाइनों के लिए सामान्य हैं।
-सामान्य तौर पर, आप अपनी इच्छानुसार कहीं भी पैरामीटर निर्दिष्ट कर सकते हैं:
-
-```py
-transcriber = pipeline(model="openai/whisper-large-v2", my_parameter=1)
-
-out = transcriber(...)  # This will use `my_parameter=1`.
-out = transcriber(..., my_parameter=2)  # This will override and use `my_parameter=2`.
-out = transcriber(...)  # This will go back to using `my_parameter=1`.
-```
-
-आइए 3 महत्वपूर्ण बातों पर गौर करें:
-
-### उपकरण
-
-यदि आप `device=0` का उपयोग करते हैं, तो पाइपलाइन स्वचालित रूप से मॉडल को निर्दिष्ट डिवाइस पर डाल देती है।
-यह इस पर ध्यान दिए बिना काम करेगा कि आप PyTorch या Tensorflow का उपयोग कर रहे हैं या नहीं।
-
-```py
-transcriber = pipeline(model="openai/whisper-large-v2", device=0)
-```
-
-यदि मॉडल एकल GPU के लिए बहुत बड़ा है और आप PyTorch का उपयोग कर रहे हैं, तो आप `device_map="auto"` को स्वचालित रूप से सेट कर सकते हैं
-निर्धारित करें कि मॉडल वज़न को कैसे लोड और संग्रहीत किया जाए। `device_map` तर्क का उपयोग करने के लिए 🤗 [Accelerate] (https://huggingface.co/docs/accelerate) की आवश्यकता होती है
-पैकेट:
-
-```bash
-pip install --upgrade accelerate
-```
-
-निम्नलिखित कोड स्वचालित रूप से सभी डिवाइसों में मॉडल भार को लोड और संग्रहीत करता है:
-
-```py
-transcriber = pipeline(model="openai/whisper-large-v2", device_map="auto")
-```
-
-ध्यान दें कि यदि `device_map='auto'` पारित हो गया है, तो अपनी `pipeline` को चालू करते समय `device=device` तर्क जोड़ने की कोई आवश्यकता नहीं है क्योंकि आपको कुछ अप्रत्याशित व्यवहार का सामना करना पड़ सकता है!
-
-### बैच का आकार
-
-डिफ़ॉल्ट रूप से, पाइपलाइनें [यहां] (https://huggingface.co/docs/transformers/main_classes/pipelines#pipeline-batching) विस्तार से बताए गए कारणों के लिए बैच अनुमान नहीं लगाएंगी। इसका कारण यह है कि बैचिंग आवश्यक रूप से तेज़ नहीं है, और वास्तव में कुछ मामलों में काफी धीमी हो सकती है।
-
-लेकिन अगर यह आपके उपयोग के मामले में काम करता है, तो आप इसका उपयोग कर सकते हैं:
-
-```py
-transcriber = pipeline(model="openai/whisper-large-v2", device=0, batch_size=2)
-audio_filenames = [f"https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/{i}.flac" for i in range(1, 5)]
-texts = transcriber(audio_filenames)
-```
-
-यह प्रदान की गई 4 ऑडियो फाइलों पर पाइपलाइन चलाता है, लेकिन यह उन्हें 2 के बैच में पास करेगा
-आपसे किसी और कोड की आवश्यकता के बिना मॉडल (जो एक जीपीयू पर है, जहां बैचिंग से मदद मिलने की अधिक संभावना है) पर जाएं।
-आउटपुट हमेशा उसी से मेल खाना चाहिए जो आपको बैचिंग के बिना प्राप्त हुआ होगा। इसका उद्देश्य केवल पाइपलाइन से अधिक गति प्राप्त करने में आपकी सहायता करना है।
-
-पाइपलाइनें बैचिंग की कुछ जटिलताओं को भी कम कर सकती हैं क्योंकि, कुछ पाइपलाइनों के लिए, एक एकल आइटम (जैसे एक लंबी ऑडियो फ़ाइल) को एक मॉडल द्वारा संसाधित करने के लिए कई भागों में विभाजित करने की आवश्यकता होती है। पाइपलाइन आपके लिए यह [*chunk batching*](./main_classes/pipelines#pipeline-chunk-batching) करती है।
-
-### कार्य विशिष्ट प्राचल
-
-सभी कार्य कार्य विशिष्ट प्राचल प्रदान करते हैं जो आपको अपना काम पूरा करने में मदद करने के लिए अतिरिक्त लचीलेपन और विकल्पों की अनुमति देते हैं।
-उदाहरण के लिए, [`transformers.AutomaticSpeechRecognitionPipeline.__call__`] विधि में एक `return_timestamps` प्राचल है जो वीडियो उपशीर्षक के लिए आशाजनक लगता है:
-
-
-```py
->>> transcriber = pipeline(model="openai/whisper-large-v2", return_timestamps=True)
->>> transcriber("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac")
-{'text': ' I have a dream that one day this nation will rise up and live out the true meaning of its creed.', 'chunks': [{'timestamp': (0.0, 11.88), 'text': ' I have a dream that one day this nation will rise up and live out the true meaning of its'}, {'timestamp': (11.88, 12.38), 'text': ' creed.'}]}
-```
-
-जैसा कि आप देख सकते हैं, मॉडल ने पाठ का अनुमान लगाया और **when** विभिन्न वाक्यों का उच्चारण किया गया तो आउटपुट भी दिया।
-
-प्रत्येक कार्य के लिए कई प्राचल उपलब्ध हैं, इसलिए यह देखने के लिए कि आप किसके साथ छेड़छाड़ कर सकते हैं, प्रत्येक कार्य का API संदर्भ देखें!
-उदाहरण के लिए, [`~transformers.AutomaticSpeechRecognitionPipeline`] में एक `chunk_length_s` प्राचल है जो सहायक है
-वास्तव में लंबी ऑडियो फ़ाइलों पर काम करने के लिए (उदाहरण के लिए, संपूर्ण फिल्मों या घंटे-लंबे वीडियो को उपशीर्षक देना) जो आमतौर पर एक मॉडल होता है
-अपने आप संभाल नहीं सकता:
-
-```python
->>> transcriber = pipeline(model="openai/whisper-large-v2", chunk_length_s=30, return_timestamps=True)
->>> transcriber("https://huggingface.co/datasets/sanchit-gandhi/librispeech_long/resolve/main/audio.wav")
-{'text': " Chapter 16. I might have told you of the beginning of this liaison in a few lines, but I wanted you to see every step by which we came.  I, too, agree to whatever Marguerite wished, Marguerite to be unable to live apart from me. It was the day after the evening...
-```
-
-यदि आपको कोई ऐसा पैरामीटर नहीं मिल रहा है जो वास्तव में आपकी मदद करेगा, तो बेझिझक [अनुरोध करें](https://github.com/huggingface/transformers/issues/new?assignees=&labels=feature&template=feature-request.yml)!
-
-
-## डेटासेट पर पाइपलाइनों का उपयोग करना
-
-पाइपलाइन बड़े डेटासेट पर भी अनुमान चला सकती है। ऐसा करने का सबसे आसान तरीका हम एक पुनरावर्तक का उपयोग करने की सलाह देते हैं:
-
-```py
-def data():
-    for i in range(1000):
-        yield f"My example {i}"
-
-
-pipe = pipeline(model="gpt2", device=0)
-generated_characters = 0
-for out in pipe(data()):
-    generated_characters += len(out[0]["generated_text"])
-```
-
-पुनरावर्तक `data()` प्रत्येक परिणाम और पाइपलाइन स्वचालित रूप से उत्पन्न करता है
-पहचानता है कि इनपुट पुनरावर्तनीय है और डेटा प्राप्त करना शुरू कर देगा
-यह इसे GPU पर प्रोसेस करना जारी रखता है (यह हुड के तहत [DataLoader](https://pytorch.org/docs/stable/data.html#torch.utils.data.DataLoader) का उपयोग करता है)।
-यह महत्वपूर्ण है क्योंकि आपको संपूर्ण डेटासेट के लिए मेमोरी आवंटित करने की आवश्यकता नहीं है
-और आप जितनी जल्दी हो सके GPU को फीड कर सकते हैं।
-
-चूंकि बैचिंग से चीज़ें तेज़ हो सकती हैं, इसलिए यहां `batch_size` प्राचल को ट्यून करने का प्रयास करना उपयोगी हो सकता है।
-
-किसी डेटासेट पर पुनरावृति करने का सबसे सरल तरीका बस एक को 🤗 [Dataset](https://github.com/huggingface/datasets/) से लोड करना है:
-
-```py
-# KeyDataset is a util that will just output the item we're interested in.
-from transformers.pipelines.pt_utils import KeyDataset
-from datasets import load_dataset
-
-pipe = pipeline(model="hf-internal-testing/tiny-random-wav2vec2", device=0)
-dataset = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation[:10]")
-
-for out in pipe(KeyDataset(dataset, "audio")):
-    print(out)
-```
-
-
-## वेबसर्वर के लिए पाइपलाइनों का उपयोग करना
-
-<Tip>
-एक अनुमान इंजन बनाना एक जटिल विषय है जो अपने आप में उपयुक्त है
-पृष्ठ।
-</Tip>
-
-[Link](./pipeline_webserver)
-
-## विज़न पाइपलाइन
-
-दृष्टि कार्यों के लिए [`pipeline`] का उपयोग करना व्यावहारिक रूप से समान है।
-
-अपना कार्य निर्दिष्ट करें और अपनी छवि क्लासिफायरियर को भेजें। छवि एक लिंक, एक स्थानीय पथ या बेस64-एन्कोडेड छवि हो सकती है। उदाहरण के लिए, बिल्ली की कौन सी प्रजाति नीचे दिखाई गई है?
-
-![pipeline-cat-chonk](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg)
-
-```py
->>> from transformers import pipeline
-
->>> vision_classifier = pipeline(model="google/vit-base-patch16-224")
->>> preds = vision_classifier(
-...     images="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"
-... )
->>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
->>> preds
-[{'score': 0.4335, 'label': 'lynx, catamount'}, {'score': 0.0348, 'label': 'cougar, puma, catamount, mountain lion, painter, panther, Felis concolor'}, {'score': 0.0324, 'label': 'snow leopard, ounce, Panthera uncia'}, {'score': 0.0239, 'label': 'Egyptian cat'}, {'score': 0.0229, 'label': 'tiger cat'}]
-```
-
-## पाठ पाइपलाइन
-
-NLP कार्यों के लिए [`pipeline`] का उपयोग करना व्यावहारिक रूप से समान है।
-
-```py
->>> from transformers import pipeline
-
->>> # This model is a `zero-shot-classification` model.
->>> # It will classify text, except you are free to choose any label you might imagine
->>> classifier = pipeline(model="facebook/bart-large-mnli")
->>> classifier(
-...     "I have a problem with my iphone that needs to be resolved asap!!",
-...     candidate_labels=["urgent", "not urgent", "phone", "tablet", "computer"],
-... )
-{'sequence': 'I have a problem with my iphone that needs to be resolved asap!!', 'labels': ['urgent', 'phone', 'computer', 'not urgent', 'tablet'], 'scores': [0.504, 0.479, 0.013, 0.003, 0.002]}
-```
-
-## बहुविध पाइपलाइन
-
-[`pipeline`] एक से अधिक तौर-तरीकों का समर्थन करती है। उदाहरण के लिए, एक दृश्य प्रश्न उत्तर (VQA) कार्य पाठ और छवि को जोड़ता है। अपनी पसंद के किसी भी छवि लिंक और छवि के बारे में कोई प्रश्न पूछने के लिए स्वतंत्र महसूस करें। छवि एक URL या छवि का स्थानीय पथ हो सकती है।
-
-उदाहरण के लिए, यदि आप इस [invoice image](https://huggingface.co/spaces/impira/docquery/resolve/2359223c1837a7587402bda0f2643382a6eefeab/invoice.png) का उपयोग करते हैं:
-
-```py
->>> from transformers import pipeline
-
->>> vqa = pipeline(model="impira/layoutlm-document-qa")
->>> vqa(
-...     image="https://huggingface.co/spaces/impira/docquery/resolve/2359223c1837a7587402bda0f2643382a6eefeab/invoice.png",
-...     question="What is the invoice number?",
-... )
-[{'score': 0.42515, 'answer': 'us-001', 'start': 16, 'end': 16}]
-```
-
-<Tip>
-
-ऊपर दिए गए उदाहरण को चलाने के लिए आपको 🤗 ट्रांसफॉर्मर के अलावा [`pytesseract`](https://pypi.org/project/pytesseract/) इंस्टॉल करना होगा:
-
-```bash
-sudo apt install -y tesseract-ocr
-pip install pytesseract
-```
-
-</Tip>
-
-## 🤗 `त्वरण` के साथ बड़े मॉडलों पर `pipeline` का उपयोग करना:
-
-आप 🤗 `accelerate` का उपयोग करके बड़े मॉडलों पर आसानी से `pipeline` चला सकते हैं! पहले सुनिश्चित करें कि आपने `accelerate` को `pip install accelerate` के साथ इंस्टॉल किया है।
-
-सबसे पहले `device_map='auto'` का उपयोग करके अपना मॉडल लोड करें! हम अपने उदाहरण के लिए `facebook/opt-1.3b` का उपयोग करेंगे।
-
-```py
-# pip install accelerate
-import torch
-from transformers import pipeline
-
-pipe = pipeline(model="facebook/opt-1.3b", torch_dtype=torch.bfloat16, device_map="auto")
-output = pipe("This is a cool example!", do_sample=True, top_p=0.95)
-```
-
-यदि आप `bitsandbytes` इंस्टॉल करते हैं और `load_in_8bit=True` तर्क जोड़ते हैं तो आप 8-बिट लोडेड मॉडल भी पास कर सकते हैं
-
-```py
-# pip install accelerate bitsandbytes
-import torch
-from transformers import pipeline
-
-pipe = pipeline(model="facebook/opt-1.3b", device_map="auto", model_kwargs={"load_in_8bit": True})
-output = pipe("This is a cool example!", do_sample=True, top_p=0.95)
-```
-
-ध्यान दें कि आप चेकपॉइंट को किसी भी हगिंग फेस मॉडल से बदल सकते हैं जो BLOOM जैसे बड़े मॉडल लोडिंग का समर्थन करता है!

docs/source/ja/_toctree.yml

@@ -1,268 +1,14 @@
 - sections:
   - local: index
     title: 🤗 Transformers
-  - local: quicktour
-    title: クイックツアー
   - local: installation
     title: インストール
-  title: Get started
+  title: はじめに
 - sections:
-  - local: pipeline_tutorial
-    title: パイプラインを使用して推論を実行する
-  - local: autoclass_tutorial
-    title: AutoClass を使用して移植可能なコードを作成する
-  - local: preprocessing
-    title: データの前処理
-  - local: training
-    title: 事前トレーニングされたモデルを微調整する
-  - local: run_scripts
-    title: スクリプトを使用してトレーニングする
   - local: accelerate
-    title: 🤗 Accelerate を使用して分散トレーニングをセットアップする
-  - local: peft
-    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
-- sections:
-  - isExpanded: false
-  #   sections:
-  #   - local: tasks/sequence_classification
-  #     title: Text classification
-  #   - local: tasks/token_classification
-  #     title: Token classification
-  #   - local: tasks/question_answering
-  #     title: Question answering
-  #   - local: tasks/language_modeling
-  #     title: Causal language modeling
-  #   - local: tasks/masked_language_modeling
-  #     title: Masked language modeling
-  #   - local: tasks/translation
-  #     title: Translation
-  #   - local: tasks/summarization
-  #     title: Summarization
-  #   - local: tasks/multiple_choice
-  #     title: Multiple choice
-  #   title: Natural Language Processing
-  # - isExpanded: false
-  #   sections:
-  #   - local: tasks/audio_classification
-  #     title: Audio classification
-  #   - local: tasks/asr
-  #     title: Automatic speech recognition
-  #   title: Audio
-  # - isExpanded: false
-  #   sections:
-  #   - local: tasks/image_classification
-  #     title: Image classification
-  #   - local: tasks/semantic_segmentation
-  #     title: Semantic segmentation
-  #   - local: tasks/video_classification
-  #     title: Video classification
-  #   - local: tasks/object_detection
-  #     title: Object detection
-  #   - local: tasks/zero_shot_object_detection
-  #     title: Zero-shot object detection
-  #   - local: tasks/zero_shot_image_classification
-  #     title: Zero-shot image classification
-  #   - local: tasks/monocular_depth_estimation
-  #     title: Depth estimation
-  #   - local: tasks/image_to_image
-  #     title: Image-to-Image
-  #   - local: tasks/knowledge_distillation_for_image_classification
-  #     title: Knowledge Distillation for Computer Vision
-  #   title: Computer Vision
-  # - isExpanded: false
-  #   sections:
-  #   - local: tasks/image_captioning
-  #     title: Image captioning
-  #   - local: tasks/document_question_answering
-  #     title: Document Question Answering
-  #   - local: tasks/visual_question_answering
-  #     title: Visual Question Answering
-  #   - local: tasks/text-to-speech
-  #     title: Text to speech
-  #   title: Multimodal
-  - isExpanded: false
-    sections:
-    - local: generation_strategies
-      title: 生成戦略をカスタマイズする
-    title: Generation
-  # - isExpanded: false
-  #   sections:
-  #   - local: tasks/idefics
-  #     title: Image tasks with IDEFICS
-  #   - local: tasks/prompting
-  #     title: LLM prompting guide
-  #   title: Prompting
-    title: Task Guides
-- sections:
-  - local: fast_tokenizers
-    title: 🤗 トークナイザーの高速トークナイザーを使用する
-  - local: multilingual
-    title: 多言語モデルで推論を実行する
-  - local: create_a_model
-    title: モデル固有の API を使用する
-  - local: custom_models
-    title: カスタムモデルを共有する
-  - local: chat_templating
-    title: チャットモデルのテンプレート
-  - local: serialization
-    title: ONNX へのエクスポート
-  - local: tflite
-    title: TFLite へのエクスポート
-  - local: torchscript
-    title: トーチスクリプトへのエクスポート
-  - local: benchmarks
-    title: ベンチマーク
-  - 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
-- sections:
-  - local: performance
-    title: 概要
-  - sections:
-    - local: perf_train_gpu_one
-      title: 単一の GPU で効率的にトレーニングするための方法とツール
-    - local: perf_train_gpu_many
-      title: 複数の GPU と並列処理
-    - local: perf_train_cpu
-      title: CPU での効率的なトレーニング
-    - local: perf_train_cpu_many
-      title: 分散CPUトレーニング
-    - local: perf_train_tpu
-      title: TPU に関するトレーニング
-    - local: perf_train_tpu_tf
-      title: TensorFlow を使用した TPU のトレーニング
-    - local: perf_train_special
-      title: 特殊なハードウェアに関するトレーニング
-    - local: perf_hardware
-      title: トレーニング用のカスタム ハードウェア
-    - local: hpo_train
-      title: Trainer API を使用したハイパーパラメータ検索
-    title: 効率的なトレーニングテクニック
-  - sections:
-    - local: perf_infer_cpu
-      title: CPUでの推論
-    - local: perf_infer_gpu_one
-      title: 1 つの GPU での推論
-    - local: perf_infer_gpu_many
-      title: 多くの GPU での推論
-    - local: perf_infer_special
-      title: 特殊なハードウェアでの推論
-    title: 推論の最適化
-  - local: big_models
-    title: 大きなモデルのインスタンス化
-  - local: tf_xla
-    title: TensorFlowモデルのXLA統合
-  - local: perf_torch_compile
-    title: torch.compile()を使用した推論の最適化
-  title: パフォーマンスとスケーラビリティ
-- sections:
-  - local: add_new_model
-    title: 🤗 Transformersにモデルを追加する方法
-  - local: add_tensorflow_model
-    title: 🤗 TransformersモデルをTensorFlowに変換する方法
-  - local: testing
-    title: テスト
-  - local: pr_checks
-    title: プルリクエストのチェック
-  title: 貢献する
-- sections:
-  - local: philosophy
-    title: フィロソフィー
-  - local: glossary
-    title: 用語集
-  - local: task_summary
-    title: 🤗 Transformersの機能
-  - local: tasks_explained
-    title: 🤗 Transformersがタスクを解決する方法
-  - local: model_summary
-    title: Transformerモデルファミリー
-  - local: tokenizer_summary
-    title: トークナイザーの概要
-  - local: attention
-    title: 注意機構
-  - local: pad_truncation
-    title: パディングと切り詰め
-  - local: bertology
-    title: BERTology
-  - local: perplexity
-    title: 固定長モデルのパープレキシティ
-  - local: pipeline_webserver
-    title: Webサーバー推論用パイプライン
-  - local: model_memory_anatomy
-    title: モデルトレーニングの解剖学
-  title: コンセプチュアルガイド
+    title: 🤗 Accelerate を用いた分散学習
+  title: チュートリアル
 - 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
-      title: コールバック
-    - local: main_classes/configuration
-      title: 構成
-    - local: main_classes/data_collator
-      title: データ照合者
-    - local: main_classes/keras_callbacks
-      title: Keras コールバック
-    - local: main_classes/logging
-      title: ロギング
-    - local: main_classes/model
-      title: モデル
-    - local: main_classes/text_generation
-      title: テキストの生成
-    - local: main_classes/onnx
-      title: ONNX
-    - local: main_classes/optimizer_schedules
-      title: 最適化
-    - local: main_classes/output
-      title: モデルの出力
-    - local: main_classes/pipelines
-      title: パイプライン
-    - local: main_classes/processors
-      title: プロセッサー
-    - local: main_classes/quantization
-      title: 量子化
-    - local: main_classes/tokenizer
-      title: トークナイザー
-    - local: main_classes/trainer
-      title: トレーナー
-    - local: main_classes/deepspeed
-      title: ディープスピードの統合
-    - local: main_classes/feature_extractor
-      title: 特徴抽出器
-    - local: main_classes/image_processor
-      title: 画像処理プロセッサ
-    title: 主要なクラス
-  - sections:
-    - local: internal/modeling_utils
-      title: カスタムレイヤーとユーティリティ
-    - local: internal/pipelines_utils
-      title: パイプライン用のユーティリティ
-    - local: internal/tokenization_utils
-      title: ト=ークナイザー用のユーティリティ
-    - local: internal/trainer_utils
-      title: トレーナー用ユーティリティ
-    - local: internal/generation_utils
-      title: 発電用ユーティリティ
-    - local: internal/image_processing_utils
-      title: 画像プロセッサ用ユーティリティ
-    - local: internal/audio_utils
-      title: オーディオ処理用のユーティリティ
-    - local: internal/file_utils
-      title: 一般公共事業
-    - local: internal/time_series_utils
-      title: 時系列用のユーティリティ
-    title: 内部ヘルパー
-  title: API
+    - local: multilingual
+      title: 推論のための多言語モデル

docs/source/ja/add_new_model.md

@@ -1,756 +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
-
-⚠️ このファイルはMarkdown形式ですが、特定の文法が含まれており、通常のMarkdownビューアーでは正しく表示されない場合があります。
-
--->
-
-# How to add a model to 🤗 Transformers?
-
-🤗 Transformersライブラリは、コミュニティの貢献者のおかげで新しいモデルを提供できることがよくあります。
-しかし、これは難しいプロジェクトであり、🤗 Transformersライブラリと実装するモデルについての深い知識が必要です。
-Hugging Faceでは、コミュニティの多くの人々に積極的にモデルを追加する力を与えようと努力しており、
-このガイドをまとめて、PyTorchモデルを追加するプロセスを説明します（[PyTorchがインストールされていることを確認してください](https://pytorch.org/get-started/locally/)）。
-
-<Tip>
-
-TensorFlowモデルを実装する興味がある場合は、[🤗 TransformersモデルをTensorFlowに変換する方法](add_tensorflow_model)ガイドを参照してみてください！
-
-</Tip>
-
-この過程で、以下のことを学びます：
-
-- オープンソースのベストプラクティスに関する洞察
-- 最も人気のある深層学習ライブラリの設計原則を理解する
-- 大規模なモデルを効率的にテストする方法を学ぶ
-- `black`、`ruff`、および`make fix-copies`などのPythonユーティリティを統合して、クリーンで読みやすいコードを確保する方法を学ぶ
-
-Hugging Faceチームのメンバーがサポートを提供するので、一人ぼっちになることはありません。 🤗 ❤️
-
-さあ、始めましょう！🤗 Transformersで見たいモデルについての[New model addition](https://github.com/huggingface/transformers/issues/new?assignees=&labels=New+model&template=new-model-addition.yml)のイシューを開いてください。
-特定のモデルを提供することに特にこだわりがない場合、[New model label](https://github.com/huggingface/transformers/labels/New%20model)で未割り当てのモデルリクエストがあるかどうかを確認して、それに取り組むことができます。
-
-新しいモデルリクエストを開いたら、最初のステップは🤗 Transformersをよく理解することです！
-
-## General overview of 🤗 Transformers
-
-まず、🤗 Transformersの一般的な概要を把握する必要があります。🤗 Transformersは非常に意見が分かれるライブラリですので、
-ライブラリの哲学や設計選択について同意できない可能性があります。ただし、私たちの経験から、ライブラリの基本的な設計選択と哲学は、
-🤗 Transformersを効率的にスケーリングし、適切なレベルで保守コストを抑えるために不可欠です。
-
-ライブラリの理解を深めるための良い出発点は、[哲学のドキュメント](philosophy)を読むことです。
-私たちの作業方法の結果、すべてのモデルに適用しようとするいくつかの選択肢があります：
-
-- 一般的に、抽象化よりも構成が優先されます。
-- コードの重複は、読みやすさやアクセス可能性を大幅に向上させる場合、必ずしも悪いわけではありません。
-- モデルファイルはできるだけ自己完結的であるべきで、特定のモデルのコードを読む際には、理想的には該当する`modeling_....py`ファイルのみを見る必要があります。
-
-私たちの意見では、このライブラリのコードは単なる製品を提供する手段だけでなく、*例えば、推論のためにBERTを使用する能力*などの製品そのもの.
-
-### Overview of models
-
-モデルを正常に追加するためには、モデルとその設定、[`PreTrainedModel`]、および[`PretrainedConfig`]の相互作用を理解することが重要です。
-例示的な目的で、🤗 Transformersに追加するモデルを「BrandNewBert」と呼びます。
-
-以下をご覧ください：
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_overview.png"/>
-
-ご覧のように、🤗 Transformersでは継承を使用していますが、抽象化のレベルを最小限に保っています。
-ライブラリ内のどのモデルにも、抽象化のレベルが2つを超えることはありません。
-`BrandNewBertModel` は `BrandNewBertPreTrainedModel` を継承し、さらに[`PreTrainedModel`]を継承しています。
-これだけです。
-一般的なルールとして、新しいモデルは[`PreTrainedModel`]にのみ依存するようにしたいと考えています。
-すべての新しいモデルに自動的に提供される重要な機能は、[`~PreTrainedModel.from_pretrained`]および
-[`~PreTrainedModel.save_pretrained`]です。
-これらはシリアライゼーションとデシリアライゼーションに使用されます。
-`BrandNewBertModel.forward`などの他の重要な機能は、新しい「modeling_brand_new_bert.py」スクリプトで完全に定義されるべきです。
-次に、特定のヘッドレイヤーを持つモデル（たとえば `BrandNewBertForMaskedLM` ）が `BrandNewBertModel` を継承するのではなく、
-抽象化のレベルを低く保つために、そのフォワードパスで `BrandNewBertModel` を呼び出すコンポーネントとして使用されるようにしたいと考えています。
-新しいモデルには常に `BrandNewBertConfig` という設定クラスが必要です。この設定は常に[`PreTrainedModel`]の属性として保存され、
-したがって、`BrandNewBertPreTrainedModel`から継承するすべてのクラスで`config`属性を介してアクセスできます。
-
-```python
-model = BrandNewBertModel.from_pretrained("brandy/brand_new_bert")
-model.config  # model has access to its config
-```
-
-モデルと同様に、設定は[`PretrainedConfig`]から基本的なシリアル化および逆シリアル化の機能を継承しています。注意すべきは、設定とモデルは常に2つの異なる形式にシリアル化されることです - モデルは*pytorch_model.bin*ファイルに、設定は*config.json*ファイルにシリアル化されます。[`~PreTrainedModel.save_pretrained`]を呼び出すと、自動的に[`~PretrainedConfig.save_pretrained`]も呼び出され、モデルと設定の両方が保存されます。
-
-### Code style
-
-新しいモデルをコーディングする際には、Transformersは意見があるライブラリであり、コードの書き方に関していくつかの独自の考え方があります :-)
-
-1. モデルのフォワードパスはモデリングファイルに完全に記述され、ライブラリ内の他のモデルとは完全に独立している必要があります。他のモデルからブロックを再利用したい場合、コードをコピーしてトップに`# Copied from`コメントを付けて貼り付けます（良い例は[こちら](https://github.com/huggingface/transformers/blob/v4.17.0/src/transformers/models/roberta/modeling_roberta.py#L160)、コピーに関する詳細なドキュメンテーションは[ここ](pr_checks#check-copies)を参照してください）。
-2. コードは完全に理解可能でなければなりません。これは記述的な変数名を選択し、省略形を避けるべきであることを意味します。例えば、`act`ではなく`activation`が好まれます。1文字の変数名は、forループ内のインデックスでない限り、強く非推奨です。
-3. より一般的に、魔法のような短いコードよりも長くて明示的なコードを好みます。
-4. PyTorchでは`nn.Sequential`をサブクラス化せずに、`nn.Module`をサブクラス化し、フォワードパスを記述し、コードを使用する他の人が簡単にデバッグできるようにします。プリントステートメントやブレークポイントを追加してデバッグできるようにします。
-5. 関数のシグネチャは型アノテーションを付けるべきです。その他の部分に関しては、型アノテーションよりも良い変数名が読みやすく理解しやすいことがあります。
-
-### Overview of tokenizers
-
-まだ完了していません :-( このセクションは近日中に追加されます！
-
-## Step-by-step recipe to add a model to 🤗 Transformers
-
-モデルを追加する方法は人それぞれ異なるため、他のコントリビューターが🤗 Transformersにモデルを追加する際の要約を確認することが非常に役立つ場合があります。以下は、他のコントリビューターが🤗 Transformersにモデルをポートする際のコミュニティブログ投稿のリストです。
-
-1. [GPT2モデルのポーティング](https://medium.com/huggingface/from-tensorflow-to-pytorch-265f40ef2a28) by [Thomas](https://huggingface.co/thomwolf)
-2. [WMT19 MTモデルのポーティング](https://huggingface.co/blog/porting-fsmt) by [Stas](https://huggingface.co/stas)
-
-経験から言えることは、モデルを追加する際に最も重要なことは次のようになります：
-
-- 車輪の再発明をしないでください！新しい🤗 Transformersモデルのために追加するコードのほとんどはすでに🤗 Transformers内のどこかに存在しています。類似した既存のモデルやトークナイザを見つけるために、いくつかの時間をかけて探すことが重要です。[grep](https://www.gnu.org/software/grep/)と[rg](https://github.com/BurntSushi/ripgrep)はあなたの友達です。モデルのトークナイザは1つのモデル実装に基づいているかもしれませんが、モデルのモデリングコードは別の実装に基づいていることがあることに注意してください。例えば、FSMTのモデリングコードはBARTに基づいており、FSMTのトークナイザコードはXLMに基づいています。
-- これは科学的な課題よりもエンジニアリングの課題です。モデルの論文の理論的な側面をすべて理解しようとするよりも、効率的なデバッグ環境を作成するために時間を費やすべきです。
-- 行き詰まった場合は助けを求めてください！モデルは🤗 Transformersのコアコンポーネントであり、Hugging Faceではモデルを追加するための各ステップでお手伝いするのを喜んでいます。進行がないことに気付いた場合は、進展していないことを気にしないでください。
-
-以下では、🤗 Transformersにモデルをポートする際に最も役立つと考えられる一般的なレシピを提供しようとしています。
-
-次のリストは、モデルを追加するために行う必要があるすべてのことの要約であり、To-Doリストとして使用できます：
-
-- ☐ （オプション）モデルの理論的な側面を理解しました
-- ☐ 🤗 Transformersの開発環境を準備しました
-- ☐ オリジナルのリポジトリのデバッグ環境をセットアップしました
-- ☐ `forward()` パスをオリジナルのリポジトリとチェックポイントで正常に実行するスクリプトを作成しました
-- ☐ モデルの骨格を🤗 Transformersに正常に追加しました
-- ☐ オリジナルのチェックポイントを🤗 Transformersのチェックポイントに正常に変換しました
-- ☐ 🤗 Transformersで実行される `forward()` パスを正常に実行し、オリジナルのチェックポイントと同一の出力を得ました
-- ☐ 🤗 Transformersでのモデルテストを完了しました
-- ☐ 🤗 Transformersにトークナイザを正常に追加しました
-- ☐ エンドツーエンドの統合テストを実行しました
-- ☐ ドキュメントを完成させました
-- ☐ モデルのウェイトをHubにアップロードしました
-- ☐ プルリクエストを提出しました
-- ☐ （オプション）デモノートブックを追加しました
-
-まず、通常、`BrandNewBert`の理論的な理解を深めることをお勧めします。
-ただし、もしモデルの理論的な側面を「実務中に理解する」方が好ましい場合、`BrandNewBert`のコードベースに直接アクセスするのも問題ありません。
-このオプションは、エンジニアリングのスキルが理論的なスキルよりも優れている場合、
-`BrandNewBert`の論文を理解するのに苦労している場合、または科学的な論文を読むよりもプログラミングを楽しんでいる場合に適しています。
-
-### 1. (Optional) Theoretical aspects of BrandNewBert
-
-BrandNewBertの論文がある場合、その説明を読むための時間を取るべきです。論文の中には理解が難しい部分があるかもしれません。
-その場合でも心配しないでください。目標は論文の深い理論的理解を得ることではなく、
-🤗 Transformersでモデルを効果的に再実装するために必要な情報を抽出することです。
-ただし、理論的な側面にあまり多くの時間をかける必要はありません。代わりに、実践的な側面に焦点を当てましょう。具体的には次の点です：
-
-- *brand_new_bert*はどの種類のモデルですか？ BERTのようなエンコーダーのみのモデルですか？ GPT2のようなデコーダーのみのモデルですか？ BARTのようなエンコーダー-デコーダーモデルですか？
-  [model_summary](model_summary)を参照して、これらの違いについて詳しく知りたい場合があります。
-- *brand_new_bert*の応用分野は何ですか？ テキスト分類ですか？ テキスト生成ですか？ Seq2Seqタスク、例えば要約ですか？
-- モデルをBERT/GPT-2/BARTとは異なるものにする新しい機能は何ですか？
-- 既存の[🤗 Transformersモデル](https://huggingface.co/transformers/#contents)の中で*brand_new_bert*に最も似ているモデルはどれですか？
-- 使用されているトークナイザの種類は何ですか？ SentencePieceトークナイザですか？ WordPieceトークナイザですか？ BERTやBARTで使用されているトークナイザと同じですか？
-
-モデルのアーキテクチャの良い概要を得たと感じたら、Hugging Faceチームに質問を送ることができます。
-これにはモデルのアーキテクチャ、注意層などに関する質問が含まれるかもしれません。
-私たちは喜んでお手伝いします。
-
-### 2. Next prepare your environment
-
-1. リポジトリのページで「Fork」ボタンをクリックして、[リポジトリ](https://github.com/huggingface/transformers)をフォークします。
-   これにより、コードのコピーがGitHubユーザーアカウントの下に作成されます。
-
-2. ローカルディスクにある`transformers`フォークをクローンし、ベースリポジトリをリモートとして追加します：
-
-```bash
-git clone https://github.com/[your Github handle]/transformers.git
-cd transformers
-git remote add upstream https://github.com/huggingface/transformers.git
-```
-
-```bash
-python -m venv .env
-source .env/bin/activate
-pip install -e ".[dev]"
-```
-
-3. 開発環境をセットアップするために、次のコマンドを実行してください：
-
-```bash
-python -m venv .env
-source .env/bin/activate
-pip install -e ".[dev]"
-```
-
-お使いのOSに応じて、およびTransformersのオプションの依存関係の数が増えているため、このコマンドでエラーが発生する可能性があります。
-その場合は、作業しているDeep Learningフレームワーク（PyTorch、TensorFlow、および/またはFlax）をインストールし、次の手順を実行してください：
-
-```bash
-pip install -e ".[quality]"
-```
-
-これはほとんどのユースケースには十分であるはずです。その後、親ディレクトリに戻ることができます。
-
-```bash
-cd ..
-```
-
-4. Transformersに*brand_new_bert*のPyTorchバージョンを追加することをお勧めします。PyTorchをインストールするには、
-   https://pytorch.org/get-started/locally/ の指示に従ってください。
-
-   **注意:** CUDAをインストールする必要はありません。新しいモデルをCPUで動作させることで十分です。
-
-5. *brand_new_bert*を移植するには、元のリポジトリへのアクセスも必要です。
-
-```bash
-git clone https://github.com/org_that_created_brand_new_bert_org/brand_new_bert.git
-cd brand_new_bert
-pip install -e .
-```
-
-
-*brand_new_bert*を🤗 Transformersにポートするための開発環境を設定しました。
-
-### 3.-4. Run a pretrained checkpoint using the original repository
-
-最初に、オリジナルの*brand_new_bert*リポジトリで作業します。通常、オリジナルの実装は非常に「研究的」であり、ドキュメンテーションが不足していたり、コードが理解しにくいことがあります。しかし、これが*brand_new_bert*を再実装する動機となるべきです。Hugging Faceでは、主要な目標の1つが、動作するモデルを取り、それをできるだけ**アクセス可能でユーザーフレンドリーで美しい**ものに書き直すことです。これは、🤗 Transformersにモデルを再実装する最も重要な動機です - 複雑な新しいNLP技術を**誰にでも**アクセス可能にしようとする試みです。
-
-まず、オリジナルのリポジトリに入り込むことから始めるべきです。
-
-公式の事前学習済みモデルをオリジナルのリポジトリで正常に実行することは、通常、**最も困難な**ステップです。
-私たちの経験から、オリジナルのコードベースに慣れるのに時間をかけることが非常に重要です。以下のことを理解する必要があります：
-
-- 事前学習済みの重みをどこで見つけるか？
-- 対応するモデルに事前学習済みの重みをロードする方法は？
-- モデルから独立してトークナイザを実行する方法は？
-- 1つのフォワードパスを追跡して、単純なフォワードパスに必要なクラスと関数がわかるようにします。通常、これらの関数だけを再実装する必要があります。
-- モデルの重要なコンポーネントを特定できること：モデルのクラスはどこにありますか？モデルのサブクラス、*例* EncoderModel、DecoderModelがありますか？自己注意レイヤーはどこにありますか？複数の異なる注意レイヤー、*例* *自己注意*、*クロスアテンション*などが存在しますか？
-- オリジナルのリポジトリの環境でモデルをデバッグする方法は？*print*ステートメントを追加する必要があるか、*ipdb*のような対話型デバッガを使用できるか、PyCharmのような効率的なIDEを使用してモデルをデバッグする必要がありますか？
-
-重要なのは、ポーティングプロセスを開始する前に、オリジナルのリポジトリでコードを**効率的に**デバッグできることです！また、これはオープンソースライブラリで作業していることを覚えておいてください。オリジナルのリポジトリでコードを調べる誰かを歓迎するために、問題をオープンにしたり、プルリクエストを送信したりすることをためらわないでください。このリポジトリのメンテナーは、彼らのコードを調べてくれる人に対して非常に喜んでいる可能性が高いです！
-
-この段階では、オリジナルのモデルのデバッグにどのような環境と戦略を使用するかは、あなた次第です。最初にオリジナルのリポジトリに関するコードをデバッグできることが非常に重要です。また、GPU環境をセットアップすることはお勧めしません。まず、CPU上で作業し、モデルがすでに🤗 Transformersに正常にポートされていることを確認します。最後に、モデルがGPU上でも期待通りに動作するかどうかを検証する必要があります。
-
-一般的に、オリジナルのモデルを実行するための2つのデバッグ環境があります：
-
--  [Jupyter notebooks](https://jupyter.org/) / [google colab](https://colab.research.google.com/notebooks/intro.ipynb)
--  ローカルなPythonスクリプト。
-
-Jupyterノートブックは、セルごとに実行できるため、論理的なコンポーネントをより分割し、中間結果を保存できるため、デバッグサイクルが速くなるという利点があります。また、ノートブックは他の共同作業者と簡単に共有できることが多く、Hugging Faceチームに助けを求める場合に非常に役立つ場合があります。Jupyterノートブックに精通している場合、それ
-
-
-```python
-model = BrandNewBertModel.load_pretrained_checkpoint("/path/to/checkpoint/")
-input_ids = [0, 4, 5, 2, 3, 7, 9]  # vector of input ids
-original_output = model.predict(input_ids)
-```
-
-デバッグ戦略については、通常、いくつかの選択肢があります：
-
-- 元のモデルを多くの小さなテスト可能なコンポーネントに分解し、それぞれに対して前方パスを実行して検証します
-- 元のモデルを元のトークナイザと元のモデルにのみ分解し、それらに対して前方パスを実行し、検証のために中間のプリントステートメントまたはブレークポイントを使用します
-
-再度、どの戦略を選択するかはあなた次第です。元のコードベースに依存することが多く、元のコードベースに応じて一方または他方が有利なことがあります。
-
-元のコードベースがモデルを小さなサブコンポーネントに分解できる場合、*例えば*元のコードベースが簡単にイーガーモードで実行できる場合、それを行う価値が通常あります。最初からより難しい方法を選択することにはいくつかの重要な利点があります：
-
-- 後で元のモデルを🤗 Transformersの実装と比較する際に、各コンポーネントが対応する🤗 Transformers実装のコンポーネントと一致することを自動的に検証できるため、視覚的な比較に依存せずに済みます
-- 大きな問題を小さな問題に分解する、つまり個々のコンポーネントのみをポーティングする問題に分割するのに役立ち、作業を構造化するのに役立ちます
-- モデルを論理的な意味のあるコンポーネントに分割することで、モデルの設計をよりよく理解しやすくし、モデルをよりよく理解するのに役立ちます
-- 後で、コンポーネントごとのテストを行うことで、コードを変更し続ける際にリグレッションが発生しないことを確認するのに役立ちます
-
-[Lysandreの](https://gist.github.com/LysandreJik/db4c948f6b4483960de5cbac598ad4ed) ELECTRAの統合チェックは、これがどのように行われるかの良い例です。
-
-ただし、元のコードベースが非常に複雑で、中間コンポーネントをコンパイルモードで実行することしか許可しない場合、モデルを小さなテスト可能なサブコンポーネントに分解することが時間がかかりすぎるか、不可能であることがあります。
-良い例は[T5のMeshTensorFlow](https://github.com/tensorflow/mesh/tree/master/mesh_tensorflow)ライブラリであり、非常に複雑でモデルをサブコンポーネントに分解する簡単な方法を提供しないことがあります。このようなライブラリでは、通常、プリントステートメントを検証することに依存します。
-
-どの戦略を選択しても、推奨される手順は通常同じで、最初のレイヤーからデバッグを開始し、最後のレイヤーからデバッグを行うべきです。
-
-通常、以下の順序で次のレイヤーからの出力を取得することをお勧めします：
-
-1. モデルに渡された入力IDを取得する
-2. 単語の埋め込みを取得する
-3. 最初のTransformerレイヤーの入力を取得する
-4. 最初のTransformerレイヤーの出力を取得する
-5. 次のn - 1つのTransformerレイヤーの出力を取得する
-6. BrandNewBertモデル全体の出力を取得する
-
-入力IDは整数の配列である必要があり、*例：* `input_ids = [0, 4, 4, 3, 2, 4, 1, 7, 19]` のようになります。
-
-以下のレイヤーの出力は多次元の浮動小数点配列であることが多く、次のようになることがあります：
-
-
-```
-[[
- [-0.1465, -0.6501,  0.1993,  ...,  0.1451,  0.3430,  0.6024],
- [-0.4417, -0.5920,  0.3450,  ..., -0.3062,  0.6182,  0.7132],
- [-0.5009, -0.7122,  0.4548,  ..., -0.3662,  0.6091,  0.7648],
- ...,
- [-0.5613, -0.6332,  0.4324,  ..., -0.3792,  0.7372,  0.9288],
- [-0.5416, -0.6345,  0.4180,  ..., -0.3564,  0.6992,  0.9191],
- [-0.5334, -0.6403,  0.4271,  ..., -0.3339,  0.6533,  0.8694]]],
-```
-
-🤗 Transformersに追加されるすべてのモデルは、統合テストを数回合格することが期待されており、元のモデルと🤗 Transformersで再実装されたバージョンが、0.001の精度までまったく同じ出力を提供する必要があります。
-異なるライブラリフレームワークで同じモデルを書いた場合、わずかに異なる出力を返すことが正常であるため、誤差許容値として1e-3（0.001）を受け入れています。モデルがほぼ同じ出力を返すだけでは不十分で、ほぼ同一である必要があります。そのため、🤗 Transformersバージョンの中間出力を元の*brand_new_bert*の実装の中間出力と複数回にわたって比較することになるでしょう。その際、元のリポジトリの**効率的な**デバッグ環境が非常に重要です。以下は、デバッグ環境をできるだけ効率的にするためのアドバイスです。
-
-- 中間結果をデバッグする最適な方法を見つける。元のリポジトリはPyTorchで書かれていますか？その場合、元のモデルをより小さなサブコンポーネントに分解して中間値を取得する長いスクリプトを書くことがおそらく適切です。元のリポジトリがTensorflow 1で書かれている場合、[tf.print](https://www.tensorflow.org/api_docs/python/tf/print)などのTensorFlowのプリント操作を使用して中間値を出力する必要があるかもしれません。元のリポジトリがJaxで書かれている場合、フォワードパスの実行時にモデルが**jittedされていない**ことを確認してください。例：[このリンク](https://github.com/google/jax/issues/196)をチェック。
-- 使用可能な最小の事前学習済みチェックポイントを使用します。チェックポイントが小さいほど、デバッグサイクルが速くなります。事前学習済みモデルがフォワードパスに10秒以上かかる場合、効率的ではありません。非常に大きなチェックポイントしか利用できない場合、新しい環境でランダムに初期化されたウェイトを持つダミーモデルを作成し、それらのウェイトを🤗 Transformersバージョンのモデルと比較する方が良いかもしれません。
-- 元のリポジトリでフォワードパスを呼び出す最も簡単な方法を使用していることを確認してください。理想的には、元のリポジトリで**単一のフォワードパス**を呼び出す関数を見つけたいです。これは通常「predict」、「evaluate」、「forward」、「__call__」と呼ばれます。複数回「forward」を呼び出す関数をデバッグしたくありません。例：テキストを生成するために「autoregressive_sample」、「generate」と呼ばれる関数。
-- トークナイゼーションとモデルの「フォワード」パスを分離しようとしてください。元のリポジトリが入力文字列を入力する必要がある例を示す場合、フォワードコール内で文字列入力が入力IDに変更される場所を特定し、このポイントから開始します。これは、スクリプトを自分で書くか、入力文字列ではなく入力IDを直接入力できるように元のコードを変更する必要があるかもしれません。
-- デバッグセットアップ内のモデルがトレーニングモードではないことを確認してください。トレーニングモードでは、モデル内の複数のドロップアウトレイヤーのためにランダムな出力が生成されることがあります。デバッグ環境のフォワードパスが**決定論的**であることを確認し、ドロップアウトレイヤーが使用されないようにします。または、新しい実装が同じフレームワーク内にある場合、*transformers.utils.set_seed*を使用してください。
-
-以下のセクションでは、*brand_new_bert*についてこれを具体的にどのように行うかについての詳細/ヒントを提供します。
-
-### 5.-14. Port BrandNewBert to 🤗 Transformers
-
-次に、ついに新しいコードを🤗 Transformersに追加できます。🤗 Transformersのフォークのクローンに移動してください：
-
-```bash
-cd transformers
-```
-
-特別なケースとして、既存のモデルと完全に一致するアーキテクチャのモデルを追加する場合、
-[このセクション](#write-a-conversion-script)で説明されているように、変換スクリプトを追加するだけで済みます。
-この場合、既存のモデルの完全なモデルアーキテクチャを再利用できます。
-
-それ以外の場合、新しいモデルの生成を開始します。ここで2つの選択肢があります：
-
-- `transformers-cli add-new-model-like`を使用して既存のモデルのような新しいモデルを追加します
-- `transformers-cli add-new-model`を使用して、テンプレートから新しいモデルを追加します（モデルのタイプに応じてBERTまたはBartのように見えます）
-
-どちらの場合でも、モデルの基本情報を入力するための質問事項が表示されます。
-2番目のコマンドを実行するには、`cookiecutter`をインストールする必要があります。
-詳細については[こちら](https://github.com/huggingface/transformers/tree/main/templates/adding_a_new_model)をご覧ください。
-
-**主要な huggingface/transformers リポジトリでプルリクエストを開く**
-
-自動生成されたコードを適応し始める前に、🤗 Transformers に「作業中（WIP）」プルリクエストを開くタイミングです。
-例：「[WIP] *brand_new_bert* を追加」などです。
-これにより、ユーザーと Hugging Face チームが🤗 Transformers にモデルを統合する作業を並行して行うことができます。
-
-以下の手順を実行してください：
-
-1. メインブランチから分かりやすい名前のブランチを作成します。
-
-```bash
-git checkout -b add_brand_new_bert
-```
-
-2. 自動生成されたコードをコミットしてください:
-
-```bash
-git add .
-git commit
-```
-
-3. 現在の main ブランチにフェッチしてリベース
-
-```bash
-git fetch upstream
-git rebase upstream/main
-```
-
-4. 変更をあなたのアカウントにプッシュするには、次のコマンドを使用します：
-
-```bash
-git push -u origin a-descriptive-name-for-my-changes
-```
-
-5. 満足したら、GitHub上のフォークのウェブページに移動します。[プルリクエスト]をクリックします。将来の変更に備えて、Hugging Face チームのメンバーのGitHubハンドルをレビュアーとして追加してください。
-
-6. GitHubのプルリクエストウェブページの右側にある「ドラフトに変換」をクリックして、PRをドラフトに変更します。
-
-以下では、進捗があった場合は常に作業をコミットし、プッシュしてプルリクエストに表示されるようにしてください。さらに、定期的にメインからの最新の変更を取り込むために、次のように行うことを忘れないでください：
-
-```bash
-git fetch upstream
-git merge upstream/main
-```
-
-一般的に、モデルや実装に関する質問はPull Request (PR) で行い、PR内で議論し、解決します。
-これにより、Hugging Face チームは新しいコードをコミットする際や質問がある場合に常に通知を受けることができます。
-質問や問題が解決された際に、問題や質問が理解されやすいように、Hugging Face チームにコードを指摘することが非常に役立ちます。
-
-このためには、「Files changed」タブに移動してすべての変更を表示し、質問したい行に移動して「+」シンボルをクリックしてコメントを追加します。
-質問や問題が解決された場合は、作成されたコメントの「Resolve」ボタンをクリックできます。
-
-同様に、Hugging Face チームはコードをレビューする際にコメントを開きます。
-PR上でのほとんどの質問はGitHub上で行うことをお勧めします。
-一般的な質問に関しては、公にはあまり役立たない質問については、SlackやメールでHugging Face チームに連絡することもできます。
-
-**5. 生成されたモデルコードを"brand_new_bert"に適応させる**
-
-最初に、モデル自体に焦点を当て、トークナイザには気にしないでください。
-関連するコードは、生成されたファイル`src/transformers/models/brand_new_bert/modeling_brand_new_bert.py`および`src/transformers/models/brand_new_bert/configuration_brand_new_bert.py`で見つかるはずです。
-
-さて、ついにコーディングを始めることができます :smile:。
-`src/transformers/models/brand_new_bert/modeling_brand_new_bert.py`にある生成されたコードは、エンコーダーのみのモデルであればBERTと同じアーキテクチャを持っているか、エンコーダー-デコーダーモデルであればBARTと同じアーキテクチャを持っているはずです。
-この段階では、モデルの理論的な側面について学んだことを思い出すべきです。つまり、「このモデルはBERTまたはBARTとどのように異なるのか？」ということです。
-これらの変更を実装しますが、これは通常、セルフアテンションレイヤー、正規化レイヤーの順序などを変更することを意味します。
-再び、あなたのモデルがどのように実装されるべきかをより良く理解するために、Transformers内に既存のモデルの類似アーキテクチャを見ることが役立つことがあります。
-
-この時点では、コードが完全に正確またはクリーンである必要はありません。
-むしろ、まずは必要なコードの最初の*クリーンでない*コピー＆ペーストバージョンを
-`src/transformers/models/brand_new_bert/modeling_brand_new_bert.py`に追加し、必要なコードがすべて追加されていると感じるまで改善/修正を反復的に行うことがお勧めです。
-私たちの経験から、必要なコードの最初のバージョンを迅速に追加し、次のセクションで説明する変換スクリプトを使用してコードを繰り返し改善/修正する方が効率的であることが多いです。
-この時点で動作する必要があるのは、🤗 Transformersの"brand_new_bert"の実装をインスタンス化できることだけです。つまり、以下のコマンドが機能する必要があります：
-
-```python
-from transformers import BrandNewBertModel, BrandNewBertConfig
-
-model = BrandNewBertModel(BrandNewBertConfig())
-```
-
-上記のコマンドは、`BrandNewBertConfig()` で定義されたデフォルトパラメータに従ってモデルを作成し、
-すべてのコンポーネントの `init()` メソッドが正常に動作することを確認します。
-
-すべてのランダムな初期化は、`BrandnewBertPreTrainedModel` クラスの `_init_weights` メソッドで行う必要があります。
-このメソッドは、設定変数に依存するすべてのリーフモジュールを初期化する必要があります。以下は、BERT の `_init_weights` メソッドの例です：
-
-```py
-def _init_weights(self, module):
-    """Initialize the weights"""
-    if isinstance(module, nn.Linear):
-        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-        if module.bias is not None:
-            module.bias.data.zero_()
-    elif isinstance(module, nn.Embedding):
-        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-        if module.padding_idx is not None:
-            module.weight.data[module.padding_idx].zero_()
-    elif isinstance(module, nn.LayerNorm):
-        module.bias.data.zero_()
-        module.weight.data.fill_(1.0)
-```
-
-特定のモジュールに特別な初期化が必要な場合、カスタムスキームをさらに持つことができます。たとえば、
-`Wav2Vec2ForPreTraining`では、最後の2つの線形層には通常のPyTorchの`nn.Linear`の初期化が必要ですが、
-他のすべての層は上記のような初期化を使用する必要があります。これは以下のようにコーディングされています：
-
-```py
-def _init_weights(self, module):
-    """Initialize the weights"""
-    if isinstnace(module, Wav2Vec2ForPreTraining):
-        module.project_hid.reset_parameters()
-        module.project_q.reset_parameters()
-        module.project_hid._is_hf_initialized = True
-        module.project_q._is_hf_initialized = True
-    elif isinstance(module, nn.Linear):
-        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-        if module.bias is not None:
-            module.bias.data.zero_()
-```
-
-`_is_hf_initialized`フラグは、サブモジュールを一度だけ初期化することを確実にするために内部で使用されます。
-`module.project_q`と`module.project_hid`のためにそれを`True`に設定することで、
-カスタム初期化が後で上書きされないようにし、`_init_weights`関数がそれらに適用されないようにします。
-
-**6. 変換スクリプトを書く**
-
-次に、*brand_new_bert* の元のリポジトリでデバッグに使用したチェックポイントを、新しく作成した 🤗 Transformers 実装の *brand_new_bert* と互換性のあるチェックポイントに変換できる変換スクリプトを書く必要があります。
-変換スクリプトをゼロから書くことはお勧めされませんが、代わりに 🤗 Transformers で既に存在する類似のモデルを同じフレームワークで変換したスクリプトを調べることが良いでしょう。
-通常、既存の変換スクリプトをコピーして、自分のユースケースにわずかに適応させることで十分です。
-Hugging Face チームに既存のモデルに類似した変換スクリプトを教えてもらうことも躊躇しないでください。
-
-- TensorFlowからPyTorchにモデルを移植している場合、良い出発点はBERTの変換スクリプトかもしれません [here](https://github.com/huggingface/transformers/blob/7acfa95afb8194f8f9c1f4d2c6028224dbed35a2/src/transformers/models/bert/modeling_bert.py#L91)
-- PyTorchからPyTorchにモデルを移植している場合、良い出発点はBARTの変換スクリプトかもしれません [here](https://github.com/huggingface/transformers/blob/main/src/transformers/models/bart/convert_bart_original_pytorch_checkpoint_to_pytorch.py)
-
-以下では、PyTorchモデルが層の重みをどのように保存し、層の名前を定義するかについて簡単に説明します。
-PyTorchでは、層の名前は層に与えるクラス属性の名前によって定義されます。
-PyTorchで `SimpleModel` というダミーモデルを定義しましょう：
-
-```python
-from torch import nn
-
-
-class SimpleModel(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.dense = nn.Linear(10, 10)
-        self.intermediate = nn.Linear(10, 10)
-        self.layer_norm = nn.LayerNorm(10)
-```
-
-これで、このモデル定義のインスタンスを作成し、`dense`、`intermediate`、`layer_norm`のすべての重みをランダムな重みで埋めたモデルを作成できます。モデルのアーキテクチャを確認するために、モデルを印刷してみましょう。
-
-```python
-model = SimpleModel()
-
-print(model)
-```
-これは以下を出力します：
-
-```
-SimpleModel(
-  (dense): Linear(in_features=10, out_features=10, bias=True)
-  (intermediate): Linear(in_features=10, out_features=10, bias=True)
-  (layer_norm): LayerNorm((10,), eps=1e-05, elementwise_affine=True)
-)
-```
-
-層の名前はPyTorchのクラス属性の名前によって定義されています。特定の層の重み値を出力することができます：
-
-
-```python
-print(model.dense.weight.data)
-```
-
-ランダムに初期化された重みを確認するために
-
-```
-tensor([[-0.0818,  0.2207, -0.0749, -0.0030,  0.0045, -0.1569, -0.1598,  0.0212,
-         -0.2077,  0.2157],
-        [ 0.1044,  0.0201,  0.0990,  0.2482,  0.3116,  0.2509,  0.2866, -0.2190,
-          0.2166, -0.0212],
-        [-0.2000,  0.1107, -0.1999, -0.3119,  0.1559,  0.0993,  0.1776, -0.1950,
-         -0.1023, -0.0447],
-        [-0.0888, -0.1092,  0.2281,  0.0336,  0.1817, -0.0115,  0.2096,  0.1415,
-         -0.1876, -0.2467],
-        [ 0.2208, -0.2352, -0.1426, -0.2636, -0.2889, -0.2061, -0.2849, -0.0465,
-          0.2577,  0.0402],
-        [ 0.1502,  0.2465,  0.2566,  0.0693,  0.2352, -0.0530,  0.1859, -0.0604,
-          0.2132,  0.1680],
-        [ 0.1733, -0.2407, -0.1721,  0.1484,  0.0358, -0.0633, -0.0721, -0.0090,
-          0.2707, -0.2509],
-        [-0.1173,  0.1561,  0.2945,  0.0595, -0.1996,  0.2988, -0.0802,  0.0407,
-          0.1829, -0.1568],
-        [-0.1164, -0.2228, -0.0403,  0.0428,  0.1339,  0.0047,  0.1967,  0.2923,
-          0.0333, -0.0536],
-        [-0.1492, -0.1616,  0.1057,  0.1950, -0.2807, -0.2710, -0.1586,  0.0739,
-          0.2220,  0.2358]]).
-```
-
-スクリプト内の変換スクリプトでは、ランダムに初期化された重みを、対応するチェックポイント内の正確な重みで埋める必要があります。例えば、以下のように翻訳します：
-
- 
-```python
-# retrieve matching layer weights, e.g. by
-# recursive algorithm
-layer_name = "dense"
-pretrained_weight = array_of_dense_layer
-
-model_pointer = getattr(model, "dense")
-
-model_pointer.weight.data = torch.from_numpy(pretrained_weight)
-```
-
-PyTorchモデルの各ランダム初期化された重みと対応する事前学習済みチェックポイントの重みが
-**形状と名前の両方**で正確に一致することを確認する必要があります。
-これを行うために、形状に対するassertステートメントを追加し、チェックポイントの重みの名前を出力することが
-**必要不可欠**です。例えば、次のようなステートメントを追加する必要があります：
-
-
-```python
-assert (
-    model_pointer.weight.shape == pretrained_weight.shape
-), f"Pointer shape of random weight {model_pointer.shape} and array shape of checkpoint weight {pretrained_weight.shape} mismatched"
-```
-
-また、両方の重みの名前を印刷して、一致していることを確認する必要があります。例えば、次のようにします：
-
-```python
-logger.info(f"Initialize PyTorch weight {layer_name} from {pretrained_weight.name}")
-```
-
-もし形状または名前のいずれかが一致しない場合、おそらく誤って🤗 Transformersの実装に初期化されたレイヤーに間違ったチェックポイントの重みを割り当ててしまった可能性があります。
-
-誤った形状は、おそらく`BrandNewBertConfig()`での設定パラメーターが、変換したいチェックポイントで使用されたものと正確に一致しないためです。
-ただし、PyTorchのレイヤーの実装によっては、重みを事前に転置する必要がある場合もあります。
-
-最後に、**すべて**の必要な重みが初期化されていることを確認し、初期化に使用されなかったすべてのチェックポイントの重みを表示して、モデルが正しく変換されていることを確認してください。
-変換トライアルが誤った形状ステートメントまたは誤った名前割り当てで失敗するのは完全に正常です。
-これはおそらく、`BrandNewBertConfig()`で誤ったパラメーターを使用したか、🤗 Transformersの実装に誤ったアーキテクチャがあるか、🤗 Transformersの実装の1つのコンポーネントの`init()`関数にバグがあるか、チェックポイントの重みの1つを転置する必要があるためです。
-
-このステップは、以前のステップと繰り返すべきです。すべてのチェックポイントの重みが正しく🤗 Transformersモデルに読み込まれるまで繰り返すべきです。
-🤗 Transformers実装に正しくチェックポイントを読み込んだ後、選択したフォルダーにモデルを保存できます `/path/to/converted/checkpoint/folder`。このフォルダには`pytorch_model.bin`ファイルと`config.json`ファイルの両方が含まれるはずです。
-
-
-```python
-model.save_pretrained("/path/to/converted/checkpoint/folder")
-```
-
-**7. 順伝播（forward pass）の実装**
-
-🤗 Transformers実装で事前学習済みの重みを正しく読み込んだ後、順伝播が正しく実装されていることを確認する必要があります。[元のリポジトリを理解する](#34-run-a-pretrained-checkpoint-using-the-original-repository)で、元のリポジトリを使用してモデルの順伝播を実行するスクリプトをすでに作成しました。今度は、元のリポジトリの代わりに🤗 Transformers実装を使用して類似のスクリプトを作成する必要があります。以下のようになります：
-
-```python
-model = BrandNewBertModel.from_pretrained("/path/to/converted/checkpoint/folder")
-input_ids = [0, 4, 4, 3, 2, 4, 1, 7, 19]
-output = model(input_ids).last_hidden_states
-```
-
-🤗 Transformersの実装と元のモデルの実装が最初の実行で完全に同じ出力を提供しないか、
-フォワードパスでエラーが発生する可能性が非常に高いです。失望しないでください - これは予想されていることです！
-まず、フォワードパスがエラーをスローしないことを確認する必要があります。
-間違った次元が使用され、*次元の不一致*エラーや、誤ったデータ型オブジェクトが使用されることがよくあります。
-例えば、`torch.long`ではなく`torch.float32`が使用されます。特定のエラーを解決できない場合は、
-Hugging Faceチームに助けを求めることを躊躇しないでください。
-
-🤗 Transformers実装が正しく機能することを確認する最終的な部分は、出力が`1e-3`の精度で同等であることを確認することです。
-まず、出力の形状が同一であること、つまりスクリプトの🤗 Transformers実装と元の実装の両方で`outputs.shape`が同じ値を生成する必要があります。
-次に、出力値が同一であることを確認する必要があります。
-これは新しいモデルを追加する際の最も難しい部分の1つです。
-出力が同一でない理由の一般的な間違いは以下の通りです。
-
-- 一部のレイヤーが追加されていない、つまり*活性化*レイヤーが追加されていないか、リザバル接続が忘れられている
-- 単語埋め込み行列が結ばれていない
-- オリジナルの実装がオフセットを使用しているため、誤った位置埋め込みが使用されている
-- フォワードパス中にドロップアウトが適用されています。これを修正するには、*model.trainingがFalse*であることを確認し、フォワードパス中に誤ってドロップアウトレイヤーがアクティブ化されないようにします。
-*つまり* [PyTorchのfunctional dropout](https://pytorch.org/docs/stable/nn.functional.html?highlight=dropout#torch.nn.functional.dropout)に*model.training*を渡します。
-
-問題を修正する最良の方法は、通常、元の実装と🤗 Transformers実装のフォワードパスを並べて表示し、違いがあるかどうかを確認することです。
-理想的には、フォワードパスの両方の実装の中間出力をデバッグ/プリントアウトして、🤗 Transformers実装が元の実装と異なる出力を示すネットワーク内の正確な位置を見つけることができます。
-最初に、両方のスクリプトのハードコーディングされた`input_ids`が同一であることを確認します。
-次に、`input_ids`の最初の変換（通常、単語埋め込み）の出力が同一であることを確認します。
-その後、ネットワークの最後のレイヤーまで作業を進めます。
-いずれかの時点で、2つの実装間で違いがあることに気付くはずで、それにより🤗 Transformers実装のバグの場所が特定されます。
-経験上、元の実装と🤗 Transformers実装のフォワードパスの同じ位置に多くのプリントステートメントを追加し、
-中間プレゼンテーションで同じ値を示すプリントステートメントを段階的に削除するのがシンプルかつ効果的な方法です。
-
-両方の実装が同じ出力を生成することに自信を持っている場合、`torch.allclose(original_output, output, atol=1e-3)`を使用して出力を確認すると、最も難しい部分が完了します！
-おめでとうございます - 完了する作業は簡単なものになるはずです 😊。
-
-**8. 必要なすべてのモデルテストを追加**
-
-この時点で、新しいモデルが正常に追加されました。
-ただし、モデルがまだ必要な設計に完全に準拠していない可能性が非常に高いです。
-🤗 Transformersと完全に互換性があることを確認するために、すべての一般的なテストがパスする必要があります。
-Cookiecutterはおそらくモデル用のテストファイルを自動的に追加しているはずで、おそらく同じディレクトリに`tests/models/brand_new_bert/test_modeling_brand_new_bert.py`として存在します。
-このテストファイルを実行して、すべての一般的なテストがパスすることを確認してください：
-
-```bash
-pytest tests/models/brand_new_bert/test_modeling_brand_new_bert.py
-```
-
-すべての一般的なテストを修正したら、今度は実行したすべての素晴らしい作業が適切にテストされていることを確認することが非常に重要です。これにより、
-
-- a) コミュニティは*brand_new_bert*の特定のテストを見ることで、あなたの作業を簡単に理解できます。
-- b) モデルへの将来の変更がモデルの重要な機能を壊さないようにすることができます。
-
-まず、統合テストを追加する必要があります。これらの統合テストは、基本的にはデバッグスクリプトと同じことを行います。これらのモデルテストのテンプレートはCookiecutterによって既に追加されており、「BrandNewBertModelIntegrationTests」と呼ばれています。このテストを記入するだけです。これらのテストが合格していることを確認するには、次のコマンドを実行します。
-
-```bash
-RUN_SLOW=1 pytest -sv tests/models/brand_new_bert/test_modeling_brand_new_bert.py::BrandNewBertModelIntegrationTests
-```
-
-<Tip>
-
-Windowsを使用している場合、`RUN_SLOW=1`を`SET RUN_SLOW=1`に置き換えてください。
-
-</Tip>
-
-次に、*brand_new_bert*に特有のすべての特徴は、別個のテスト内で追加されるべきです。
-`BrandNewBertModelTester`/`BrandNewBertModelTest`の下に。この部分はよく忘れられますが、2つの点で非常に役立ちます：
-
-- モデルの追加中に獲得した知識をコミュニティに伝え、*brand_new_bert*の特別な機能がどのように動作するかを示すことによって、知識の共有を支援します。
-- 将来の貢献者は、これらの特別なテストを実行することでモデルへの変更を迅速にテストできます。
-
-**9. トークナイザの実装**
-
-次に、*brand_new_bert*のトークナイザを追加する必要があります。通常、トークナイザは🤗 Transformersの既存のトークナイザと同等か非常に似ています。
-
-トークナイザが正しく動作することを確認するためには、まず、元のリポジトリ内で文字列を入力し、`input_ids`を返すスクリプトを作成することをお勧めします。
-このスクリプトは、次のように見えるかもしれません（疑似コードで示します）：
-
-```python
-input_str = "This is a long example input string containing special characters .$?-, numbers 2872 234 12 and words."
-model = BrandNewBertModel.load_pretrained_checkpoint("/path/to/checkpoint/")
-input_ids = model.tokenize(input_str)
-```
-
-オリジナルのリポジトリを詳しく調査し、正しいトークナイザの関数を見つける必要があるかもしれません。
-または、オリジナルのリポジトリのクローンを変更して、`input_ids`だけを出力するようにする必要があるかもしれません。
-オリジナルのリポジトリを使用した機能的なトークナイゼーションスクリプトを作成した後、
-🤗 Transformers向けの類似したスクリプトを作成する必要があります。
-以下のように見えるべきです：
-
-```python
-from transformers import BrandNewBertTokenizer
-
-input_str = "This is a long example input string containing special characters .$?-, numbers 2872 234 12 and words."
-
-tokenizer = BrandNewBertTokenizer.from_pretrained("/path/to/tokenizer/folder/")
-
-input_ids = tokenizer(input_str).input_ids
-```
-
-`input_ids`が同じ値を生成した場合、最終ステップとしてトークナイザのテストファイルも追加するべきです。
-
-*brand_new_bert*のモデルングテストファイルと同様に、*brand_new_bert*のトークナイズテストファイルには、いくつかのハードコードされた統合テストが含まれるべきです。
-
-**10. エンドツーエンド統合テストの実行**
-
-トークナイザを追加した後、`🤗 Transformers`内の`tests/models/brand_new_bert/test_modeling_brand_new_bert.py`に
-モデルとトークナイザの両方を使用するいくつかのエンドツーエンド統合テストも追加する必要があります。
-このようなテストは、🤗 Transformersの実装が期待どおりに機能することを示すべきです。
-意味のあるテキスト対テキストのサンプルが含まれます。有用なテキスト対テキストのサンプルには、ソースからターゲットへの翻訳ペア、記事から要約へのペア、質問から回答へのペアなどが含まれます。
-ポートされたチェックポイントがダウンストリームタスクでファインチューニングされていない場合、モデルのテストに依存するだけで十分です。
-モデルが完全に機能していることを確認するために、すべてのテストをGPU上で実行することもお勧めします。
-モデルの内部テンソルに`.to(self.device)`ステートメントを追加するのを忘れる可能性があるため、そのようなテストではエラーが表示されることがあります。
-GPUにアクセスできない場合、Hugging Faceチームが代わりにこれらのテストを実行できます。
-
-**11. ドキュメントの追加**
-
-これで、*brand_new_bert*の必要なすべての機能が追加されました - ほぼ完了です！残りの追加すべきことは、良いドキュメントとドキュメントページです。
-Cookiecutterが`docs/source/model_doc/brand_new_bert.md`というテンプレートファイルを追加しているはずで、これを記入する必要があります。
-モデルのユーザーは通常、モデルを使用する前にまずこのページを見ます。したがって、ドキュメンテーションは理解しやすく簡潔である必要があります。
-モデルの使用方法を示すためにいくつかの*Tips*を追加することはコミュニティにとって非常に役立ちます。ドキュメンテーションに関しては、Hugging Faceチームに問い合わせることをためらわないでください。
-
-次に、`src/transformers/models/brand_new_bert/modeling_brand_new_bert.py`に追加されたドキュメンテーション文字列が正しいこと、およびすべての必要な入力および出力を含んでいることを確認してください。
-ドキュメンテーションの書き方とドキュメンテーション文字列のフォーマットについて詳細なガイドが[こちら](writing-documentation)にあります。
-ドキュメンテーションは通常、コミュニティとモデルの最初の接触点であるため、コードと同じくらい注意深く扱うべきであることを常に念頭に置いてください。
-
-**コードのリファクタリング**
-
-素晴らしい、これで*brand_new_bert*に必要なすべてのコードが追加されました。
-この時点で、次のようなポテンシャルなコードスタイルの誤りを訂正するために以下を実行する必要があります：
-
-```bash
-make style
-```
-
-あなたのコーディングスタイルが品質チェックをパスすることを確認してください:
-
-```bash
-make quality
-```
-
-🤗 Transformersの非常に厳格なデザインテストには、まだ合格していない可能性があるいくつかの他のテストが存在するかもしれません。
-これは、ドキュメント文字列に情報が不足しているか、名前が間違っていることが原因であることが多いです。Hugging Faceチームは、ここで詰まっている場合には必ず助けてくれるでしょう。
-
-最後に、コードが正しく機能することを確認した後、コードをリファクタリングするのは常に良いアイデアです。
-すべてのテストがパスした今、追加したコードを再度確認してリファクタリングを行うのは良いタイミングです。
-
-これでコーディングの部分は完了しました、おめでとうございます！ 🎉 あなたは素晴らしいです！ 😎
-
-**12. モデルをモデルハブにアップロード**
-
-最後のパートでは、すべてのチェックポイントをモデルハブに変換してアップロードし、各アップロードしたモデルチェックポイントにモデルカードを追加する必要があります。
-モデルハブの機能について詳しくは、[Model sharing and uploading Page](model_sharing)を読んで理解できます。
-ここでは、*brand_new_bert*の著者組織の下にモデルをアップロードできるように必要なアクセス権を取得するために、Hugging Faceチームと協力する必要があります。
-`transformers`のすべてのモデルに存在する`push_to_hub`メソッドは、チェックポイントをハブにプッシュする迅速かつ効率的な方法です。
-以下に、少しのコードスニペットを示します：
-
-```python
-brand_new_bert.push_to_hub("brand_new_bert")
-# Uncomment the following line to push to an organization.
-# brand_new_bert.push_to_hub("<organization>/brand_new_bert")
-```
-
-各チェックポイントに適切なモデルカードを作成する価値があります。モデルカードは、この特定のチェックポイントの特性をハイライトするべきです。例えば、このチェックポイントはどのデータセットで事前学習/ファインチューニングされたか、どのような下流タスクでモデルを使用すべきかを示すべきです。また、モデルの正しい使用方法に関するコードも含めるべきです。
-
-**13.（オプション）ノートブックの追加**
-
-*brand_new_bert*を推論または下流タスクのファインチューニングにどのように詳細に使用できるかを示すノートブックを追加することは非常に役立ちます。これはあなたのPRをマージするために必須ではありませんが、コミュニティにとって非常に有用です。
-
-**14. 完成したPRの提出**
-
-プログラミングが完了したら、最後のステップに移動し、PRをメインブランチにマージしましょう。通常、Hugging Faceチームはこの時点で既にあなたをサポートしているはずですが、PRに良い説明を追加し、コードにコメントを追加して、レビュアーに特定の設計の選択肢を指摘したい場合はコメントを追加することも価値があります。
-
-### Share your work!!
-
-さあ、コミュニティからあなたの作業に対する評価を得る時が来ました！モデルの追加を完了することは、TransformersおよびNLPコミュニティにとって重要な貢献です。あなたのコードとポートされた事前学習済みモデルは、何百人、何千人という開発者や研究者によって確実に使用されるでしょう。あなたの仕事に誇りを持ち、コミュニティとあなたの成果を共有しましょう。
-
-**あなたはコミュニティの誰でも簡単にアクセスできる別のモデルを作成しました！ 🤯**
-
-

docs/source/ja/add_tensorflow_model.md

@@ -1,296 +0,0 @@
-<!--
-Copyright 2023 The HuggingFace Team. All rights reserved.
-
-ライセンス：Apache License、バージョン2.0（「ライセンス」）に基づいています。このファイルは、ライセンスに準拠していない限り、使用できません。ライセンスのコピーは以下から入手できます：
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-適用法に従って必要な場合、または書面で同意した場合を除き、ライセンスの下で配布されるソフトウェアは、「AS IS」の基盤で、明示または黙示を問わず、いかなる保証や条件も含みません。ライセンスの詳細については、ライセンス文書をご覧ください。
-
-⚠️ このファイルはMarkdown形式ですが、弊社のドキュメントビルダー（MDXに類似した特定の構文を含む）を含むため、お使いのMarkdownビューアでは正しく表示されない場合があります。
-
--->
-
-
-# How to convert a 🤗 Transformers model to TensorFlow?
-
-🤗 Transformersを使用するために複数のフレームワークが利用可能であることは、アプリケーションを設計する際にそれぞれの強みを活かす柔軟性を提供しますが、
-互換性をモデルごとに追加する必要があることを意味します。しかし、幸いなことに
-既存のモデルにTensorFlow互換性を追加することは、[ゼロから新しいモデルを追加すること](add_new_model)よりも簡単です！
-大規模なTensorFlowモデルの詳細を理解したり、主要なオープンソースの貢献を行ったり、
-選択したモデルをTensorFlowで有効にするためのガイドです。
-
-このガイドは、コミュニティのメンバーであるあなたに、TensorFlowモデルの重みおよび/または
-アーキテクチャを🤗 Transformersで使用するために、Hugging Faceチームからの最小限の監視で貢献できる力を与えます。新しいモデルを書くことは小さな偉業ではありませんが、
-このガイドを読むことで、それがローラーコースターのようなものから散歩のようなものになることを願っています🎢🚶。
-このプロセスをますます簡単にするために、私たちの共通の経験を活用することは非常に重要ですので、
-このガイドの改善を提案することを強くお勧めします！
-
-さらに詳しく調べる前に、以下のリソースをチェックすることをお勧めします。🤗 Transformersが初めての場合：
-
-- [🤗 Transformersの一般的な概要](add_new_model#general-overview-of-transformers)
-- [Hugging FaceのTensorFlow哲学](https://huggingface.co/blog/tensorflow-philosophy)
-
-このガイドの残りの部分では、新しいTensorFlowモデルアーキテクチャを追加するために必要なもの、
-PyTorchをTensorFlowモデルの重みに変換する手順、およびMLフレームワーク間の不一致を効率的にデバッグする方法について学びます。それでは始めましょう！
-
-<Tip>
-
-使用したいモデルに対応するTensorFlowアーキテクチャがすでに存在するかどうかわからないですか？
-
-&nbsp;
-
-選択したモデルの`config.json`の`model_type`フィールドをチェックしてみてください
-（[例](https://huggingface.co/bert-base-uncased/blob/main/config.json#L14)）。
-🤗 Transformersの該当するモデルフォルダに、名前が"modeling_tf"で始まるファイルがある場合、それは対応するTensorFlow
-アーキテクチャを持っていることを意味します（[例](https://github.com/huggingface/transformers/tree/main/src/transformers/models/bert)）。
-
-</Tip>
-
-## Step-by-step guide to add TensorFlow model architecture code
-
-大規模なモデルアーキテクチャを設計する方法はさまざまであり、その設計を実装する方法もさまざまです。
-しかし、[🤗 Transformersの一般的な概要](add_new_model#general-overview-of-transformers)から
-思い出していただけるかもしれませんが、私たちは意見のあるグループです - 🤗 Transformersの使いやすさは一貫性のある設計の選択肢に依存しています。経験から、TensorFlowモデルを追加する際に重要なことをいくつかお伝えできます：
-
-- 車輪を再発明しないでください！ほとんどの場合、確認すべき少なくとも2つの参照実装があります。それは、
-あなたが実装しているモデルのPyTorchバージョンと、同じ種類の問題に対する他のTensorFlowモデルです。
-- 優れたモデル実装は時間の試練を乗り越えます。これは、コードがきれいだからではなく、コードが明確で、デバッグしやすく、
-構築しやすいからです。TensorFlow実装でPyTorch実装と一致するパターンを複製し、PyTorch実装との不一致を最小限に抑えることで、
-あなたの貢献が長期間にわたって有用であることを保証します。
-- 行き詰まったら助けを求めてください！ 🤗 Transformersチームはここにいますし、おそらくあなたが直面している同じ問題に対する解決策を見つけています。
-
-TensorFlowモデルアーキテクチャを追加するために必要なステップの概要は次のとおりです：
-1. 変換したいモデルを選択
-2. transformersの開発環境を準備
-3. （オプション）理論的な側面と既存の実装を理解
-4. モデルアーキテクチャを実装
-5. モデルのテストを実装
-6. プルリクエストを提出
-7. （オプション）デモを構築して世界と共有
-
-### 1.-3. Prepare your model contribution
-
-**1. 変換したいモデルを選択する**
-
-まず、基本から始めましょう。最初に知っておく必要があることは、変換したいアーキテクチャです。
-特定のアーキテクチャを決めていない場合、🤗 Transformers チームに提案を求めることは、影響を最大限にする素晴らしい方法です。
-チームは、TensorFlow サイドで不足している最も注目されるアーキテクチャに向けてガイドします。
-TensorFlow で使用したい特定のモデルに、🤗 Transformers に既に TensorFlow アーキテクチャの実装が存在しているが、重みが不足している場合、
-このページの[重みの追加セクション](#adding-tensorflow-weights-to-hub)に直接移動してください。
-
-簡単にするために、このガイドの残りの部分では、TensorFlow バージョンの *BrandNewBert* を貢献することを決定したと仮定しています
-（これは、[新しいモデルの追加ガイド](add_new_model)での例と同じです）。
-
-<Tip>
-
-TensorFlow モデルのアーキテクチャに取り組む前に、それを行うための進行中の取り組みがないかを再確認してください。
-GitHub ページの[プルリクエスト](https://github.com/huggingface/transformers/pulls?q=is%3Apr)で `BrandNewBert` を検索して、
-TensorFlow 関連のプルリクエストがないことを確認できます。
-
-</Tip>
-
-
-**2. transformers 開発環境の準備**
-
-モデルアーキテクチャを選択したら、意向を示すためにドラフト PR を開くための環境を設定してください。
-以下の手順に従って、環境を設定し、ドラフト PR を開いてください。
-
-1. リポジトリのページで 'Fork' ボタンをクリックして、[リポジトリ](https://github.com/huggingface/transformers)をフォークします。
-   これにより、コードのコピーが GitHub ユーザーアカウントの下に作成されます。
-
-2. ローカルディスクにある 'transformers' フォークをクローンし、ベースリポジトリをリモートとして追加します:
-
-```bash
-git clone https://github.com/[your Github handle]/transformers.git
-cd transformers
-git remote add upstream https://github.com/huggingface/transformers.git
-```
-
-3. 開発環境を設定します。たとえば、以下のコマンドを実行してください：
-
-```bash
-git clone https://github.com/[your Github handle]/transformers.git
-cd transformers
-git remote add upstream https://github.com/huggingface/transformers.git
-```
-
-依存関係が増えているため、OSに応じて、Transformersのオプションの依存関係の数が増えるかもしれません。その場合は、TensorFlowをインストールしてから次のコマンドを実行してください。
-
-```bash
-pip install -e ".[quality]"
-```
-
-**注意:** CUDAをインストールする必要はありません。新しいモデルをCPUで動作させることが十分です。
-
-4. メインブランチからわかりやすい名前のブランチを作成してください。
-
-```bash
-git checkout -b add_tf_brand_new_bert
-```
-5. 現在のmainブランチにフェッチしてリベースする
-
-```bash
-git fetch upstream
-git rebase upstream/main
-```
-
-6. `transformers/src/models/brandnewbert/`に`modeling_tf_brandnewbert.py`という名前の空の`.py`ファイルを追加します。これはあなたのTensorFlowモデルファイルです。
-
-7. 以下を使用して変更内容をアカウントにプッシュします：
-
-```bash
-git add .
-git commit -m "initial commit"
-git push -u origin add_tf_brand_new_bert
-```
-
-8. GitHub上でフォークしたウェブページに移動し、「プルリクエスト」をクリックします。将来の変更に備えて、Hugging Face チームのメンバーのGitHubハンドルをレビュアーとして追加してください。
-
-9. GitHubのプルリクエストウェブページの右側にある「ドラフトに変換」をクリックして、プルリクエストをドラフトに変更します。
-
-これで、🤗 Transformers内に*BrandNewBert*をTensorFlowに移植するための開発環境が設定されました。
-
-**3. (任意) 理論的な側面と既存の実装を理解する**
-
-*BrandNewBert*の論文が存在する場合、その記述的な作業を読む時間を取るべきです。論文には理解が難しい大きなセクションがあるかもしれません。その場合でも問題ありません - 心配しないでください！目標は論文の理論的な理解を深めることではなく、🤗 Transformersを使用してTensorFlowでモデルを効果的に再実装するために必要な情報を抽出することです。とは言え、理論的な側面にあまり時間をかける必要はありません。代わりに、既存のモデルのドキュメンテーションページ（たとえば、[BERTのモデルドキュメント](model_doc/bert)など）に焦点を当てるべきです。
-
-実装するモデルの基本を把握した後、既存の実装を理解することは重要です。これは、動作する実装がモデルに対する期待と一致することを確認する絶好の機会であり、TensorFlow側での技術的な課題を予測することもできます。
-
-情報の多さに圧倒されていると感じるのは完全に自然です。この段階ではモデルのすべての側面を理解する必要はありません。ただし、[フォーラム](https://discuss.huggingface.co/)で急な質問を解決することを強くお勧めします。
-
-
-### 4. Model implementation
-
-さあ、いよいよコーディングを始めましょう。お勧めする出発点は、PyTorchファイルそのものです。
-`src/transformers/models/brand_new_bert/`内の`modeling_brand_new_bert.py`の内容を
-`modeling_tf_brand_new_bert.py`にコピーします。このセクションの目標は、
-🤗 Transformersのインポート構造を更新し、`TFBrandNewBert`と
-`TFBrandNewBert.from_pretrained(model_repo, from_pt=True)`を正常に読み込む動作するTensorFlow *BrandNewBert*モデルを
-インポートできるようにすることです。
-
-残念ながら、PyTorchモデルをTensorFlowに変換する明確な方法はありません。ただし、プロセスをできるだけスムーズにするためのヒントを以下に示します：
-
-- すべてのクラスの名前の前に `TF` を付けます（例： `BrandNewBert` は `TFBrandNewBert` になります）。
-- ほとんどのPyTorchの操作には、直接TensorFlowの代替があります。たとえば、`torch.nn.Linear` は `tf.keras.layers.Dense` に対応し、`torch.nn.Dropout` は `tf.keras.layers.Dropout` に対応します。特定の操作について不明確な場合は、[TensorFlowのドキュメント](https://www.tensorflow.org/api_docs/python/tf)または[PyTorchのドキュメント](https://pytorch.org/docs/stable/)を参照できます。
-- 🤗 Transformersのコードベースにパターンが見つかります。特定の操作に直接的な代替がない場合、誰かがすでに同じ問題に対処している可能性が高いです。
-- デフォルトでは、PyTorchと同じ変数名と構造を維持します。これにより、デバッグや問題の追跡、修正の追加が容易になります。
-- 一部のレイヤーには、各フレームワークで異なるデフォルト値があります。注目すべき例は、バッチ正規化レイヤーの epsilon です（PyTorchでは`1e-5`、[TensorFlowでは](https://www.tensorflow.org/api_docs/python/tf/keras/layers/BatchNormalization) `1e-3` です）。ドキュメントを再確認してください！
-- PyTorchの `nn.Parameter` 変数は通常、TF Layerの `build()` 内で初期化する必要があります。次の例を参照してください：[PyTorch](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_vit_mae.py#L212) / [TensorFlow](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_tf_vit_mae.py#L220)
-- PyTorchモデルに関数の上部に `#copied from ...` がある場合、TensorFlowモデルも同じアーキテクチャからその関数を借りることができる可能性が高いです。TensorFlowアーキテクチャがある場合です。
-- TensorFlow関数内で `name`属性を正しく設定することは、`from_pt=True`のウェイトのクロスロードロードを行うために重要です。通常、`name`はPyTorchコード内の対応する変数の名前です。`name`が正しく設定されていない場合、モデルウェイトのロード時にエラーメッセージで表示されます。
-- ベースモデルクラス `BrandNewBertModel` のロジックは実際には `TFBrandNewBertMainLayer` にあります。これはKerasレイヤーのサブクラスです（[例](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L719)）。`TFBrandNewBertModel` は、単にこのレイヤーのラッパーです。
-- モデルを読み込むためには、Kerasモデルをビルドする必要があります。そのため、`TFBrandNewBertPreTrainedModel` はモデルへの入力の例、`dummy_inputs` を持つ必要があります（[例](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L916)）。
-- 表示が止まった場合は、助けを求めてください。私たちはあなたのお手伝いにここにいます！ 🤗
-
-モデルファイル自体だけでなく、モデルクラスと関連するドキュメンテーションページへのポインターも追加する必要があります。他のPRのパターンに従ってこの部分を完了できます
-（[例](https://github.com/huggingface/transformers/pull/18020/files)）。
-以下は手動での変更が必要な一覧です：
-- *BrandNewBert*のすべてのパブリッククラスを `src/transformers/__init__.py` に含める
-- *BrandNewBert*クラスを `src/transformers/models/auto/modeling_tf_auto.py` の対応するAutoクラスに追加
-- ドキュメンテーションテストファイルのリストにモデリングファイルを追加する `utils/documentation_tests.txt`
-- `src/transformers/utils/dummy_tf_objects.py` に関連する *BrandNewBert* に関連する遅延ロードクラスを追加
-- `src/transformers/models/brand_new_bert/__init__.py` でパブリッククラスのインポート構造を更新
-- `docs/source/en/model_doc/brand_new_bert.md` に *BrandNewBert* のパブリックメソッドのドキュメンテーションポインターを追加
-- `docs/source/en/model_doc/brand_new_bert.md` の *BrandNewBert* の貢献者リストに自分自身を追加
-- 最後に、`docs/source/en/index.md` の *BrandNewBert* のTensorFlow列に緑色のチェックマーク ✅ を追加
-
-モデルアーキテクチャが準備できていることを確認するために、以下のチェックリストを実行してください：
-1. 訓練時に異なる動作をするすべてのレイヤー（例：Dropout）は、`training`引数を使用して呼び出され、それが最上位クラスから伝播されます。
-2. 可能な限り `#copied from ...` を使用しました
-3. `TFBrandNewBertMainLayer` およびそれを使用するすべてのクラスの `call` 関数が `@unpack_inputs` でデコレートされています
-4. `TFBrandNewBertMainLayer` は `@keras_serializable` でデコレートされています
-5. PyTorchウェイトからTensorFlowウェイトを使用してTensorFlowモデルをロードできます `TFBrandNewBert.from_pretrained(model_repo, from_pt=True)`
-6. 予期される入力形式を使用してTensorFlowモデルを呼び出すことができます
-
-
-### 5. Add model tests
-
-やったね、TensorFlowモデルを実装しました！
-今度は、モデルが期待通りに動作することを確認するためのテストを追加する時間です。
-前のセクションと同様に、`tests/models/brand_new_bert/`ディレクトリ内の`test_modeling_brand_new_bert.py`ファイルを`test_modeling_tf_brand_new_bert.py`にコピーし、必要なTensorFlowの置換を行うことをお勧めします。
-今の段階では、すべての`.from_pretrained()`呼び出しで、既存のPyTorchの重みをロードするために`from_pt=True`フラグを使用する必要があります。
-
-作業が完了したら、テストを実行する準備が整いました！ 😬
-
-```bash
-NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
-py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
-```
-
-最も可能性の高い結果は、多くのエラーが表示されることです。心配しないでください、これは予想される動作です！
-MLモデルのデバッグは非常に難しいとされており、成功の鍵は忍耐力（と`breakpoint()`）です。私たちの経験では、
-最も難しい問題はMLフレームワーク間の微妙な不一致から発生し、これについてはこのガイドの最後にいくつかのポインタを示します。
-他の場合では、一般的なテストが直接モデルに適用できない場合もあり、その場合はモデルのテストクラスレベルでオーバーライドを提案します。
-問題の種類に関係なく、詰まった場合は、ドラフトのプルリクエストで助けを求めることをためらわないでください。
-
-すべてのテストがパスしたら、おめでとうございます。あなたのモデルはほぼ🤗 Transformersライブラリに追加する準備が整いました！🎉
-
-**6. プルリクエストを提出する**
-
-実装とテストが完了したら、プルリクエストを提出する準備が整いました。コードをプッシュする前に、
-コードフォーマットユーティリティである `make fixup` 🪄 を実行してください。
-これにより、自動的なチェックに失敗する可能性のあるフォーマットの問題が自動的に修正されます。
-
-これで、ドラフトプルリクエストを実際のプルリクエストに変換する準備が整いました。
-これを行うには、「レビュー待ち」ボタンをクリックし、Joao（`@gante`）とMatt（`@Rocketknight1`）をレビュワーとして追加します。
-モデルプルリクエストには少なくとも3人のレビュワーが必要ですが、モデルに適切な追加のレビュワーを見つけるのは彼らの責任です。
-
-すべてのレビュワーがプルリクエストの状態に満足したら、最後のアクションポイントは、`.from_pretrained()` 呼び出しで `from_pt=True` フラグを削除することです。
-TensorFlowのウェイトが存在しないため、それらを追加する必要があります！これを行う方法については、以下のセクションを確認してください。
-
-最後に、TensorFlowのウェイトがマージされ、少なくとも3人のレビューアが承認し、すべてのCIチェックが
-成功した場合、テストをローカルで最後にもう一度確認してください。
-
-```bash
-NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
-py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
-```
-
-そして、あなたのPRをマージします！マイルストーン達成おめでとうございます 🎉
-
-**7. (Optional) デモを作成して世界と共有**
-
-オープンソースの最も難しい部分の1つは、発見です。あなたの素晴らしいTensorFlowの貢献が存在することを他のユーザーがどのように知ることができるでしょうか？適切なコミュニケーションです！ 📣
-
-コミュニティとモデルを共有する主要な方法は2つあります。
-- デモを作成します。これにはGradioデモ、ノートブック、およびモデルを紹介するための他の楽しい方法が含まれます。[コミュニティ駆動のデモ](https://huggingface.co/docs/transformers/community)にノートブックを追加することを強くお勧めします。
-- TwitterやLinkedInなどのソーシャルメディアでストーリーを共有します。あなたの仕事に誇りを持ち、コミュニティとあなたの成果を共有するべきです - あなたのモデルは今や世界中の何千人ものエンジニアや研究者によって使用される可能性があります 🌍！私たちはあなたの投稿をリツイートして共同体と共有するお手伝いを喜んでします。
-
-## Adding TensorFlow weights to 🤗 Hub
-
-TensorFlowモデルのアーキテクチャが🤗 Transformersで利用可能な場合、PyTorchの重みをTensorFlowの重みに変換することは簡単です！
-
-以下がその方法です：
-1. ターミナルでHugging Faceアカウントにログインしていることを確認してください。コマンド`huggingface-cli login`を使用してログインできます（アクセストークンは[こちら](https://huggingface.co/settings/tokens)で見つけることができます）。
-2. `transformers-cli pt-to-tf --model-name foo/bar`というコマンドを実行します。ここで、`foo/bar`は変換したいPyTorchの重みを含むモデルリポジトリの名前です。
-3. 上記のコマンドで作成された🤗 Hub PRに`@joaogante`と`@Rocketknight1`をタグ付けします。
-
-それだけです！ 🎉
-
-## Debugging mismatches across ML frameworks 🐛
-
-新しいアーキテクチャを追加したり、既存のアーキテクチャのTensorFlowの重みを作成したりする際、PyTorchとTensorFlow間の不一致についてのエラーに遭遇することがあります。
-場合によっては、PyTorchとTensorFlowのモデルアーキテクチャがほぼ同一であるにもかかわらず、不一致を指摘するエラーが表示されることがあります。
-どうしてでしょうか？ 🤔
-
-まず最初に、なぜこれらの不一致を理解することが重要かについて話しましょう。多くのコミュニティメンバーは🤗 Transformersモデルをそのまま使用し、モデルが期待どおりに動作すると信頼しています。
-2つのフレームワーク間で大きな不一致があると、少なくとも1つのフレームワークのリファレンス実装に従ってモデルが動作しないことを意味します。
-これにより、モデルは実行されますが性能が低下する可能性があり、静かな失敗が発生する可能性があります。これは、全く実行されないモデルよりも悪いと言えるかもしれません！そのため、モデルのすべての段階でのフレームワークの不一致が`1e-5`未満であることを目指しています。
-
-数値計算の問題と同様に、詳細については細かいところにあります。そして、詳細指向の技術である以上、秘密の要素は忍耐です。
-この種の問題に遭遇した場合のお勧めのワークフローは次のとおりです：
-1. 不一致の原因を特定します。変換中のモデルにはおそらく特定の点までほぼ同一の内部変数があります。
-   両方のフレームワークのアーキテクチャに`breakpoint()`ステートメントを配置し、トップダウンの方法で数値変数の値を比較し、問題の原因を見つけます。
-2. 問題の原因を特定したら、🤗 Transformersチームと連絡を取りましょう。同様の問題に遭遇したことがあるかもしれず、迅速に解決策を提供できるかもしれません。最終手段として、StackOverflowやGitHubの問題など、人気のあるページをスキャンします。
-3. 解決策が見当たらない場合、問題を掘り下げる必要があることを意味します。良いニュースは、問題の原因を特定したことです。したがって、問題のある命令に焦点を当て、モデルの残りを抽象化できます！悪いニュースは、その命令のソース実装に進む必要があることです。一部の場合では、リファレンス実装に問題があるかもしれません - 上流リポジトリで問題を開くのを控えないでください。
-
-🤗 Transformersチームとの話し合いで、不一致を修正することが困難であることが判明することがあります。
-出力レイヤーのモデルで不一致が非常に小さい場合（ただし、隠れた状態では大きい可能性がある）、モデルを配布するためにそれを無視することにするかもしれません。
-上記で言及した`pt-to-tf` CLIには、重み変換時にエラーメッセージを無視するための`--max-error`フラグがあります。
-
-
-
-
-
-

docs/source/ja/attention.md

@@ -1,52 +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.
-
-⚠️ このファイルはMarkdown形式ですが、ドキュメンテーションビルダー用の特定の構文を含んでおり、Markdownビューアーでは正しく表示されないことに注意してください。
--->
-
-# Attention mechanism
-
-ほとんどのTransformerモデルは、アテンション行列が正方形であるという意味で完全なアテンションを使用します。
-これは、長いテキストを扱う場合に計算のボトルネックとなることがあります。LongformerやReformerは、より効率的でトレーニングを高速化するためにアテンション行列のスパースバージョンを使用しようとするモデルです。
-
-## LSH attention
-
-[Reformer](#reformer)はLSH（局所的に散在ハッシュ）アテンションを使用します。
-ソフトマックス(QK^t)では、行列QK^tの中で（ソフトマックス次元で）最も大きな要素のみが有用な寄与を提供します。
-したがって、各クエリqについて、クエリqに近いキーkのみを考慮できます。
-qとkが近いかどうかを決定するために、ハッシュ関数が使用されます。
-アテンションマスクは変更され、現在のトークンをマスク化します（最初の位置を除く）。
-なぜなら、それはクエリとキーが等しい（つまり非常に似ている）クエリとキーを提供するからです。
-ハッシュは多少ランダムかもしれないため、実際にはいくつかのハッシュ関数が使用され（n_roundsパラメータで決定されます）、それらが平均化されます。
-
-## Local attention
-
-[Longformer](#longformer)はローカルアテンションを使用します。
-しばしば、ローカルコンテキスト（例：左右の2つのトークンは何ですか？）は、特定のトークンに対して行動を起こすのに十分です。
-また、小さなウィンドウを持つアテンションレイヤーを積み重ねることで、最後のレイヤーはウィンドウ内のトークンだけでなく、ウィンドウ内のトークンを超えて受容野を持つようになり、文全体の表現を構築できます。
-
-一部の事前選択された入力トークンにはグローバルアテンションも与えられます。
-これらの少数のトークンに対して、アテンション行列はすべてのトークンにアクセスでき、このプロセスは対称的です。
-他のすべてのトークンは、これらの特定のトークンにアクセスできます（ローカルウィンドウ内のトークンに加えて）。
-これは、論文の図2dに示されており、以下はサンプルのアテンションマスクです：
-
-<div class="flex justify-center">
-    <img scale="50 %" align="center" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/local_attention_mask.png"/>
-</div>
-
-
-## Other tricks
-
-### Axial positional encodings
-
-[Reformer](#reformer)は軸方向の位置エンコーディングを使用しています。伝統的なトランスフォーマーモデルでは、位置エンコーディングEはサイズが \\(l\\) × \\(d\\) の行列で、\\(l\\) はシーケンスの長さ、\\(d\\) は隠れ状態の次元です。非常に長いテキストを扱う場合、この行列は非常に大きく、GPU上で大量のスペースを占有します。これを緩和するために、軸方向の位置エンコーディングは、この大きな行列Eを2つの小さな行列E1とE2に分解します。それぞれの行列はサイズ \\(l_{1} \times d_{1}\\) および \\(l_{2} \times d_{2}\\) を持ち、 \\(l_{1} \times l_{2} = l\\) および \\(d_{1} + d_{2} = d\\) という条件を満たします（長さの積を考えると、これがはるかに小さくなります）。行列E内の時刻 \\(j\\) の埋め込みは、E1内の時刻 \\(j \% l1\\) の埋め込みとE2内の時刻 \\(j // l1\\) の埋め込みを連結することによって得られます。
-

docs/source/ja/autoclass_tutorial.md

@@ -1,161 +0,0 @@
-<!--
-著作権 2023 The HuggingFace Team。全著作権所有。
-
-Apache License、Version 2.0（以下「ライセンス」と呼びます）に基づくライセンスで、
-ライセンスに従わない限り、このファイルを使用できません。
-ライセンスのコピーは以下から入手できます：
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-適用法に従うか、書面による同意がある限り、ライセンスの下でソフトウェアは配布されます。
-ライセンスに基づく特定の言語での条件を確認するか、ライセンスを参照してください。
-このファイルはMarkdown形式ですが、doc-builder（MDXに類似したもの）の特定の構文を含んでおり、
-お使いのMarkdownビューアで正しくレンダリングされない場合があります。
-
--->
-
-# AutoClassを使用して事前学習済みインスタンスをロードする
-
-さまざまなTransformerアーキテクチャが存在するため、自分のタスクに合ったモデルを作成するのは難しいことがあります。
-🤗 Transformersのコア哲学の一環として、ライブラリを使用しやすく、シンプルで柔軟にするために、
-`AutoClass`は与えられたチェックポイントから正しいアーキテクチャを自動的に推論してロードします。
-`from_pretrained()`メソッドを使用すると、事前学習済みモデルを素早くロードできるため、モデルをゼロからトレーニングするために時間とリソースを費やす必要がありません。
-この種のチェックポイントに依存しないコードを生成することは、
-コードが1つのチェックポイントで動作すれば、アーキテクチャが異なっていても、同じタスクに向けてトレーニングされた場合は別のチェックポイントでも動作することを意味します。
-
-<Tip>
-
-アーキテクチャはモデルの骨格を指し、チェックポイントは特定のアーキテクチャの重みです。
-たとえば、[BERT](https://huggingface.co/bert-base-uncased)はアーキテクチャであり、`bert-base-uncased`はチェックポイントです。
-モデルはアーキテクチャまたはチェックポイントのどちらを指す一般的な用語です。
-
-</Tip>
-
-このチュートリアルでは、以下を学習します：
-
-* 事前学習済みトークナイザをロードする。
-* 事前学習済み画像プロセッサをロードする。
-* 事前学習済み特徴量抽出器をロードする。
-* 事前学習済みプロセッサをロードする。
-* 事前学習済みモデルをロードする。
-
-## AutoTokenizer
-
-ほとんどのNLPタスクはトークナイザで始まります。トークナイザは入力をモデルで処理できる形式に変換します。
-
-[`AutoTokenizer.from_pretrained`]を使用してトークナイザをロードします：
-
-```py
->>> from transformers import AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
-```
-
-
-次に、以下のように入力をトークナイズします：
-
-```py
->>> sequence = "In a hole in the ground there lived a hobbit."
->>> print(tokenizer(sequence))
-{'input_ids': [101, 1999, 1037, 4920, 1999, 1996, 2598, 2045, 2973, 1037, 7570, 10322, 4183, 1012, 102], 
- 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
- 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
-```
-
-## AutoImageProcessor
-
-ビジョンタスクの場合、画像プロセッサが画像を正しい入力形式に変換します。
-
-```py
->>> from transformers import AutoImageProcessor
-
->>> image_processor = AutoImageProcessor.from_pretrained("google/vit-base-patch16-224")
-```
-
-## AutoFeatureExtractor
-
-オーディオタスクの場合、特徴量抽出器がオーディオ信号を正しい入力形式に変換します。
-
-[`AutoFeatureExtractor.from_pretrained`]を使用して特徴量抽出器をロードします.
-
-```py
->>> from transformers import AutoFeatureExtractor
-
->>> feature_extractor = AutoFeatureExtractor.from_pretrained(
-...     "ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition"
-... )
-```
-
-## AutoProcessor
-
-マルチモーダルタスクの場合、2つの前処理ツールを組み合わせるプロセッサが必要です。たとえば、
-[LayoutLMV2](model_doc/layoutlmv2)モデルは画像を処理するための画像プロセッサとテキストを処理するためのトークナイザが必要です。
-プロセッサはこれらの両方を組み合わせます。
-
-[`AutoProcessor.from_pretrained`]を使用してプロセッサをロードします：
-
-```py
->>> from transformers import AutoProcessor
-
->>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased")
-```
-
-## AutoModel
-
-<frameworkcontent>
-<pt>
-最後に、`AutoModelFor`クラスは特定のタスクに対して事前学習済みモデルをロードできます（使用可能なタスクの完全な一覧については[こちら](model_doc/auto)を参照）。
-たとえば、[`AutoModelForSequenceClassification.from_pretrained`]を使用してシーケンス分類用のモデルをロードできます：
-
-```py
->>> from transformers import AutoModelForSequenceClassification
-
->>> model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-同じチェックポイントを再利用して異なるタスクのアーキテクチャをロードできます：
-
-```py
->>> from transformers import AutoModelForTokenClassification
-
->>> model = AutoModelForTokenClassification.from_pretrained("distilbert-base-uncased")
-```
-
-<Tip warning={true}>
-
-PyTorchモデルの場合、  `from_pretrained()`メソッドは内部で`torch.load()`を使用し、内部的には`pickle`を使用しており、セキュリティの問題が知られています。
-一般的には、信頼性のないソースから取得した可能性があるモデルや改ざんされた可能性のあるモデルをロードしないでください。
-このセキュリティリスクは、`Hugging Face Hub`でホストされている公開モデルに対して部分的に緩和されており、各コミットでマルウェアのスキャンが行われています。
-GPGを使用した署名済みコミットの検証などのベストプラクティスについては、Hubのドキュメンテーションを参照してください。
-
-TensorFlowおよびFlaxのチェックポイントには影響がなく、`from_pretrained`メソッドの`from_tf`および`from_flax`引数を使用してPyTorchアーキテクチャ内でロードできます。
-
-</Tip>
-
-一般的に、事前学習済みモデルのインスタンスをロードするために`AutoTokenizer`クラスと`AutoModelFor`クラスの使用をお勧めします。
-これにより、常に正しいアーキテクチャをロードできます。
-次の[tutorial](preprocessing)では、新しくロードしたトークナイザ、画像プロセッサ、特徴量抽出器、およびプロセッサを使用して、ファインチューニング用にデータセットを前処理する方法を学びます。
-</pt>
-<tf>
-最後に、`TFAutoModelFor`クラスは特定のタスクに対して事前学習済みモデルをロードできます（使用可能なタスクの完全な一覧についてはこちらを参照）。
-たとえば、[`TFAutoModelForSequenceClassification.from_pretrained`]を使用してシーケンス分類用のモデルをロードできます：
-
-```py
->>> from transformers import TFAutoModelForSequenceClassification
-
->>> model = TFAutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-同じチェックポイントを再利用して異なるタスクのアーキテクチャをロードできます：
-
-```py
->>> from transformers import TFAutoModelForTokenClassification
-
->>> model = TFAutoModelForTokenClassification.from_pretrained("distilbert-base-uncased")
-```
-
-一般的には、事前学習済みモデルのインスタンスをロードするために`AutoTokenizer`クラスと`TFAutoModelFor`クラスの使用をお勧めします。
-これにより、常に正しいアーキテクチャをロードできます。
-次の[tutorial](preproccesing)では、新しくロードしたトークナイザ、画像プロセッサ、特徴量抽出器、およびプロセッサを使用して、ファインチューニング用にデータセットを前処理する方法を学びます。
-</tf>
-</frameworkcontent>

docs/source/ja/benchmarks.md

@@ -1,381 +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.
-
-⚠️ このファイルはMarkdownですが、Hugging Faceのdoc-builder（MDXに類似）向けの特定の構文を含んでいるため、
-Markdownビューアでは正しく表示されないことに注意してください。
--->
-
-# Benchmarks
-
-<Tip warning={true}>
-
-Hugging Faceのベンチマークツールは非推奨であり、Transformerモデルの速度とメモリの複雑さを測定するために外部のベンチマークライブラリを使用することをお勧めします。
-
-</Tip>
-
-[[open-in-colab]]
-
-🤗 Transformersモデルをベンチマークし、ベストプラクティス、すでに利用可能なベンチマークについて見てみましょう。
-
-🤗 Transformersモデルをベンチマークする方法について詳しく説明したノートブックは[こちら](https://github.com/huggingface/notebooks/tree/main/examples/benchmark.ipynb)で利用できます。
-
-## How to benchmark 🤗 Transformers models
-
-[`PyTorchBenchmark`]クラスと[`TensorFlowBenchmark`]クラスを使用すると、🤗 Transformersモデルを柔軟にベンチマークできます。
-ベンチマーククラスを使用すると、_ピークメモリ使用量_ および _必要な時間_ を _推論_ および _トレーニング_ の両方について測定できます。
-
-<Tip>
-
-ここでの _推論_ は、単一のフォワードパスによって定義され、 _トレーニング_ は単一のフォワードパスと
-バックワードパスによって定義されます。
-
-</Tip>
-
-ベンチマーククラス[`PyTorchBenchmark`]と[`TensorFlowBenchmark`]は、それぞれのベンチマーククラスに対する適切な設定を含む [`PyTorchBenchmarkArguments`] および [`TensorFlowBenchmarkArguments`] タイプのオブジェクトを必要とします。
-[`PyTorchBenchmarkArguments`] および [`TensorFlowBenchmarkArguments`] はデータクラスであり、それぞれのベンチマーククラスに対するすべての関連する設定を含んでいます。
-次の例では、タイプ _bert-base-cased_ のBERTモデルをベンチマークする方法が示されています。
-
-<frameworkcontent>
-<pt>
-```py
->>> from transformers import PyTorchBenchmark, PyTorchBenchmarkArguments
-
->>> args = PyTorchBenchmarkArguments(models=["bert-base-uncased"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512])
->>> benchmark = PyTorchBenchmark(args)
-```
-</pt>
-<tf>
-```py
->>> from transformers import TensorFlowBenchmark, TensorFlowBenchmarkArguments
-
->>> args = TensorFlowBenchmarkArguments(
-...     models=["bert-base-uncased"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512]
-... )
->>> benchmark = TensorFlowBenchmark(args)
-```
-</tf>
-</frameworkcontent>
-
-
-ここでは、ベンチマーク引数のデータクラスに対して、`models`、`batch_sizes`
-および`sequence_lengths`の3つの引数が指定されています。引数`models`は必須で、
-[モデルハブ](https://huggingface.co/models)からのモデル識別子の`リスト`を期待し
-ます。`batch_sizes`と`sequence_lengths`の2つの`リスト`引数は
-モデルのベンチマーク対象となる`input_ids`のサイズを定義します。
-ベンチマーク引数データクラスを介して設定できる他の多くのパラメータがあります。これらの詳細については、直接ファイル
-`src/transformers/benchmark/benchmark_args_utils.py`、
-`src/transformers/benchmark/benchmark_args.py`（PyTorch用）、および`src/transformers/benchmark/benchmark_args_tf.py`（Tensorflow用）
-を参照するか、次のシェルコマンドをルートから実行すると、PyTorchとTensorflowのそれぞれに対して設定可能なすべてのパラメータの記述的なリストが表示されます。
-
-<frameworkcontent>
-<pt>
-```bash
-python examples/pytorch/benchmarking/run_benchmark.py --help
-```
-
-インスタンス化されたベンチマークオブジェクトは、単に `benchmark.run()` を呼び出すことで実行できます。
-
-
-```py
->>> results = benchmark.run()
->>> print(results)
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length     Time in s                  
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             0.006     
-bert-base-uncased          8               32            0.006     
-bert-base-uncased          8              128            0.018     
-bert-base-uncased          8              512            0.088     
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length    Memory in MB 
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             1227
-bert-base-uncased          8               32            1281
-bert-base-uncased          8              128            1307
-bert-base-uncased          8              512            1539
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: PyTorch
-- use_torchscript: False
-- framework_version: 1.4.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 08:58:43.371351
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</pt>
-<tf>
-```bash
-python examples/tensorflow/benchmarking/run_benchmark_tf.py --help
-```
-
-インスタンス化されたベンチマークオブジェクトは、単に `benchmark.run()` を呼び出すことで実行できます。
-
-
-
-```py
->>> results = benchmark.run()
->>> print(results)
->>> results = benchmark.run()
->>> print(results)
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length     Time in s                  
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             0.005
-bert-base-uncased          8               32            0.008
-bert-base-uncased          8              128            0.022
-bert-base-uncased          8              512            0.105
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length    Memory in MB 
---------------------------------------------------------------------------------
-bert-base-uncased          8               8             1330
-bert-base-uncased          8               32            1330
-bert-base-uncased          8              128            1330
-bert-base-uncased          8              512            1770
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: Tensorflow
-- use_xla: False
-- framework_version: 2.2.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 09:26:35.617317
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</tf>
-</frameworkcontent>
-
-デフォルトでは、_推論時間_ と _必要なメモリ_ がベンチマークされます。
-上記の例の出力では、最初の2つのセクションが _推論時間_ と _推論メモリ_ 
-に対応する結果を示しています。さらに、計算環境に関するすべての関連情報、
-例えば GPU タイプ、システム、ライブラリのバージョンなどが、_ENVIRONMENT INFORMATION_ の下に表示されます。この情報は、[`PyTorchBenchmarkArguments`] 
-および [`TensorFlowBenchmarkArguments`] に引数 `save_to_csv=True` 
-を追加することで、オプションで _.csv_ ファイルに保存することができます。この場合、各セクションは別々の _.csv_ ファイルに保存されます。_.csv_ 
-ファイルへのパスは、データクラスの引数を使用してオプションで定義できます。
-
-モデル識別子、例えば `bert-base-uncased` を使用して事前学習済みモデルをベンチマークする代わりに、利用可能な任意のモデルクラスの任意の設定をベンチマークすることもできます。この場合、ベンチマーク引数と共に設定の `list` を挿入する必要があります。
-
-
-<frameworkcontent>
-<pt>
-```py
->>> from transformers import PyTorchBenchmark, PyTorchBenchmarkArguments, BertConfig
-
->>> args = PyTorchBenchmarkArguments(
-...     models=["bert-base", "bert-384-hid", "bert-6-lay"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512]
-... )
->>> config_base = BertConfig()
->>> config_384_hid = BertConfig(hidden_size=384)
->>> config_6_lay = BertConfig(num_hidden_layers=6)
-
->>> benchmark = PyTorchBenchmark(args, configs=[config_base, config_384_hid, config_6_lay])
->>> benchmark.run()
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length       Time in s                  
---------------------------------------------------------------------------------
-bert-base                  8              128            0.006
-bert-base                  8              512            0.006
-bert-base                  8              128            0.018     
-bert-base                  8              512            0.088     
-bert-384-hid              8               8             0.006     
-bert-384-hid              8               32            0.006     
-bert-384-hid              8              128            0.011     
-bert-384-hid              8              512            0.054     
-bert-6-lay                 8               8             0.003     
-bert-6-lay                 8               32            0.004     
-bert-6-lay                 8              128            0.009     
-bert-6-lay                 8              512            0.044
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length      Memory in MB 
---------------------------------------------------------------------------------
-bert-base                  8               8             1277
-bert-base                  8               32            1281
-bert-base                  8              128            1307     
-bert-base                  8              512            1539     
-bert-384-hid              8               8             1005     
-bert-384-hid              8               32            1027     
-bert-384-hid              8              128            1035     
-bert-384-hid              8              512            1255     
-bert-6-lay                 8               8             1097     
-bert-6-lay                 8               32            1101     
-bert-6-lay                 8              128            1127     
-bert-6-lay                 8              512            1359
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: PyTorch
-- use_torchscript: False
-- framework_version: 1.4.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 09:35:25.143267
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</pt>
-<tf>
-```py
->>> from transformers import TensorFlowBenchmark, TensorFlowBenchmarkArguments, BertConfig
-
->>> args = TensorFlowBenchmarkArguments(
-...     models=["bert-base", "bert-384-hid", "bert-6-lay"], batch_sizes=[8], sequence_lengths=[8, 32, 128, 512]
-... )
->>> config_base = BertConfig()
->>> config_384_hid = BertConfig(hidden_size=384)
->>> config_6_lay = BertConfig(num_hidden_layers=6)
-
->>> benchmark = TensorFlowBenchmark(args, configs=[config_base, config_384_hid, config_6_lay])
->>> benchmark.run()
-====================       INFERENCE - SPEED - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length       Time in s                  
---------------------------------------------------------------------------------
-bert-base                  8               8             0.005
-bert-base                  8               32            0.008
-bert-base                  8              128            0.022
-bert-base                  8              512            0.106
-bert-384-hid              8               8             0.005
-bert-384-hid              8               32            0.007
-bert-384-hid              8              128            0.018
-bert-384-hid              8              512            0.064
-bert-6-lay                 8               8             0.002
-bert-6-lay                 8               32            0.003
-bert-6-lay                 8              128            0.0011
-bert-6-lay                 8              512            0.074
---------------------------------------------------------------------------------
-
-====================      INFERENCE - MEMORY - RESULT       ====================
---------------------------------------------------------------------------------
-Model Name             Batch Size     Seq Length      Memory in MB 
---------------------------------------------------------------------------------
-bert-base                  8               8             1330
-bert-base                  8               32            1330
-bert-base                  8              128            1330
-bert-base                  8              512            1770
-bert-384-hid              8               8             1330
-bert-384-hid              8               32            1330
-bert-384-hid              8              128            1330
-bert-384-hid              8              512            1540
-bert-6-lay                 8               8             1330
-bert-6-lay                 8               32            1330
-bert-6-lay                 8              128            1330
-bert-6-lay                 8              512            1540
---------------------------------------------------------------------------------
-
-====================        ENVIRONMENT INFORMATION         ====================
-
-- transformers_version: 2.11.0
-- framework: Tensorflow
-- use_xla: False
-- framework_version: 2.2.0
-- python_version: 3.6.10
-- system: Linux
-- cpu: x86_64
-- architecture: 64bit
-- date: 2020-06-29
-- time: 09:38:15.487125
-- fp16: False
-- use_multiprocessing: True
-- only_pretrain_model: False
-- cpu_ram_mb: 32088
-- use_gpu: True
-- num_gpus: 1
-- gpu: TITAN RTX
-- gpu_ram_mb: 24217
-- gpu_power_watts: 280.0
-- gpu_performance_state: 2
-- use_tpu: False
-```
-</tf>
-</frameworkcontent>
-
-カスタマイズされたBertModelクラスの構成に対する推論時間と必要なメモリのベンチマーク
-
-この機能は、モデルをトレーニングする際にどの構成を選択すべきかを決定する際に特に役立つことがあります。
-
-## Benchmark best practices
-
-このセクションでは、モデルをベンチマークする際に注意すべきいくつかのベストプラクティスをリストアップしています。
-
-- 現在、単一デバイスのベンチマークしかサポートされていません。GPUでベンチマークを実行する場合、コードを実行するデバイスをユーザーが指定することを推奨します。
-  これはシェルで`CUDA_VISIBLE_DEVICES`環境変数を設定することで行えます。例：`export CUDA_VISIBLE_DEVICES=0`を実行してからコードを実行します。
-- `no_multi_processing`オプションは、テストおよびデバッグ用にのみ`True`に設定すべきです。正確なメモリ計測を確保するために、各メモリベンチマークを別々のプロセスで実行することをお勧めします。これにより、`no_multi_processing`が`True`に設定されます。
-- モデルのベンチマーク結果を共有する際には、常に環境情報を記述するべきです。異なるGPUデバイス、ライブラリバージョンなどでベンチマーク結果が大きく異なる可能性があるため、ベンチマーク結果単体ではコミュニティにとってあまり有用ではありません。
-
-## Sharing your benchmark
-
-以前、すべての利用可能なコアモデル（当時10モデル）に対して、多くの異なる設定で推論時間のベンチマークが行われました：PyTorchを使用し、TorchScriptの有無、TensorFlowを使用し、XLAの有無などです。これらのテストはすべてCPUで行われました（TensorFlow XLAを除く）。
-
-このアプローチの詳細については、[次のブログポスト](https://medium.com/huggingface/benchmarking-transformers-pytorch-and-tensorflow-e2917fb891c2)に詳しく説明されており、結果は[こちら](https://docs.google.com/spreadsheets/d/1sryqufw2D0XlUH4sq3e9Wnxu5EAQkaohzrJbd5HdQ_w/edit?usp=sharing)で利用できます。
-
-新しいベンチマークツールを使用すると、コミュニティとベンチマーク結果を共有することがこれまで以上に簡単になります。
-
-- [PyTorchベンチマーク結果](https://github.com/huggingface/transformers/tree/main/examples/pytorch/benchmarking/README.md)。
-- [TensorFlowベンチマーク結果](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/benchmarking/README.md)。

docs/source/ja/bertology.md

@@ -1,34 +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.
-
--->
-
-
-# BERTology
-
-大規模なトランスフォーマー、例えばBERTの内部動作を調査する研究領域が急成長しています（これを「BERTology」とも呼びます）。この分野の良い例は以下です：
-
-- BERT Rediscovers the Classical NLP Pipeline by Ian Tenney, Dipanjan Das, Ellie Pavlick:
-  [論文リンク](https://arxiv.org/abs/1905.05950)
-- Are Sixteen Heads Really Better than One? by Paul Michel, Omer Levy, Graham Neubig: [論文リンク](https://arxiv.org/abs/1905.10650)
-- What Does BERT Look At? An Analysis of BERT's Attention by Kevin Clark, Urvashi Khandelwal, Omer Levy, Christopher D. Manning: [論文リンク](https://arxiv.org/abs/1906.04341)
-- CAT-probing: A Metric-based Approach to Interpret How Pre-trained Models for Programming Language Attend Code Structure: [論文リンク](https://arxiv.org/abs/2210.04633)
-
-この新しい分野の発展を支援するために、BERT/GPT/GPT-2モデルにいくつかの追加機能を組み込み、人々が内部表現にアクセスできるようにしました。これらの機能は、主にPaul Michel氏の優れた研究（[論文リンク](https://arxiv.org/abs/1905.10650)）に基づいています。具体的には、以下の機能が含まれています：
-
-- BERT/GPT/GPT-2のすべての隠れ状態にアクセスすることができます。
-- BERT/GPT/GPT-2の各ヘッドの注意重みにアクセスできます。
-- ヘッドの出力値と勾配を取得し、ヘッドの重要性スコアを計算し、[論文リンク](https://arxiv.org/abs/1905.10650)で説明されているようにヘッドを削減できます。
-
-これらの機能を理解し、使用するのを支援するために、特定のサンプルスクリプト「[bertology.py](https://github.com/huggingface/transformers/tree/main/examples/research_projects/bertology/run_bertology.py)」を追加しました。このスクリプトは、GLUEで事前トレーニングされたモデルから情報を抽出し、ヘッドを削減する役割を果たします。

docs/source/ja/big_models.md

@@ -1,130 +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.
-
--->
-
-# Instantiating a big model
-
-非常に大規模な事前学習済みモデルを使用する場合、RAMの使用量を最小限に抑えることは課題の1つです。通常のPyTorchのワークフローは次のとおりです：
-
-1. ランダムな重みを持つモデルを作成します。
-2. 事前学習済みの重みをロードします。
-3. これらの事前学習済みの重みをランダムなモデルに配置します。
-
-ステップ1と2の両方がメモリにモデルの完全なバージョンを必要とし、ほとんどの場合は問題ありませんが、モデルのサイズが数ギガバイトになると、これらの2つのコピーをRAMから排除することができなくなる可能性があります。さらに悪いことに、分散トレーニングを実行するために`torch.distributed`を使用している場合、各プロセスは事前学習済みモデルをロードし、これらの2つのコピーをRAMに保存します。
-
-<Tip>
-
-ランダムに作成されたモデルは、メモリ内に「空の」テンソルで初期化されます。これらのランダムな値は、メモリの特定のチャンクにあったものを使用します（したがって、ランダムな値はその時点でのメモリチャンク内の値です）。モデル/パラメータの種類に適した分布（たとえば、正規分布）に従うランダムな初期化は、ステップ3で初期化されていない重みに対して、できるだけ高速に実行されます！
-
-</Tip>
-
-このガイドでは、Transformersがこの問題に対処するために提供するソリューションを探ります。なお、これは現在も開発が進行中の分野であり、将来、ここで説明されているAPIがわずかに変更される可能性があることに注意してください。
-
-## Sharded checkpoints
-
-バージョン4.18.0から、10GBを超えるサイズのモデルチェックポイントは自動的に複数の小さな部分に分割されます。`model.save_pretrained(save_dir)`を実行する際に1つの単一のチェックポイントを持つ代わりに、いくつかの部分的なチェックポイント（それぞれのサイズが<10GB）と、パラメータ名をそれらが格納されているファイルにマップするインデックスが生成されます。
-
-`max_shard_size`パラメータでシャーディング前の最大サイズを制御できるため、例として通常サイズのモデルと小さなシャードサイズを使用します。従来のBERTモデルを使用してみましょう。
-
-
-```py
-from transformers import AutoModel
-
-model = AutoModel.from_pretrained("bert-base-cased")
-```
-
-もし[`~PreTrainedModel.save_pretrained`]を使用して保存する場合、新しいフォルダが2つのファイルを含む形で作成されます: モデルの設定情報とその重み情報です。
-
-```py
->>> import os
->>> import tempfile
-
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir)
-...     print(sorted(os.listdir(tmp_dir)))
-['config.json', 'pytorch_model.bin']
-```
-
-最大シャードサイズを200MBに設定します：
-
-```py
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     print(sorted(os.listdir(tmp_dir)))
-['config.json', 'pytorch_model-00001-of-00003.bin', 'pytorch_model-00002-of-00003.bin', 'pytorch_model-00003-of-00003.bin', 'pytorch_model.bin.index.json']
-```
-
-モデルの設定の上に、3つの異なる重みファイルと、`index.json`ファイルが見られます。これは私たちのインデックスです。
-このようなチェックポイントは、[`~PreTrainedModel.from_pretrained`]メソッドを使用して完全に再ロードできます：
-
-```py
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     new_model = AutoModel.from_pretrained(tmp_dir)
-```
-
-主要な利点は、大規模なモデルの場合、上記のワークフローのステップ2において、各チェックポイントのシャードが前のシャードの後にロードされ、RAMのメモリ使用量をモデルのサイズと最大のシャードのサイズを合わせたものに制限できることです。
-
-内部では、インデックスファイルが使用され、どのキーがチェックポイントに存在し、対応する重みがどこに格納されているかを判断します。このインデックスは通常のJSONファイルのように読み込むことができ、辞書として取得できます。
-
-
-```py
->>> import json
-
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     with open(os.path.join(tmp_dir, "pytorch_model.bin.index.json"), "r") as f:
-...         index = json.load(f)
-
->>> print(index.keys())
-dict_keys(['metadata', 'weight_map'])
-```
-
-メタデータには現時点ではモデルの総サイズのみが含まれています。
-将来的には他の情報を追加する予定です：
-
-```py
->>> index["metadata"]
-{'total_size': 433245184}
-```
-
-重みマップはこのインデックスの主要な部分であり、各パラメータ名（通常はPyTorchモデルの`state_dict`で見つかるもの）をその格納されているファイルにマップします：
-
-```py
->>> index["weight_map"]
-{'embeddings.LayerNorm.bias': 'pytorch_model-00001-of-00003.bin',
- 'embeddings.LayerNorm.weight': 'pytorch_model-00001-of-00003.bin',
- ...
-```
-
-直接モデル内で[`~PreTrainedModel.from_pretrained`]を使用せずに、
-シャーディングされたチェックポイントをロードしたい場合（フルチェックポイントの場合に`model.load_state_dict()`を使用するように行う方法）、[`~modeling_utils.load_sharded_checkpoint`]を使用する必要があります：
-
-
-```py
->>> from transformers.modeling_utils import load_sharded_checkpoint
-
->>> with tempfile.TemporaryDirectory() as tmp_dir:
-...     model.save_pretrained(tmp_dir, max_shard_size="200MB")
-...     load_sharded_checkpoint(model, tmp_dir)
-```
-
-
-## Low memory loading
-
-シャードされたチェックポイントは、上記のワークフローのステップ2におけるメモリ使用量を削減しますが、
-低メモリの環境でそのモデルを使用するために、Accelerateライブラリに基づいた当社のツールを活用することをお勧めします。
-
-詳細については、以下のガイドをご覧ください：[Accelerateを使用した大規模モデルの読み込み](./main_classes/model#large-model-loading)

docs/source/ja/chat_templating.md

@@ -1,249 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Templates for Chat Models
-
-## Introduction
-
-LLM（Language Model）のますます一般的な使用事例の1つは「チャット」です。
-チャットのコンテキストでは、通常の言語モデルのように単一のテキストストリングを継続するのではなく、モデルは1つ以上の「メッセージ」からなる会話を継続します。
-各メッセージには「ロール」とメッセージテキストが含まれます。
-
-最も一般的に、これらのロールはユーザーからのメッセージには「ユーザー」、モデルからのメッセージには「アシスタント」が割り当てられます。
-一部のモデルは「システム」ロールもサポートしています。
-システムメッセージは通常会話の開始時に送信され、モデルの動作方法に関する指示が含まれます。
-
-すべての言語モデル、チャット用に微調整されたモデルを含むすべてのモデルは、トークンのリニアシーケンスで動作し、ロールに特有の特別な処理を持ちません。
-つまり、ロール情報は通常、メッセージ間に制御トークンを追加して注入され、メッセージの境界と関連するロールを示すことで提供されます。
-
-残念ながら、トークンの使用方法については（まだ！）標準が存在せず、異なるモデルはチャット用のフォーマットや制御トークンが大きく異なる形式でトレーニングされています。
-これはユーザーにとって実際の問題になる可能性があります。正しいフォーマットを使用しないと、モデルは入力に混乱し、パフォーマンスが本来よりも遥かに低下します。
-これが「チャットテンプレート」が解決しようとする問題です。
-
-チャット会話は通常、各辞書が「ロール」と「コンテンツ」のキーを含み、単一のチャットメッセージを表すリストとして表現されます。
-チャットテンプレートは、指定されたモデルの会話を単一のトークン化可能なシーケンスにどのようにフォーマットするかを指定するJinjaテンプレートを含む文字列です。
-トークナイザとこの情報を保存することにより、モデルが期待する形式の入力データを取得できるようになります。
-
-さっそく、`BlenderBot` モデルを使用した例を示して具体的にしましょう。`BlenderBot` のデフォルトテンプレートは非常にシンプルで、ほとんどが対話のラウンド間に空白を追加するだけです。
-
-
-```python
->>> from transformers import AutoTokenizer
->>> tokenizer = AutoTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
-
->>> chat = [
-...   {"role": "user", "content": "Hello, how are you?"},
-...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
-...   {"role": "user", "content": "I'd like to show off how chat templating works!"},
-... ]
-
->>> tokenizer.apply_chat_template(chat, tokenize=False)
-" Hello, how are you?  I'm doing great. How can I help you today?   I'd like to show off how chat templating works!</s>"
-```
-
-指定された通り、チャット全体が単一の文字列にまとめられています。デフォルトの設定である「tokenize=True」を使用すると、
-その文字列もトークン化されます。しかし、より複雑なテンプレートが実際にどのように機能するかを確認するために、
-「meta-llama/Llama-2-7b-chat-hf」モデルを使用してみましょう。ただし、このモデルはゲート付きアクセスを持っており、
-このコードを実行する場合は[リポジトリでアクセスをリクエスト](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)する必要があります。
-
-```python
->> from transformers import AutoTokenizer
->> tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
-
->> chat = [
-...   {"role": "user", "content": "Hello, how are you?"},
-...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
-...   {"role": "user", "content": "I'd like to show off how chat templating works!"},
-... ]
-
->> tokenizer.use_default_system_prompt = False
->> tokenizer.apply_chat_template(chat, tokenize=False)
-"<s>[INST] Hello, how are you? [/INST] I'm doing great. How can I help you today? </s><s>[INST] I'd like to show off how chat templating works! [/INST]"
-```
-
-今回、トークナイザは制御トークン [INST] と [/INST] を追加しました。これらはユーザーメッセージの開始と終了を示すためのものです（ただし、アシスタントメッセージには適用されません！）
-
-## How do chat templates work?
-
-モデルのチャットテンプレートは、`tokenizer.chat_template`属性に格納されています。チャットテンプレートが設定されていない場合、そのモデルクラスのデフォルトテンプレートが代わりに使用されます。`BlenderBot`のテンプレートを見てみましょう:
-
-```python
-
->>> from transformers import AutoTokenizer
->>> tokenizer = AutoTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
-
->>> tokenizer.default_chat_template
-"{% for message in messages %}{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}{{ message['content'] }}{% if not loop.last %}{{ '  ' }}{% endif %}{% endfor %}{{ eos_token }}"
-```
-
-
-これは少し抑圧的ですね。可読性を高めるために、新しい行とインデントを追加しましょう。
-各ブロックの直前の空白と、ブロックの直後の最初の改行は、デフォルトでJinjaの `trim_blocks` および `lstrip_blocks` フラグを使用して削除します。
-これにより、インデントと改行を含むテンプレートを書いても正常に機能することができます。
-
-```
-{% for message in messages %}
-    {% if message['role'] == 'user' %}
-        {{ ' ' }}
-    {% endif %}
-    {{ message['content'] }}
-    {% if not loop.last %}
-        {{ '  ' }}
-    {% endif %}
-{% endfor %}
-{{ eos_token }}
-```
-
-これが初めて見る方へ、これは[Jinjaテンプレート](https://jinja.palletsprojects.com/en/3.1.x/templates/)です。
-Jinjaはテキストを生成するためのシンプルなコードを記述できるテンプレート言語です。多くの点で、コードと
-構文はPythonに似ています。純粋なPythonでは、このテンプレートは次のようになるでしょう：
-
-```python
-for idx, message in enumerate(messages):
-    if message['role'] == 'user':
-        print(' ')
-    print(message['content'])
-    if not idx == len(messages) - 1:  # Check for the last message in the conversation
-        print('  ')
-print(eos_token)
-```
-
-実際に、このテンプレートは次の3つのことを行います：
-1. 各メッセージに対して、メッセージがユーザーメッセージである場合、それの前に空白を追加し、それ以外の場合は何も表示しません。
-2. メッセージの内容を追加します。
-3. メッセージが最後のメッセージでない場合、その後に2つのスペースを追加します。最後のメッセージの後にはEOSトークンを表示します。
-
-これは非常にシンプルなテンプレートです。制御トークンを追加しないし、モデルに対する指示を伝える一般的な方法である「システム」メッセージをサポートしていません。
-ただし、Jinjaはこれらのことを行うための多くの柔軟性を提供しています！
-LLaMAがフォーマットする方法に類似した入力をフォーマットするためのJinjaテンプレートを見てみましょう
-（実際のLLaMAテンプレートはデフォルトのシステムメッセージの処理や、一般的なシステムメッセージの処理が若干異なるため、
-実際のコードではこのテンプレートを使用しないでください！）
-
-
-```
-{% for message in messages %}
-    {% if message['role'] == 'user' %}
-        {{ bos_token + '[INST] ' + message['content'] + ' [/INST]' }}
-    {% elif message['role'] == 'system' %}
-        {{ '<<SYS>>\\n' + message['content'] + '\\n<</SYS>>\\n\\n' }}
-    {% elif message['role'] == 'assistant' %}
-        {{ ' '  + message['content'] + ' ' + eos_token }}
-    {% endif %}
-{% endfor %}
-```
-
-願わくば、少し見つめていただければ、このテンプレートが何を行っているかがわかるかもしれません。
-このテンプレートは、各メッセージの「役割」に基づいて特定のトークンを追加します。これらのトークンは、メッセージを送信した人を表すものです。
-ユーザー、アシスタント、およびシステムメッセージは、それらが含まれるトークンによってモデルによって明確に区別されます。
-
-
-## How do I create a chat template?
-
-簡単です。単純にJinjaテンプレートを書いて、`tokenizer.chat_template`を設定します。
-他のモデルから既存のテンプレートを始点にして、必要に応じて編集すると便利かもしれません！
-例えば、上記のLLaMAテンプレートを取って、アシスタントメッセージに"[ASST]"と"[/ASST]"を追加できます。
-
-```
-{% for message in messages %}
-    {% if message['role'] == 'user' %}
-        {{ bos_token + '[INST] ' + message['content'].strip() + ' [/INST]' }}
-    {% elif message['role'] == 'system' %}
-        {{ '<<SYS>>\\n' + message['content'].strip() + '\\n<</SYS>>\\n\\n' }}
-    {% elif message['role'] == 'assistant' %}
-        {{ '[ASST] '  + message['content'] + ' [/ASST]' + eos_token }}
-    {% endif %}
-{% endfor %}
-```
-
-次に、単に`tokenizer.chat_template`属性を設定してください。
-次回、[`~PreTrainedTokenizer.apply_chat_template`]を使用する際に、新しいテンプレートが使用されます！
-この属性は`tokenizer_config.json`ファイルに保存されるため、[`~utils.PushToHubMixin.push_to_hub`]を使用して
-新しいテンプレートをHubにアップロードし、みんなが正しいテンプレートを使用していることを確認できます！
-
-```python
-template = tokenizer.chat_template
-template = template.replace("SYS", "SYSTEM")  # Change the system token
-tokenizer.chat_template = template  # Set the new template
-tokenizer.push_to_hub("model_name")  # Upload your new template to the Hub!
-```
-
-[`~PreTrainedTokenizer.apply_chat_template`] メソッドは、あなたのチャットテンプレートを使用するために [`ConversationalPipeline`] クラスによって呼び出されます。
-したがって、正しいチャットテンプレートを設定すると、あなたのモデルは自動的に [`ConversationalPipeline`] と互換性があるようになります。
-
-
-## What are "default" templates?
-
-チャットテンプレートの導入前に、チャットの処理はモデルクラスレベルでハードコードされていました。
-後方互換性のために、このクラス固有の処理をデフォルトテンプレートとして保持し、クラスレベルで設定されています。
-モデルにチャットテンプレートが設定されていない場合、ただしモデルクラスのデフォルトテンプレートがある場合、
-`ConversationalPipeline`クラスや`apply_chat_template`などのメソッドはクラステンプレートを使用します。
-トークナイザのデフォルトのチャットテンプレートを確認するには、`tokenizer.default_chat_template`属性をチェックしてください。
-
-これは、後方互換性のために純粋に行っていることで、既存のワークフローを壊さないようにしています。
-モデルにとってクラステンプレートが適切である場合でも、デフォルトテンプレートをオーバーライドして
-`chat_template`属性を明示的に設定することを強くお勧めします。これにより、ユーザーにとって
-モデルがチャット用に正しく構成されていることが明確になり、デフォルトテンプレートが変更されたり廃止された場合に備えることができます。
-
-## What template should I use?
-
-すでにチャットのトレーニングを受けたモデルのテンプレートを設定する場合、テンプレートがトレーニング中にモデルが見たメッセージのフォーマットとまったく一致することを確認する必要があります。
-そうでない場合、性能の低下を経験する可能性が高いです。これはモデルをさらにトレーニングしている場合でも同様です - チャットトークンを一定に保つと、おそらく最高の性能が得られます。
-これはトークン化と非常に類似しており、通常はトレーニング中に使用されたトークン化と正確に一致する場合に、推論またはファインチューニングの際に最良の性能が得られます。
-
-一方、ゼロからモデルをトレーニングするか、チャットのためにベース言語モデルをファインチューニングする場合、適切なテンプレートを選択する自由度があります。
-LLM（Language Model）はさまざまな入力形式を処理できるほどスマートです。クラス固有のテンプレートがないモデル用のデフォルトテンプレートは、一般的なユースケースに対して良い柔軟な選択肢です。
-これは、[ChatMLフォーマット](https://github.com/openai/openai-python/blob/main/chatml.md)に従ったもので、多くのユースケースに適しています。次のようになります：
-
-```
-{% for message in messages %}
-    {{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}
-{% endfor %}
-```
-
-If you like this one, here it is in one-liner form, ready to copy into your code:
-
-```
-tokenizer.chat_template = "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}"
-```
-
-このテンプレートは、各メッセージを「``」トークンで囲み、役割を文字列として単純に記述します。
-これにより、トレーニングで使用する役割に対する柔軟性が得られます。出力は以下のようになります：
-
-
-```
-<|im_start|>system
-You are a helpful chatbot that will do its best not to say anything so stupid that people tweet about it.<|im_end|>
-<|im_start|>user
-How are you?<|im_end|>
-<|im_start|>assistant
-I'm doing great!<|im_end|>
-```
-
-「ユーザー」、「システム」、および「アシスタント」の役割は、チャットの標準です。
-特に、[`ConversationalPipeline`]との連携をスムーズに行う場合には、これらの役割を使用することをお勧めします。ただし、これらの役割に制約はありません。テンプレートは非常に柔軟で、任意の文字列を役割として使用できます。
-
-## I want to use chat templates! How should I get started?
-
-チャットモデルを持っている場合、そのモデルの`tokenizer.chat_template`属性を設定し、[`~PreTrainedTokenizer.apply_chat_template`]を使用してテストする必要があります。
-これはモデルの所有者でない場合でも適用されます。モデルのリポジトリが空のチャットテンプレートを使用している場合、またはデフォルトのクラステンプレートを使用している場合でも、
-この属性を適切に設定できるように[プルリクエスト](https://huggingface.co/docs/hub/repositories-pull-requests-discussions)を開いてください。
-
-一度属性が設定されれば、それで完了です！ `tokenizer.apply_chat_template`は、そのモデルに対して正しく動作するようになります。これは、
-`ConversationalPipeline`などの場所でも自動的にサポートされます。
-
-モデルがこの属性を持つことを確認することで、オープンソースモデルの全コミュニティがそのフルパワーを使用できるようになります。
-フォーマットの不一致はこの分野に悩み続け、パフォーマンスに黙って影響を与えてきました。それを終わらせる時が来ました！
-

docs/source/ja/community.md

@@ -1,69 +0,0 @@
-<!--⚠️ 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.
--->
-
-# Community
-
-このページは、コミュニティによって開発された🤗 Transformersに関するリソースをまとめたものです。
-
-## Community resources:
-
-| リソース     |      説明      |      作者      |
-|:----------|:-------------|------:|
-| [Hugging Face Transformers Glossary Flashcards](https://www.darigovresearch.com/huggingface-transformers-glossary-flashcards) | [Transformers Docs Glossary](glossary)に基づいたフラッシュカードセットです。このセットは、長期の知識定着を特に考慮して設計されたオープンソースのクロスプラットフォームアプリである[Anki](https://apps.ankiweb.net/)を使用して簡単に学習/復習できる形式になっています。[フラッシュカードの使用方法に関する紹介ビデオはこちら](https://www.youtube.com/watch?v=Dji_h7PILrw)をご覧ください。 | [Darigov Research](https://www.darigovresearch.com/) |
-
-## Community notebooks:
-
-| ノートブック | 説明 | 著者 | |
-|:----------|:-------------|:-------------|------:|
-| [事前学習済みのTransformerを微調整して歌詞を生成](https://github.com/AlekseyKorshuk/huggingartists) | GPT-2モデルを微調整してお気に入りのアーティストのスタイルで歌詞を生成する方法 | [Aleksey Korshuk](https://github.com/AlekseyKorshuk) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/AlekseyKorshuk/huggingartists/blob/master/huggingartists-demo.ipynb) |
-| [Tensorflow 2でT5をトレーニング](https://github.com/snapthat/TF-T5-text-to-text) | Tensorflow 2を使用して任意のタスクに対してT5をトレーニングする方法。このノートブックはTensorflow 2を使用してSQUADで実装された質問と回答タスクを示しています。 | [Muhammad Harris](https://github.com/HarrisDePerceptron) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/snapthat/TF-T5-text-to-text/blob/master/snapthatT5/notebooks/TF-T5-Datasets%20Training.ipynb) |
-| [TPUでT5をトレーニング](https://github.com/patil-suraj/exploring-T5/blob/master/T5_on_TPU.ipynb) | TransformersとNlpを使用してSQUADでT5をトレーニングする方法 | [Suraj Patil](https://github.com/patil-suraj) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/exploring-T5/blob/master/T5_on_TPU.ipynb#scrollTo=QLGiFCDqvuil) |
-| [分類と多肢選択のためにT5を微調整](https://github.com/patil-suraj/exploring-T5/blob/master/t5_fine_tuning.ipynb) | PyTorch Lightningを使用してテキスト対テキスト形式でT5を分類と多肢選択タスクに微調整する方法 | [Suraj Patil](https://github.com/patil-suraj) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/exploring-T5/blob/master/t5_fine_tuning.ipynb) |
-| [新しいデータセットと言語でDialoGPTを微調整](https://github.com/ncoop57/i-am-a-nerd/blob/master/_notebooks/2020-05-12-chatbot-part-1.ipynb) | DialoGPTモデルを新しいデータセットでオープンダイアログ会話用の微調整する方法 | [Nathan Cooper](https://github.com/ncoop57) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ncoop57/i-am-a-nerd/blob/master/_notebooks/2020-05-12-chatbot-part-1.ipynb) |
-| [Reformerを使用した長いシーケンスモデリング](https://github.com/patrickvonplaten/notebooks/blob/master/PyTorch_Reformer.ipynb) | Reformerを使用して500,000トークンまでのシーケンスをトレーニングする方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/PyTorch_Reformer.ipynb) |
-| [要約のためにBARTを微調整](https://github.com/ohmeow/ohmeow_website/blob/master/posts/2021-05-25-mbart-sequence-classification-with-blurr.ipynb) | Blurrを使用して要約のためにBARTを微調整する方法 | [Wayde Gilliam](https://ohmeow.com/) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ohmeow/ohmeow_website/blob/master/posts/2021-05-25-mbart-sequence-classification-with-blurr.ipynb) |
-| [事前学習済みのTransformerを微調整して誰かのツイートを生成](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb) | GPT-2モデルを微調整してお気に入りのTwitterアカウントのスタイルでツイートを生成する方法 | [Boris Dayma](https://github.com/borisdayma) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb) |
-| [🤗 Hugging FaceモデルをWeights & Biasesで最適化](https://colab.research.google.com/github/wandb/examples/blob/master/colabs/huggingface/Optimize_Hugging_Face_models_with_Weights_%26_Biases.ipynb) | Hugging FaceとWeights & Biasesの統合を示す完全なチュートリアル | [Boris Dayma](https://github.com/borisdayma) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/wandb/examples/blob/master/colabs/huggingface/Optimize_Hugging_Face_models_with_Weights_%26_Biases.ipynb) |
-| [Longformerの事前学習](https://github.com/allenai/longformer/blob/master/scripts/convert_model_to_long.ipynb) | 既存の事前学習済みモデルの「長い」バージョンを構築する方法 | [Iz Beltagy](https://beltagy.net) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/allenai/longformer/blob/master/scripts/convert_model_to_long.ipynb) |
-| [QAタスクのためにLongformerを微調整](https://github.com/patil-suraj/Notebooks/blob/master/longformer_qa_training.ipynb) | QAタスクのためにLongformerモデルを微調整する方法 | [Suraj Patil](https://github.com/patil-suraj) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/Notebooks/blob/master/longformer_qa_training.ipynb) |
-| [🤗nlpを使用したモデルの評価](https://github.com/patrickvonplaten/notebooks/blob/master/How_to_evaluate_Longformer_on_TriviaQA_using_NLP.ipynb) | `nlp`を使用してTriviaQAでLongformerを評価する方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1m7eTGlPmLRgoPkkA7rkhQdZ9ydpmsdLE?usp=sharing) |
-| [感情スパン抽出のためにT5を微調整](https://github.com/enzoampil/t5-intro/blob/master/t5_qa_training_pytorch_span_extraction.ipynb) | PyTorch Lightningを使用して感情スパン抽出のためにT5を微調整する方法 | [Lorenzo Ampil](https://github.com/enzoampil) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/enzoampil/t5-intro/blob/master/t5_qa_training_pytorch_span_extraction.ipynb) |
-| [DistilBertをマルチクラス分類にファインチューニング](https://github.com/abhimishra91/transformers-tutorials/blob/master/transformers_multiclass_classification.ipynb) | PyTorchを使用してDistilBertをマルチクラス分類にファインチューニングする方法 | [Abhishek Kumar Mishra](https://github.com/abhimishra91) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/abhimishra91/transformers-tutorials/blob/master/transformers_multiclass_classification.ipynb)|
-|[BERTをマルチラベル分類にファインチューニング](https://github.com/abhimishra91/transformers-tutorials/blob/master/transformers_multi_label_classification.ipynb)|PyTorchを使用してBERTをマルチラベル分類にファインチューニングする方法|[Abhishek Kumar Mishra](https://github.com/abhimishra91) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/abhimishra91/transformers-tutorials/blob/master/transformers_multi_label_classification.ipynb)|
-|[T5を要約にファインチューニング](https://github.com/abhimishra91/transformers-tutorials/blob/master/transformers_summarization_wandb.ipynb)|PyTorchを使用してT5を要約にファインチューニングし、WandBで実験をトラッキングする方法|[Abhishek Kumar Mishra](https://github.com/abhimishra91) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/abhimishra91/transformers-tutorials/blob/master/transformers_summarization_wandb.ipynb)|
-|[ダイナミックパディング/バケッティングを使用してTransformersのファインチューニングを高速化](https://github.com/ELS-RD/transformers-notebook/blob/master/Divide_Hugging_Face_Transformers_training_time_by_2_or_more.ipynb)|ダイナミックパディング/バケッティングを使用してファインチューニングを2倍高速化する方法|[Michael Benesty](https://github.com/pommedeterresautee) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1CBfRU1zbfu7-ijiOqAAQUA-RJaxfcJoO?usp=sharing)|
-|[マスク言語モデリングのためのReformerの事前学習](https://github.com/patrickvonplaten/notebooks/blob/master/Reformer_For_Masked_LM.ipynb)|双方向セルフアテンションレイヤーを備えたReformerモデルのトレーニング方法|[Patrick von Platen](https://github.com/patrickvonplaten) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1tzzh0i8PgDQGV3SMFUGxM7_gGae3K-uW?usp=sharing)|
-|[Sci-BERTを拡張してファインチューニング](https://github.com/lordtt13/word-embeddings/blob/master/COVID-19%20Research%20Data/COVID-SciBERT.ipynb)|AllenAIのCORDデータセットで事前学習済みのSciBERTモデルの語彙を拡張し、パイプライン化する方法|[Tanmay Thakur](https://github.com/lordtt13) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1rqAR40goxbAfez1xvF3hBJphSCsvXmh8)|
-|[Trainer APIを使用してBlenderBotSmallを要約のためにファインチューニング](https://github.com/lordtt13/transformers-experiments/blob/master/Custom%20Tasks/fine-tune-blenderbot_small-for-summarization.ipynb)|カスタムデータセットでBlenderBotSmallを要約のためにファインチューニングする方法、Trainer APIを使用|[Tanmay Thakur](https://github.com/lordtt13) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/19Wmupuls7mykSGyRN_Qo6lPQhgp56ymq?usp=sharing)|
-|[ElectraをファインチューニングしてCaptum Integrated Gradientsで解釈](https://github.com/elsanns/xai-nlp-notebooks/blob/master/electra_fine_tune_interpret_captum_ig.ipynb) |Electraを感情分析のためにファインチューニングし、Captum Integrated Gradientsで予測を解釈する方法|[Eliza Szczechla](https://elsanns.github.io) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/elsanns/xai-nlp-notebooks/blob/master/electra_fine_tune_interpret_captum_ig.ipynb)|
-|[Trainerクラスを使用して非英語のGPT-2モデルをファインチューニング](https://github.com/philschmid/fine-tune-GPT-2/blob/master/Fine_tune_a_non_English_GPT_2_Model_with_Huggingface.ipynb) |Trainerクラスを使用して非英語のGPT-2モデルをファインチューニングする方法|[Philipp Schmid](https://www.philschmid.de) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/philschmid/fine-tune-GPT-2/blob/master/Fine_tune_a_non_English_GPT_2_Model_with_Huggingface.ipynb)|
-|[DistilBERTモデルをマルチラベル分類タスクのためにファインチューニング](https://github.com/DhavalTaunk08/Transformers_scripts/blob/master/Transformers_multilabel_distilbert.ipynb) |DistilBERTモデルをマルチラベル分類タスクのためにファインチューニングする方法|[Dhaval Taunk](https://github.com/DhavalTaunk08) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/DhavalTaunk08/Transformers_scripts/blob/master/Transformers_multilabel_distilbert.ipynb)|
-|[ALBERTを文ペア分類タスクのためにファインチューニング](https://github.com/NadirEM/nlp-notebooks/blob/master/Fine_tune_ALBERT_sentence_pair_classification.ipynb) |ALBERTモデルまたは他のBERTベースのモデルを文ペア分類タスクのためにファインチューニングする方法|[Nadir El Manouzi](https://github.com/NadirEM) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NadirEM/nlp-notebooks/blob/master/Fine_tune_ALBERT_sentence_pair_classification.ipynb)|
-|[RoBERTaを感情分析のためにファインチューニング](https://github.com/DhavalTaunk08/NLP_scripts/blob/master/sentiment_analysis_using_roberta.ipynb) |RoBERTaモデルを感情分析のためにファインチューニングする方法|[Dhaval Taunk](https://github.com/DhavalTaunk08) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/DhavalTaunk08/NLP_scripts/blob/master/sentiment_analysis_using_roberta.ipynb)|
-|[質問生成モデルの評価](https://github.com/flexudy-pipe/qugeev) | seq2seqトランスフォーマーモデルによって生成された質問の回答の正確さを評価する方法 | [Pascal Zoleko](https://github.com/zolekode) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1bpsSqCQU-iw_5nNoRm_crPq6FRuJthq_?usp=sharing)|
-|[DistilBERTとTensorflowを使用してテキストを分類](https://github.com/peterbayerle/huggingface_notebook/blob/main/distilbert_tf.ipynb) | TensorFlowでテキスト分類のためにDistilBERTをファインチューニングする方法 | [Peter Bayerle](https://github.com/peterbayerle) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/peterbayerle/huggingface_notebook/blob/main/distilbert_tf.ipynb)|
-|[CNN/Dailymailでのエンコーダーデコーダー要約にBERTを活用](https://github.com/patrickvonplaten/notebooks/blob/master/BERT2BERT_for_CNN_Dailymail.ipynb) | *bert-base-uncased* チェックポイントを使用してCNN/Dailymailの要約のために *EncoderDecoderModel* をウォームスタートする方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/BERT2BERT_for_CNN_Dailymail.ipynb)|
-|[BBC XSumでのエンコーダーデコーダー要約にRoBERTaを活用](https://github.com/patrickvonplaten/notebooks/blob/master/RoBERTaShared_for_BBC_XSum.ipynb) | *roberta-base* チェックポイントを使用してBBC/XSumの要約のための共有 *EncoderDecoderModel* をウォームスタートする方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/RoBERTaShared_for_BBC_XSum.ipynb)|
-|[TAPASをシーケンシャル質問応答（SQA）でファインチューニング](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/TAPAS/Fine_tuning_TapasForQuestionAnswering_on_SQA.ipynb) | シーケンシャル質問応答（SQA）データセットで *tapas-base* チェックポイントを使用して *TapasForQuestionAnswering* をファインチューニングする方法 | [Niels Rogge](https://github.com/nielsrogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/TAPAS/Fine_tuning_TapasForQuestionAnswering_on_SQA.ipynb)|
-|[TabFactでTAPASを評価](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/TAPAS/Evaluating_TAPAS_on_the_Tabfact_test_set.ipynb) | *tapas-base-finetuned-tabfact* チェックポイントを使用してファインチューニングされた *TapasForSequenceClassification* を評価する方法、🤗 datasets と 🤗 transformers ライブラリを組み合わせて使用 | [Niels Rogge](https://github.com/nielsrogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/TAPAS/Evaluating_TAPAS_on_the_Tabfact_test_set.ipynb)|
-|[翻訳のためのmBARTをファインチューニング](https://colab.research.google.com/github/vasudevgupta7/huggingface-tutorials/blob/main/translation_training.ipynb) | Seq2SeqTrainerを使用してHindiからEnglishへの翻訳のためにmBARTをファインチューニングする方法 | [Vasudev Gupta](https://github.com/vasudevgupta7) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/vasudevgupta7/huggingface-tutorials/blob/main/translation_training.ipynb)|
-|[FUNSD（フォーム理解データセット）でLayoutLMをファインチューニング](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForTokenClassification_on_FUNSD.ipynb) | スキャンされたドキュメントからの情報抽出のためにFUNSDデータセットで *LayoutLMForTokenClassification* をファインチューニングする方法 | [Niels Rogge](https://github.com/nielsrogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForTokenClassification_on_FUNSD.ipynb)|
-| [DistilGPT2のファインチューニングとテキスト生成](https://colab.research.google.com/github/tripathiaakash/DistilGPT2-Tutorial/blob/main/distilgpt2_fine_tuning.ipynb) | DistilGPT2のファインチューニングとテキスト生成方法 | [Aakash Tripathi](https://github.com/tripathiaakash) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/tripathiaakash/DistilGPT2-Tutorial/blob/main/distilgpt2_fine_tuning.ipynb)|
-| [最大8KトークンでのLEDのファインチューニング](https://github.com/patrickvonplaten/notebooks/blob/master/Fine_tune_Longformer_Encoder_Decoder_(LED)_for_Summarization_on_pubmed.ipynb) | ロングレンジ要約のためのpubmedでLEDをファインチューニングする方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/Fine_tune_Longformer_Encoder_Decoder_(LED)_for_Summarization_on_pubmed.ipynb)|
-| [ArxivでのLEDの評価](https://github.com/patrickvonplaten/notebooks/blob/master/LED_on_Arxiv.ipynb) | ロングレンジ要約のためのLEDの効果的な評価方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/LED_on_Arxiv.ipynb)|
-| [RVL-CDIP（文書画像分類データセット）でのLayoutLMのファインチューニング](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForSequenceClassification_on_RVL_CDIP.ipynb) | スキャンされた文書の分類のためのRVL-CDIPデータセットで*LayoutLMForSequenceClassification*をファインチューニングする方法 | [Niels Rogge](https://github.com/nielsrogge) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForSequenceClassification_on_RVL_CDIP.ipynb)|
-| [Wav2Vec2 CTCデコーディングとGPT2の調整](https://github.com/voidful/huggingface_notebook/blob/main/xlsr_gpt.ipynb) | 言語モデルの調整を伴うCTCシーケンスのデコーディング方法 | [Eric Lam](https://github.com/voidful) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1e_z5jQHYbO2YKEaUgzb1ww1WwiAyydAj?usp=sharing)|
-| [Trainerクラスを使用した2言語の要約用にBARTをファインチューニング](https://github.com/elsanns/xai-nlp-notebooks/blob/master/fine_tune_bart_summarization_two_langs.ipynb) | トレーナークラスを使用して2つの言語での要約用にBARTをファインチューニングする方法 | [Eliza Szczechla](https://github.com/elsanns) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/elsanns/xai-nlp-notebooks/blob/master/fine_tune_bart_summarization_two_langs.ipynb)|
-| [PubMedデータセットでBigBirdの評価](https://github.com/patrickvonplaten/notebooks/blob/master/Evaluating_Big_Bird_on_TriviaQA.ipynb) | Trivia QAの長いドキュメント質問応答でBigBirdの評価方法 | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/Evaluating_Big_Bird_on_TriviaQA.ipynb)|
-| [Wav2Vec2を使用してビデオの字幕を作成する](https://github.com/Muennighoff/ytclipcc/blob/main/wav2vec_youtube_captions.ipynb) | Wav2Vecでオーディオを転記して任意のビデオからYouTubeの字幕を作成する方法 | [Niklas Muennighoff](https://github.com/Muennighoff) |[![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Muennighoff/ytclipcc/blob/main/wav2vec_youtube_captions.ipynb) |
-| [PyTorch Lightningを使用したCIFAR-10でのVision Transformerのファインチューニング](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_PyTorch_Lightning.ipynb) | HuggingFace Transformers、Datasets、およびPyTorch Lightningを使用してCIFAR-10でVision Transformer（ViT）をファインチューニングする方法 | [Niels Rogge](https://github.com/nielsrogge) |[![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_PyTorch_Lightning.ipynb) |
-| [🤗 Trainerを使用したCIFAR-10でのVision Transformerのファインチューニング](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_the_%F0%9F%A4%97_Trainer.ipynb) | HuggingFace Transformers、Datasets、および🤗 Trainerを使用してCIFAR-10でVision Transformer（ViT）をファインチューニングする方法 | [Niels Rogge](https://github.com/nielsrogge) |[![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_the_%F0%9F%A4%97_Trainer.ipynb) |
-| [Open Entity、エンティティタイピングデータセットでLUKEの評価](https://github.com/studio-ousia/luke/blob/master/notebooks/huggingface_open_entity.ipynb) | Open Entityデータセットで*LukeForEntityClassification*の評価方法 | [Ikuya Yamada](https://github.com/ikuyamada) |[![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/studio-ousia/luke/blob/master/notebooks/huggingface_open_entity.ipynb) |
-| [TACRED、関係抽出データセットでLUKEの評価](https://github.com/studio-ousia/luke/blob/master/notebooks/huggingface_tacred.ipynb) | TACREDデータセットで*LukeForEntityPairClassification*の評価方法 | [Ikuya Yamada](https://github.com/ikuyamada) |[![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/studio-ousia/luke/blob/master/notebooks/huggingface_tacred.ipynb) |
-| [CoNLL-2003、重要なNERベンチマークでLUKEの評価](https://github.com/studio-ousia/luke/blob/master/notebooks/huggingface_conll_2003.ipynb) | CoNLL-2003データセットで*LukeForEntitySpanClassification*の評価方法 | [Ikuya Yamada](https://github.com/ikuyamada) |[![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/studio-ousia/luke/blob/master/notebooks/huggingface_conll_2003.ipynb) |
-| [PubMedデータセットでBigBird-Pegasusの評価](https://github.com/vasudevgupta7/bigbird/blob/main/notebooks/bigbird_pegasus_evaluation.ipynb) | PubMedデータセットで*BigBirdPegasusForConditionalGeneration*の評価方法 | [Vasudev Gupta](https://github.com/vasudevgupta7) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/vasudevgupta7/bigbird/blob/main/notebooks/bigbird_pegasus_evaluation.ipynb) |
-| [Wav2Vec2を使用したスピーチエモーション分類](https://github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb) | MEGAデータセットでの感情分類のための事前学習済みWav2Vec2モデルの利用方法 | [Mehrdad Farahani](https://github.com/m3hrdadfi) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb) |
-| [DETRを使用して画像内のオブジェクトを検出する](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/DETR/DETR_minimal_example_(with_DetrFeatureExtractor).ipynb) | トレーニング済み*DetrForObjectDetection*モデルを使用して画像内のオブジェクトを検出し、注意を可視化する方法 | [Niels Rogge](https://github.com/NielsRogge) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/DETR_minimal_example_(with_DetrFeatureExtractor).ipynb) |
-| [カスタムオブジェクト検出データセットでDETRをファインチューニングする](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForObjectDetection_on_custom_dataset_(balloon).ipynb) | カスタムオブジェクト検出データセットで*DetrForObjectDetection*をファインチューニングする方法 | [Niels Rogge](https://github.com/NielsRogge) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForObjectDetection_on_custom_dataset_(balloon).ipynb) |
-| [Named Entity RecognitionのためにT5をファインチューニング](https://github.com/ToluClassics/Notebooks/blob/main/T5_Ner_Finetuning.ipynb) | Named Entity RecognitionタスクでT5をファインチューニングする方法 | [Ogundepo Odunayo](https://github.com/ToluClassics) | [![Colabで開く](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1obr78FY_cBmWY5ODViCmzdY6O1KB65Vc?usp=sharing) |

docs/source/ja/create_a_model.md

@@ -1,420 +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.
-
--->
-
-# Create a custom architecture
-
-[`AutoClass`](model_doc/auto)は、モデルのアーキテクチャを自動的に推論し、事前学習済みの設定と重みをダウンロードします。一般的には、チェックポイントに依存しないコードを生成するために`AutoClass`を使用することをお勧めします。ただし、特定のモデルパラメータに対する制御をより詳細に行いたいユーザーは、いくつかの基本クラスからカスタム🤗 Transformersモデルを作成できます。これは、🤗 Transformersモデルを研究、トレーニング、または実験する興味があるユーザーに特に役立つかもしれません。このガイドでは、`AutoClass`を使用しないカスタムモデルの作成について詳しく説明します。次の方法を学びます：
-
-- モデルの設定をロードおよびカスタマイズする。
-- モデルアーキテクチャを作成する。
-- テキスト用の遅いトークナイザと高速トークナイザを作成する。
-- ビジョンタスク用の画像プロセッサを作成する。
-- オーディオタスク用の特徴抽出器を作成する。
-- マルチモーダルタスク用のプロセッサを作成する。
-
-## Configuration
-
-[設定](main_classes/configuration)は、モデルの特定の属性を指します。各モデルの設定には異なる属性があります。たとえば、すべてのNLPモデルには、`hidden_size`、`num_attention_heads`、`num_hidden_layers`、および`vocab_size`属性が共通してあります。これらの属性は、モデルを構築するための注意ヘッドの数や隠れ層の数を指定します。
-
-[DistilBERT](model_doc/distilbert)をより詳しく調べるために、[`DistilBertConfig`]にアクセスしてその属性を調べてみましょう：
-
-```py
->>> from transformers import DistilBertConfig
-
->>> config = DistilBertConfig()
->>> print(config)
-DistilBertConfig {
-  "activation": "gelu",
-  "attention_dropout": 0.1,
-  "dim": 768,
-  "dropout": 0.1,
-  "hidden_dim": 3072,
-  "initializer_range": 0.02,
-  "max_position_embeddings": 512,
-  "model_type": "distilbert",
-  "n_heads": 12,
-  "n_layers": 6,
-  "pad_token_id": 0,
-  "qa_dropout": 0.1,
-  "seq_classif_dropout": 0.2,
-  "sinusoidal_pos_embds": false,
-  "transformers_version": "4.16.2",
-  "vocab_size": 30522
-}
-```
-
-[`DistilBertConfig`]は、基本の[`DistilBertModel`]を構築するために使用されるすべてのデフォルト属性を表示します。
-すべての属性はカスタマイズ可能で、実験のためのスペースを提供します。例えば、デフォルトのモデルをカスタマイズして以下のようなことができます：
-
-- `activation`パラメータで異なる活性化関数を試す。
-- `attention_dropout`パラメータで注意確率の高いドロップアウト率を使用する。
-
-
-```py
->>> my_config = DistilBertConfig(activation="relu", attention_dropout=0.4)
->>> print(my_config)
-DistilBertConfig {
-  "activation": "relu",
-  "attention_dropout": 0.4,
-  "dim": 768,
-  "dropout": 0.1,
-  "hidden_dim": 3072,
-  "initializer_range": 0.02,
-  "max_position_embeddings": 512,
-  "model_type": "distilbert",
-  "n_heads": 12,
-  "n_layers": 6,
-  "pad_token_id": 0,
-  "qa_dropout": 0.1,
-  "seq_classif_dropout": 0.2,
-  "sinusoidal_pos_embds": false,
-  "transformers_version": "4.16.2",
-  "vocab_size": 30522
-}
-```
-
-事前学習済みモデルの属性は、[`~PretrainedConfig.from_pretrained`] 関数で変更できます：
-
-```py
->>> my_config = DistilBertConfig.from_pretrained("distilbert-base-uncased", activation="relu", attention_dropout=0.4)
-```
-
-Once you are satisfied with your model configuration, you can save it with [`PretrainedConfig.save_pretrained`]. Your configuration file is stored as a JSON file in the specified save directory.
-
-```py
->>> my_config.save_pretrained(save_directory="./your_model_save_path")
-```
-
-設定ファイルを再利用するには、[`~PretrainedConfig.from_pretrained`]を使用してそれをロードします：
-
-```py
->>> my_config = DistilBertConfig.from_pretrained("./your_model_save_path/config.json")
-```
-
-<Tip>
-
-カスタム構成ファイルを辞書として保存することも、カスタム構成属性とデフォルトの構成属性の違いだけを保存することもできます！詳細については[configuration](main_classes/configuration)のドキュメンテーションをご覧ください。
-
-</Tip>
-
-## Model
-
-次のステップは、[モデル](main_classes/models)を作成することです。モデル（アーキテクチャとも緩く言われることがあります）は、各レイヤーが何をしているか、どの操作が行われているかを定義します。構成からの `num_hidden_layers` のような属性はアーキテクチャを定義するために使用されます。
-すべてのモデルは [`PreTrainedModel`] をベースクラスとし、入力埋め込みのリサイズやセルフアテンションヘッドのプルーニングなど、共通のメソッドがいくつかあります。
-さらに、すべてのモデルは [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html)、[`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)、または [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/flax.linen.html#module) のいずれかのサブクラスでもあります。つまり、モデルはそれぞれのフレームワークの使用法と互換性があります。
-
-<frameworkcontent>
-<pt>
-モデルにカスタム構成属性をロードします：
-
-```py
->>> from transformers import DistilBertModel
-
->>> my_config = DistilBertConfig.from_pretrained("./your_model_save_path/config.json")
->>> model = DistilBertModel(my_config)
-```
-
-これにより、事前トレーニング済みの重みではなくランダムな値を持つモデルが作成されます。
-これは、トレーニングが行われるまで、まだ有用なものとして使用することはできません。
-トレーニングはコストと時間がかかるプロセスです。
-通常、トレーニングに必要なリソースの一部しか使用せず、より速くより良い結果を得るために事前学習済みモデルを使用することが良いでしょう。
-
-[`~PreTrainedModel.from_pretrained`]を使用して事前学習済みモデルを作成します：
-
-
-```py
->>> model = DistilBertModel.from_pretrained("distilbert-base-uncased")
-```
-
-事前学習済みの重みをロードする際、モデルが🤗 Transformersによって提供されている場合、デフォルトのモデル設定が自動的にロードされます。ただし、必要に応じてデフォルトのモデル設定属性の一部またはすべてを独自のもので置き換えることができます。
-
-```py
->>> model = DistilBertModel.from_pretrained("distilbert-base-uncased", config=my_config)
-```
-</pt>
-<tf>
-モデルにカスタム設定属性をロードしてください：
-
-```py
->>> from transformers import TFDistilBertModel
-
->>> my_config = DistilBertConfig.from_pretrained("./your_model_save_path/my_config.json")
->>> tf_model = TFDistilBertModel(my_config)
-```
-
-これにより、事前学習済みの重みではなくランダムな値を持つモデルが作成されます。
-このモデルを有用な目的にはまだ使用することはできません。トレーニングはコストがかかり、時間がかかるプロセスです。
-一般的には、トレーニングに必要なリソースの一部しか使用せずに、より速く優れた結果を得るために事前学習済みモデルを使用することが良いでしょう。
-
-[`~TFPreTrainedModel.from_pretrained`]を使用して事前学習済みモデルを作成します：
-
-
-```py
->>> tf_model = TFDistilBertModel.from_pretrained("distilbert-base-uncased")
-```
-
-事前学習済みの重みをロードする際、モデルが🤗 Transformersによって提供されている場合、デフォルトのモデル構成が自動的にロードされます。ただし、必要であればデフォルトのモデル構成属性の一部またはすべてを独自のもので置き換えることもできます：
-
-```py
->>> tf_model = TFDistilBertModel.from_pretrained("distilbert-base-uncased", config=my_config)
-```
-</tf>
-</frameworkcontent>
-
-
-### Model heads
-
-この時点で、ベースのDistilBERTモデルがあり、これは隠れた状態を出力します。隠れた状態はモデルのヘッドへの入力として渡され、最終的な出力を生成します。🤗 Transformersは、モデルがそのタスクをサポートしている限り、各タスクに対応する異なるモデルヘッドを提供します（つまり、DistilBERTを翻訳のようなシーケンス対シーケンスタスクに使用することはできません）。
-
-<frameworkcontent>
-<pt>
-たとえば、[`DistilBertForSequenceClassification`]は、シーケンス分類ヘッドを持つベースのDistilBERTモデルです。シーケンス分類ヘッドは、プールされた出力の上にある線形層です。
-
-```py
->>> from transformers import DistilBertForSequenceClassification
-
->>> model = DistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-新しいタスクにこのチェックポイントを簡単に再利用するには、異なるモデルヘッドに切り替えます。
-質問応答タスクの場合、[`DistilBertForQuestionAnswering`] モデルヘッドを使用します。
-質問応答ヘッドはシーケンス分類ヘッドと類似していますが、隠れ状態の出力の上に線形層があります。
-
-```py
->>> from transformers import DistilBertForQuestionAnswering
-
->>> model = DistilBertForQuestionAnswering.from_pretrained("distilbert-base-uncased")
-```
-
-</pt>
-<tf>
-例えば、[`TFDistilBertForSequenceClassification`]は、シーケンス分類ヘッドを持つベースのDistilBERTモデルです。シーケンス分類ヘッドは、プールされた出力の上にある線形層です。
-
-```py
->>> from transformers import TFDistilBertForSequenceClassification
-
->>> tf_model = TFDistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")
-```
-
-別のタスクにこのチェックポイントを簡単に再利用することができ、異なるモデルヘッドに切り替えるだけです。
-質問応答タスクの場合、[`TFDistilBertForQuestionAnswering`]モデルヘッドを使用します。
-質問応答ヘッドはシーケンス分類ヘッドと似ていますが、隠れ状態の出力の上に線形層があるだけです。
-
-
-```py
->>> from transformers import TFDistilBertForQuestionAnswering
-
->>> tf_model = TFDistilBertForQuestionAnswering.from_pretrained("distilbert-base-uncased")
-```
-</tf>
-</frameworkcontent>
-
-## Tokenizer
-
-テキストデータをモデルで使用する前に必要な最後のベースクラスは、生のテキストをテンソルに変換するための[トークナイザ](main_classes/tokenizer)です。
-🤗 Transformersで使用できる2つのタイプのトークナイザがあります：
-
-- [`PreTrainedTokenizer`]: トークナイザのPython実装です。
-- [`PreTrainedTokenizerFast`]: Rustベースの[🤗 Tokenizer](https://huggingface.co/docs/tokenizers/python/latest/)ライブラリからのトークナイザです。
-このトークナイザのタイプは、そのRust実装により、特にバッチトークナイゼーション中に高速です。
-高速なトークナイザは、トークンを元の単語または文字にマッピングする*オフセットマッピング*などの追加メソッドも提供します。
-
-両方のトークナイザは、エンコードとデコード、新しいトークンの追加、特別なトークンの管理など、共通のメソッドをサポートしています。
-
-<Tip warning={true}>
-
-すべてのモデルが高速なトークナイザをサポートしているわけではありません。
-モデルが高速なトークナイザをサポートしているかどうかを確認するには、この[表](index#supported-frameworks)をご覧ください。
-
-</Tip>
-
-独自のトークナイザをトレーニングした場合、*ボキャブラリー*ファイルからトークナイザを作成できます。
-
-```py
->>> from transformers import DistilBertTokenizer
-
->>> my_tokenizer = DistilBertTokenizer(vocab_file="my_vocab_file.txt", do_lower_case=False, padding_side="left")
-```
-
-カスタムトークナイザーから生成される語彙は、事前学習済みモデルのトークナイザーが生成する語彙とは異なることを覚えておくことは重要です。
-事前学習済みモデルを使用する場合は、事前学習済みモデルの語彙を使用する必要があります。そうしないと、入力が意味をなさなくなります。
-[`DistilBertTokenizer`]クラスを使用して、事前学習済みモデルの語彙を持つトークナイザーを作成します:
-
-
-```py
->>> from transformers import DistilBertTokenizer
-
->>> slow_tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
-```
-
-[`DistilBertTokenizerFast`]クラスを使用して高速なトークナイザを作成します：
-
-```py
->>> from transformers import DistilBertTokenizerFast
-
->>> fast_tokenizer = DistilBertTokenizerFast.from_pretrained("distilbert-base-uncased")
-```
-
-<Tip>
-
-デフォルトでは、[`AutoTokenizer`]は高速なトークナイザを読み込もうとします。`from_pretrained`内で`use_fast=False`を設定することで、この動作を無効にすることができます。
-
-</Tip>
-
-## Image Processor
-
-画像プロセッサはビジョン入力を処理します。これは基本クラス [`~image_processing_utils.ImageProcessingMixin`] を継承しています。
-
-使用するには、使用しているモデルに関連付けられた画像プロセッサを作成します。
-たとえば、画像分類に[ViT](model_doc/vit)を使用する場合、デフォルトの [`ViTImageProcessor`] を作成します。
-
-```py
->>> from transformers import ViTImageProcessor
-
->>> vit_extractor = ViTImageProcessor()
->>> print(vit_extractor)
-ViTImageProcessor {
-  "do_normalize": true,
-  "do_resize": true,
-  "image_processor_type": "ViTImageProcessor",
-  "image_mean": [
-    0.5,
-    0.5,
-    0.5
-  ],
-  "image_std": [
-    0.5,
-    0.5,
-    0.5
-  ],
-  "resample": 2,
-  "size": 224
-}
-```
-
-<Tip>
-
-カスタマイズを必要としない場合、モデルのデフォルトの画像プロセッサパラメータをロードするには、単純に`from_pretrained`メソッドを使用してください。
-
-</Tip>
-
-[`ViTImageProcessor`]のパラメータを変更して、カスタムの画像プロセッサを作成できます：
-
-
-```py
->>> from transformers import ViTImageProcessor
-
->>> my_vit_extractor = ViTImageProcessor(resample="PIL.Image.BOX", do_normalize=False, image_mean=[0.3, 0.3, 0.3])
->>> print(my_vit_extractor)
-ViTImageProcessor {
-  "do_normalize": false,
-  "do_resize": true,
-  "image_processor_type": "ViTImageProcessor",
-  "image_mean": [
-    0.3,
-    0.3,
-    0.3
-  ],
-  "image_std": [
-    0.5,
-    0.5,
-    0.5
-  ],
-  "resample": "PIL.Image.BOX",
-  "size": 224
-}
-```
-
-## Feature Extractor
-
-フィーチャー抽出器は音声入力を処理します。これは基本的な [`~feature_extraction_utils.FeatureExtractionMixin`] クラスから継承され、音声入力を処理するための [`SequenceFeatureExtractor`] クラスからも継承されることがあります。
-
-使用するには、モデルに関連付けられたフィーチャー抽出器を作成します。たとえば、音声分類に [Wav2Vec2](model_doc/wav2vec2) を使用する場合、デフォルトの [`Wav2Vec2FeatureExtractor`] を作成します。
-
-
-```py
->>> from transformers import Wav2Vec2FeatureExtractor
-
->>> w2v2_extractor = Wav2Vec2FeatureExtractor()
->>> print(w2v2_extractor)
-Wav2Vec2FeatureExtractor {
-  "do_normalize": true,
-  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
-  "feature_size": 1,
-  "padding_side": "right",
-  "padding_value": 0.0,
-  "return_attention_mask": false,
-  "sampling_rate": 16000
-}
-```
-
-<Tip>
-
-カスタマイズを行わない場合、モデルのデフォルトの特徴抽出器パラメーターをロードするには、単に `from_pretrained` メソッドを使用してください。
-
-</Tip>
-
-[`Wav2Vec2FeatureExtractor`] のパラメーターを変更して、カスタム特徴抽出器を作成できます:
-
-```py
->>> from transformers import Wav2Vec2FeatureExtractor
-
->>> w2v2_extractor = Wav2Vec2FeatureExtractor(sampling_rate=8000, do_normalize=False)
->>> print(w2v2_extractor)
-Wav2Vec2FeatureExtractor {
-  "do_normalize": false,
-  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
-  "feature_size": 1,
-  "padding_side": "right",
-  "padding_value": 0.0,
-  "return_attention_mask": false,
-  "sampling_rate": 8000
-}
-```
-
-## Processor
-
-マルチモーダルタスクをサポートするモデルに対して、🤗 Transformersは便利なプロセッサクラスを提供しています。
-このプロセッサクラスは、特徴量抽出器やトークナイザなどの処理クラスを便利にラップし、単一のオブジェクトに結合します。
-たとえば、自動音声認識タスク（ASR）用に[`Wav2Vec2Processor`]を使用してみましょう。
-ASRは音声をテキストに転写するタスクであり、音声入力を処理するために特徴量抽出器とトークナイザが必要です。
-
-音声入力を処理する特徴量抽出器を作成します：
-
-```py
->>> from transformers import Wav2Vec2FeatureExtractor
-
->>> feature_extractor = Wav2Vec2FeatureExtractor(padding_value=1.0, do_normalize=True)
-```
-
-テキスト入力を処理するトークナイザを作成します:
-
-```py
->>> from transformers import Wav2Vec2CTCTokenizer
-
->>> tokenizer = Wav2Vec2CTCTokenizer(vocab_file="my_vocab_file.txt")
-```
-
-[`Wav2Vec2Processor`]で特徴量抽出器とトークナイザを組み合わせます：
-
-
-```py
->>> from transformers import Wav2Vec2Processor
-
->>> processor = Wav2Vec2Processor(feature_extractor=feature_extractor, tokenizer=tokenizer)
-```
-
-二つの基本クラス - 設定とモデル - および追加の前処理クラス（トークナイザ、画像プロセッサ、特徴抽出器、またはプロセッサ）を使用することで、🤗 Transformers がサポートするモデルのいずれかを作成できます。これらの基本クラスは設定可能で、必要な特性を使用できます。モデルをトレーニング用に簡単にセットアップしたり、既存の事前学習済みモデルを微調整することができます。

docs/source/ja/custom_models.md

@@ -1,336 +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.
-
--->
-
-# Sharing custom models
-
-🤗 Transformersライブラリは、簡単に拡張できるように設計されています。すべてのモデルはリポジトリの特定のサブフォルダに完全にコード化されており、抽象化はありません。したがって、モデリングファイルをコピーして調整することが簡単です。
-
-新しいモデルを書いている場合、ゼロから始める方が簡単かもしれません。このチュートリアルでは、カスタムモデルとその設定をどのように書き、Transformers内で使用できるようにし、コードに依存する共同体と共有する方法を説明します。ライブラリに存在しない場合でも、誰でも使用できるようにします。
-
-これを実証するために、[timmライブラリ](https://github.com/rwightman/pytorch-image-models)のResNetクラスを[`PreTrainedModel`]にラップすることによって、ResNetモデルを使用します。
-
-## Writing a custom configuration
-
-モデルに取り組む前に、まずその設定を書きましょう。モデルの設定は、モデルを構築するために必要なすべての情報を含むオブジェクトです。次のセクションで見るように、モデルは初期化するために`config`しか受け取ることができないため、そのオブジェクトができるだけ完全である必要があります。
-
-この例では、ResNetクラスのいくつかの引数を取得し、調整したいかもしれないとします。異なる設定は、異なるタイプのResNetを提供します。その後、これらの引数を確認した後、それらの引数を単に格納します。
-
-```python
-from transformers import PretrainedConfig
-from typing import List
-
-
-class ResnetConfig(PretrainedConfig):
-    model_type = "resnet"
-
-    def __init__(
-        self,
-        block_type="bottleneck",
-        layers: List[int] = [3, 4, 6, 3],
-        num_classes: int = 1000,
-        input_channels: int = 3,
-        cardinality: int = 1,
-        base_width: int = 64,
-        stem_width: int = 64,
-        stem_type: str = "",
-        avg_down: bool = False,
-        **kwargs,
-    ):
-        if block_type not in ["basic", "bottleneck"]:
-            raise ValueError(f"`block_type` must be 'basic' or bottleneck', got {block_type}.")
-        if stem_type not in ["", "deep", "deep-tiered"]:
-            raise ValueError(f"`stem_type` must be '', 'deep' or 'deep-tiered', got {stem_type}.")
-
-        self.block_type = block_type
-        self.layers = layers
-        self.num_classes = num_classes
-        self.input_channels = input_channels
-        self.cardinality = cardinality
-        self.base_width = base_width
-        self.stem_width = stem_width
-        self.stem_type = stem_type
-        self.avg_down = avg_down
-        super().__init__(**kwargs)
-```
-
-重要なことを3つ覚えておくべきポイントは次のとおりです：
-- `PretrainedConfig` を継承する必要があります。
-- あなたの `PretrainedConfig` の `__init__` は任意の kwargs を受け入れる必要があります。
-- これらの `kwargs` は親クラスの `__init__` に渡す必要があります。
-
-継承は、🤗 Transformers ライブラリのすべての機能を取得できるようにするためです。他の2つの制約は、
-`PretrainedConfig` が設定しているフィールド以外にも多くのフィールドを持っていることから来ています。
-`from_pretrained` メソッドで設定を再ロードする場合、これらのフィールドはあなたの設定に受け入れられ、
-その後、親クラスに送信される必要があります。
-
-設定の `model_type` を定義すること（ここでは `model_type="resnet"`）は、
-自動クラスにモデルを登録したい場合を除いては必須ではありません（最後のセクションを参照）。
-
-これで、ライブラリの他のモデル設定と同様に、設定を簡単に作成して保存できます。
-以下は、resnet50d 設定を作成して保存する方法の例です：
-
-
-```py
-resnet50d_config = ResnetConfig(block_type="bottleneck", stem_width=32, stem_type="deep", avg_down=True)
-resnet50d_config.save_pretrained("custom-resnet")
-```
-
-これにより、`custom-resnet` フォルダ内に `config.json` という名前のファイルが保存されます。その後、`from_pretrained` メソッドを使用して構成を再ロードできます。
-
-
-```py
-resnet50d_config = ResnetConfig.from_pretrained("custom-resnet")
-```
-
-また、[`PretrainedConfig`] クラスの他のメソッドを使用することもできます。たとえば、[`~PretrainedConfig.push_to_hub`] を使用して、設定を直接 Hub にアップロードできます。
-
-## Writing a custom model
-
-ResNet の設定ができたので、モデルを書き始めることができます。実際には2つのモデルを書きます。1つはバッチの画像から隠れた特徴を抽出するモデル（[`BertModel`] のようなもの）で、もう1つは画像分類に適したモデル（[`BertForSequenceClassification`] のようなもの）です。
-
-前述したように、この例をシンプルに保つために、モデルの緩いラッパーのみを書きます。このクラスを書く前に行う必要がある唯一のことは、ブロックタイプと実際のブロッククラスの間のマップです。その後、すべてを `ResNet` クラスに渡して設定からモデルを定義します：
-
-```py
-from transformers import PreTrainedModel
-from timm.models.resnet import BasicBlock, Bottleneck, ResNet
-from .configuration_resnet import ResnetConfig
-
-
-BLOCK_MAPPING = {"basic": BasicBlock, "bottleneck": Bottleneck}
-
-
-class ResnetModel(PreTrainedModel):
-    config_class = ResnetConfig
-
-    def __init__(self, config):
-        super().__init__(config)
-        block_layer = BLOCK_MAPPING[config.block_type]
-        self.model = ResNet(
-            block_layer,
-            config.layers,
-            num_classes=config.num_classes,
-            in_chans=config.input_channels,
-            cardinality=config.cardinality,
-            base_width=config.base_width,
-            stem_width=config.stem_width,
-            stem_type=config.stem_type,
-            avg_down=config.avg_down,
-        )
-
-    def forward(self, tensor):
-        return self.model.forward_features(tensor)
-```
-
-画像を分類するモデルの場合、forwardメソッドを変更するだけです：
-
-```py
-import torch
-
-
-class ResnetModelForImageClassification(PreTrainedModel):
-    config_class = ResnetConfig
-
-    def __init__(self, config):
-        super().__init__(config)
-        block_layer = BLOCK_MAPPING[config.block_type]
-        self.model = ResNet(
-            block_layer,
-            config.layers,
-            num_classes=config.num_classes,
-            in_chans=config.input_channels,
-            cardinality=config.cardinality,
-            base_width=config.base_width,
-            stem_width=config.stem_width,
-            stem_type=config.stem_type,
-            avg_down=config.avg_down,
-        )
-
-    def forward(self, tensor, labels=None):
-        logits = self.model(tensor)
-        if labels is not None:
-            loss = torch.nn.cross_entropy(logits, labels)
-            return {"loss": loss, "logits": logits}
-        return {"logits": logits}
-```
-
-両方の場合、`PreTrainedModel`から継承し、`config`を使用してスーパークラスの初期化を呼び出します（通常の`torch.nn.Module`を書くときのような感じです）。
-`config_class`を設定する行は必須ではありませんが、（最後のセクションを参照）、モデルを自動クラスに登録したい場合に使用できます。
-
-<Tip>
-
-モデルがライブラリ内のモデルと非常に似ている場合、このモデルと同じ構成を再利用できます。
-
-</Tip>
-
-モデルが返す内容は何でも構いませんが、ラベルが渡されるときに損失を含む辞書を返す（`ResnetModelForImageClassification`のように行ったもの）と、
-モデルを[`Trainer`]クラス内で直接使用できるようになります。独自のトレーニングループまたは他のライブラリを使用する予定である限り、
-別の出力形式を使用することも問題ありません。
-
-さて、モデルクラスができたので、1つ作成しましょう：
-
-```py
-resnet50d = ResnetModelForImageClassification(resnet50d_config)
-```
-
-再度、[`PreTrainedModel`]のいずれかのメソッド、例えば[`~PreTrainedModel.save_pretrained`]や
-[`~PreTrainedModel.push_to_hub`]などを使用できます。次のセクションでは、モデルの重みをコードと一緒に
-Hugging Face Hub にプッシュする方法を見てみます。
-しかし、まずはモデル内に事前学習済みの重みをロードしましょう。
-
-独自のユースケースでは、おそらく独自のデータでカスタムモデルをトレーニングすることになるでしょう。
-このチュートリアルではスピードアップのために、resnet50dの事前学習済みバージョンを使用します。
-私たちのモデルはそれをラップするだけなので、これらの重みを転送するのは簡単です：
-
-
-```py
-import timm
-
-pretrained_model = timm.create_model("resnet50d", pretrained=True)
-resnet50d.model.load_state_dict(pretrained_model.state_dict())
-```
-
-さて、[`~PreTrainedModel.save_pretrained`]または[`~PreTrainedModel.push_to_hub`]を実行したときに、
-モデルのコードが保存されるようにする方法を見てみましょう。
-
-## Sending the code to the Hub
-
-<Tip warning={true}>
-
-このAPIは実験的であり、次のリリースでわずかな変更があるかもしれません。
-
-</Tip>
-
-まず、モデルが`.py`ファイルに完全に定義されていることを確認してください。
-ファイルは相対インポートを他のファイルに依存できますが、すべてのファイルが同じディレクトリにある限り（まだこの機能ではサブモジュールはサポートしていません）、問題ありません。
-この例では、現在の作業ディレクトリ内に名前が「resnet_model」のフォルダを作成し、その中に`modeling_resnet.py`ファイルと`configuration_resnet.py`ファイルを定義します。
-構成ファイルには`ResnetConfig`のコードが含まれ、モデリングファイルには`ResnetModel`と`ResnetModelForImageClassification`のコードが含まれています。
-
-
-```
-.
-└── resnet_model
-    ├── __init__.py
-    ├── configuration_resnet.py
-    └── modeling_resnet.py
-```
-
-`__init__.py`は空であっても問題ありません。Pythonが`resnet_model`をモジュールとして検出できるようにするために存在します。
-
-<Tip warning={true}>
-
-ライブラリからモデリングファイルをコピーする場合、ファイルの先頭にあるすべての相対インポートを`transformers`パッケージからインポートに置き換える必要があります。
-
-</Tip>
-
-既存の設定やモデルを再利用（またはサブクラス化）できることに注意してください。
-
-コミュニティとモデルを共有するために、次の手順に従ってください：まず、新しく作成したファイルからResNetモデルと設定をインポートします：
-
-
-```py
-from resnet_model.configuration_resnet import ResnetConfig
-from resnet_model.modeling_resnet import ResnetModel, ResnetModelForImageClassification
-```
-
-次に、`save_pretrained`メソッドを使用してこれらのオブジェクトのコードファイルをコピーし、特定のAutoクラス（特にモデルの場合）に正しく登録するようライブラリに指示する必要があります。次のように実行します：
-
-```py
-ResnetConfig.register_for_auto_class()
-ResnetModel.register_for_auto_class("AutoModel")
-ResnetModelForImageClassification.register_for_auto_class("AutoModelForImageClassification")
-```
-
-注意: 設定については自動クラスを指定する必要はありません（設定用の自動クラスは1つしかなく、[`AutoConfig`]です）が、
-モデルについては異なります。カスタムモデルは多くの異なるタスクに適している可能性があるため、
-モデルが正確な自動クラスのうちどれに適しているかを指定する必要があります。
-
-次に、前述のように設定とモデルを作成しましょう：
-
-```py
-resnet50d_config = ResnetConfig(block_type="bottleneck", stem_width=32, stem_type="deep", avg_down=True)
-resnet50d = ResnetModelForImageClassification(resnet50d_config)
-
-pretrained_model = timm.create_model("resnet50d", pretrained=True)
-resnet50d.model.load_state_dict(pretrained_model.state_dict())
-```
-
-モデルをHubに送信するには、ログインしていることを確認してください。ターミナルで次のコマンドを実行します：
-
-```bash
-huggingface-cli login
-```
-
-またはノートブックから：
-
-```py
-from huggingface_hub import notebook_login
-
-notebook_login()
-```
-
-次に、次のようにして、独自の名前空間にプッシュできます（または、メンバーである組織にプッシュできます）：
-
-```py
-resnet50d.push_to_hub("custom-resnet50d")
-```
-
-モデリングの重みとJSON形式の構成に加えて、このフォルダー「custom-resnet50d」内のモデリングおよび構成「.py」ファイルもコピーされ、結果はHubにアップロードされました。結果はこの[model repo](https://huggingface.co/sgugger/custom-resnet50d)で確認できます。
-
-詳細については、[Hubへのプッシュ方法](model_sharing)を参照してください。
-
-## Using a model with custom code
-
-自動クラスと `from_pretrained` メソッドを使用して、リポジトリ内のカスタムコードファイルと共に任意の構成、モデル、またはトークナイザを使用できます。 Hubにアップロードされるすべてのファイルとコードはマルウェアのスキャンが実施されます（詳細は[Hubセキュリティ](https://huggingface.co/docs/hub/security#malware-scanning)ドキュメンテーションを参照してください）、しかし、依然として悪意のあるコードを実行しないために、モデルコードと作者を確認する必要があります。
-`trust_remote_code=True` を設定してカスタムコードを持つモデルを使用できます：
-
-
-```py
-from transformers import AutoModelForImageClassification
-
-model = AutoModelForImageClassification.from_pretrained("sgugger/custom-resnet50d", trust_remote_code=True)
-```
-
-コミットハッシュを「revision」として渡すことも強く推奨されています。これにより、モデルの作者がコードを悪意のある新しい行で更新しなかったことを確認できます（モデルの作者を完全に信頼している場合を除きます）。
-
-
-```py
-commit_hash = "ed94a7c6247d8aedce4647f00f20de6875b5b292"
-model = AutoModelForImageClassification.from_pretrained(
-    "sgugger/custom-resnet50d", trust_remote_code=True, revision=commit_hash
-)
-```
-
-モデルリポジトリのコミット履歴をブラウジングする際には、任意のコミットのコミットハッシュを簡単にコピーできるボタンがあります。
-
-## Registering a model with custom code to the auto classes
-
-🤗 Transformersを拡張するライブラリを作成している場合、独自のモデルを含めるために自動クラスを拡張したい場合があります。
-これはコードをHubにプッシュすることとは異なり、ユーザーはカスタムモデルを取得するためにあなたのライブラリをインポートする必要があります
-（Hubからモデルコードを自動的にダウンロードするのとは対照的です）。
-
-構成に既存のモデルタイプと異なる `model_type` 属性がある限り、またあなたのモデルクラスが適切な `config_class` 属性を持っている限り、
-次のようにそれらを自動クラスに追加できます：
-
-```py
-from transformers import AutoConfig, AutoModel, AutoModelForImageClassification
-
-AutoConfig.register("resnet", ResnetConfig)
-AutoModel.register(ResnetConfig, ResnetModel)
-AutoModelForImageClassification.register(ResnetConfig, ResnetModelForImageClassification)
-```
-
-注意: `AutoConfig` にカスタム設定を登録する際の最初の引数は、カスタム設定の `model_type` と一致する必要があります。
-また、任意の自動モデルクラスにカスタムモデルを登録する際の最初の引数は、それらのモデルの `config_class` と一致する必要があります。

docs/source/ja/custom_tools.md

@@ -1,764 +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>
-
-トランスフォーマーのコンテキストでツールとエージェントが何であるかを知らない場合、
-まず[Transformers Agents](transformers_agents)ページをお読みいただくことをお勧めします。
-
-</Tip>
-
-<Tip warning={true}>
-
-Transformers Agentsは実験的なAPIであり、いつでも変更される可能性があります。
-エージェントによって返される結果は、APIや基礎となるモデルが変更される可能性があるため、変化することがあります。
-
-</Tip>
-
-カスタムツールとプロンプトを作成し、使用することは、エージェントを強化し、新しいタスクを実行させるために非常に重要です。
-このガイドでは、以下の内容を説明します：
-
-- プロンプトのカスタマイズ方法
-- カスタムツールの使用方法
-- カスタムツールの作成方法
-
-## Customizing the prompt
-
-[Transformers Agents](transformers_agents)で説明されているように、エージェントは[`~Agent.run`]および[`~Agent.chat`]モードで実行できます。
-`run`モードと`chat`モードの両方は同じロジックに基づいています。
-エージェントを駆動する言語モデルは、長いプロンプトに基づいて条件付けられ、
-次のトークンを生成して停止トークンに達するまでプロンプトを完了します。
-両者の唯一の違いは、`chat`モードの間にプロンプトが前のユーザーの入力とモデルの生成と共に拡張されることです。
-これにより、エージェントは過去の対話にアクセスでき、エージェントにあたかもメモリがあるかのように見えます。
-
-### Structure of the prompt
-
-プロンプトがどのように構築され、どのように最適化できるかを理解するために、プロンプトは大まかに4つの部分に分かれています。
-
-1. イントロダクション：エージェントの振る舞い、ツールの概念の説明。
-2. すべてのツールの説明。これはユーザーによって定義/選択されたツールでランタイム時に動的に置換される`<<all_tools>>`トークンによって定義されます。
-3. タスクとその解決策の一連の例。
-4. 現在の例と解決策の要求。
-
-各部分をよりよく理解するために、`run`プロンプトがどのように見えるかの簡略版を見てみましょう：
-
-````text
-タスクを実行するために、Pythonのシンプルなコマンドのシリーズを考えてくることがあるでしょう。
-[...]
-意味がある場合は、中間結果を表示することができます。
-
-ツール：
-- document_qa：これはドキュメント（pdf）に関する質問に答えるツールです。情報を含むドキュメントである `document` と、ドキュメントに関する質問である `question` を受け取り、質問に対する回答を含むテキストを返します。
-- image_captioner：これは画像の説明を生成するツールです。キャプションにする画像である `image` と、説明を含む英語のテキストを返すテキストを受け取ります。
-[...]
-
-タスク: "変数 `question` に関する質問に答えるための画像について回答してください。質問はフランス語です。"
-
-次のツールを使用します：質問を英語に翻訳するための `translator`、そして入力画像に関する質問に答えるための `image_qa`。
-
-回答：
-```py
-translated_question = translator(question=question, src_lang="French", tgt_lang="English")
-print(f"The translated question is {translated_question}.")
-answer = image_qa(image=image, question=translated_question)
-print(f"The answer is {answer}")
-```
-
-タスク：「`document`内で最年長の人物を特定し、その結果をバナーとして表示する。」
-
-以下のツールを使用します：`document_qa`を使用してドキュメント内で最年長の人物を見つけ、その回答に従って`image_generator`を使用して画像を生成します。
-
-回答：
-```py
-answer = document_qa(document, question="What is the oldest person?")
-print(f"The answer is {answer}.")
-image = image_generator("A banner showing " + answer)
-```
-
-[...]
-タスク: "川と湖の絵を描いてください"
-
-以下のものを使用します
-````
-
-導入部分（"Tools:"の前のテキスト）は、モデルの振る舞いと実行すべきタスクを正確に説明しています。
-この部分はおそらくエージェントが常に同じ方法で振る舞う必要があるため、カスタマイズする必要はありません。
-
-2番目の部分（"Tools"の下の箇条書き）は、`run`または`chat`を呼び出すたびに動的に追加されます。
-`agent.toolbox`内のツールの数と同じ数の箇条書きがあり、それぞれの箇条書きにはツールの名前と説明が含まれています。
-
-```text
-- <tool.name>: <tool.description>
-```
-
-もうすぐ確認しましょう。 `document_qa` ツールを読み込んで名前と説明を出力します。
-
-```py
-from transformers import load_tool
-
-document_qa = load_tool("document-question-answering")
-print(f"- {document_qa.name}: {document_qa.description}")
-```
-
-which gives:
-```text
-- document_qa: This is a tool that answers a question about a document (pdf). It takes an input named `document` which should be the document containing the information, as well as a `question` that is the question about the document. It returns a text that contains the answer to the question.
-```
-
-ツール説明:
-このツールは、2つのパートから成り立っています。最初のパートでは、ツールが何を行うかを説明し、2番目のパートでは入力引数と戻り値がどのように期待されるかを述べています。
-
-良いツール名とツールの説明は、エージェントが正しく使用するために非常に重要です。エージェントがツールについて持っている唯一の情報は、その名前と説明です。したがって、ツール名と説明の両方が正確に記述され、ツールボックス内の既存のツールのスタイルに合致することを確認する必要があります。特に、説明にはコードスタイルで名前で期待されるすべての引数が言及され、期待される型とそれらが何であるかの説明も含めるべきです。
-
-<Tip>
-
-キュレートされたTransformersツールの命名と説明を確認して、ツールがどのような名前と説明を持つべきかを理解するのに役立ちます。
-すべてのツールは[`Agent.toolbox`]プロパティで確認できます。
-
-</Tip>
-
-
-カスタマイズされた例：
-ツールの使い方をエージェントに正確に示す一連の例が含まれています。これらの例は、エージェントが実際に正確で実行可能なコードを生成する可能性を最大化するように書かれているため、非常に重要です。大規模な言語モデルは、プロンプト内のパターンを認識し、新しいデータを使用してそのパターンを繰り返すことに非常に優れています。したがって、実践で正しい実行可能なコードを生成するエージェントの可能性を最大化するように、これらの例は書かれている必要があります。
-
-以下は、一つの例です：
-
-````text
-Task: "Identify the oldest person in the `document` and create an image showcasing the result as a banner."
-
-I will use the following tools: `document_qa` to find the oldest person in the document, then `image_generator` to generate an image according to the answer.
-
-Answer:
-```py
-answer = document_qa(document, question="What is the oldest person?")
-print(f"The answer is {answer}.")
-image = image_generator("A banner showing " + answer)
-```
-
-````
-
-パターン：モデルが繰り返しを行うように指示されるパターンには、3つの部分があります。
-タスクの声明、エージェントの意図した動作の説明、そして最後に生成されるコードです。
-プロンプトの一部であるすべての例には、この正確なパターンがあり、エージェントが新しいトークンを生成する際にも
-同じパターンを再現することを確認しています。
-
-プロンプトの例はTransformersチームによって厳選され、一連の問題ステートメントで厳密に評価されます。
-これにより、エージェントのプロンプトがエージェントの実際の使用ケースを解決するためにできるだけ優れたものになります。
-
-プロンプトの最後の部分に対応しています：
-
-[こちら](https://github.com/huggingface/transformers/blob/main/src/transformers/tools/evaluate_agent.py)の問題ステートメントで厳密に評価される、エージェントのプロンプトができるだけ優れたものになるように
-慎重に選定されたプロンプト例を提供しています。
-
-```text
-Task: "Draw me a picture of rivers and lakes"
-
-I will use the following
-```
-
-
-これがエージェントに完成させるための最終的で未完成の例です。未完成の例は、実際のユーザー入力に基づいて動的に作成されます。上記の例では、ユーザーが次のように実行しました：
-
-```py
-agent.run("Draw me a picture of rivers and lakes")
-```
-
-ユーザーの入力 - つまり、タスク："川と湖の絵を描いてください"は、以下のようなプロンプトテンプレートに変換されます："タスク：<task> \n\n 次に私は以下を使用します"。
-この文は、エージェントが条件付けられたプロンプトの最終行を構成し、したがってエージェントに対して前の例とまったく同じ方法で例を終了するよう強く影響します。
-
-詳細には立ち入りませんが、チャットテンプレートは同じプロンプト構造を持ち、例はわずかに異なるスタイルを持っています。例：
-
-````text
-[...]
-
-=====
-
-Human: Answer the question in the variable `question` about the image stored in the variable `image`.
-
-Assistant: I will use the tool `image_qa` to answer the question on the input image.
-
-```py
-answer = image_qa(text=question, image=image)
-print(f"The answer is {answer}")
-```
-
-Human: I tried this code, it worked but didn't give me a good result. The question is in French
-
-Assistant: In this case, the question needs to be translated first. I will use the tool `translator` to do this.
-
-```py
-translated_question = translator(question=question, src_lang="French", tgt_lang="English")
-print(f"The translated question is {translated_question}.")
-answer = image_qa(text=translated_question, image=image)
-print(f"The answer is {answer}")
-```
-
-=====
-
-[...]
-````
-
-*Human:* `run`プロンプトの例とは対照的に、各`chat`プロンプトの例には*Human*と*Assistant*の間で1つ以上のやりとりがあります。各やりとりは、`run`プロンプトの例と同様の構造になっています。ユーザーの入力は*Human:*の後ろに追加され、エージェントにはコードを生成する前に何を行う必要があるかを最初に生成するように指示されます。やりとりは以前のやりとりに基づいて行われることがあり、ユーザーが「I tried **this** code」と入力したように、以前に生成されたエージェントのコードを参照できます。
-
-*Assistant:* `.chat`を実行すると、ユーザーの入力または*タスク*が未完了の形式に変換されます：
-
-```text
-Human: <user-input>\n\nAssistant:
-```
-
-以下のエージェントが完了するコマンドについて説明します。 `run` コマンドとは対照的に、`chat` コマンドは完了した例をプロンプトに追加します。そのため、次の `chat` ターンのためにエージェントにより多くの文脈を提供します。
-
-さて、プロンプトの構造がわかったところで、どのようにカスタマイズできるかを見てみましょう！
-
-### Writing good user inputs
-
-大規模な言語モデルはユーザーの意図を理解する能力がますます向上していますが、エージェントが正しいタスクを選択するのを助けるために、できるだけ正確に記述することが非常に役立ちます。できるだけ正確であるとは何を意味するのでしょうか？
-
-エージェントは、プロンプトでツール名とその説明のリストを見ています。ツールが追加されるほど、エージェントが正しいツールを選択するのが難しくなり、正しいツールの連続を選択するのはさらに難しくなります。共通の失敗例を見てみましょう。ここではコードのみを返すことにします。
-
-
-```py
-from transformers import HfAgent
-
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
-
-agent.run("Show me a tree", return_code=True)
-```
-
-gives:
-
-```text
-==Explanation from the agent==
-I will use the following tool: `image_segmenter` to create a segmentation mask for the image.
-
-
-==Code generated by the agent==
-mask = image_segmenter(image, prompt="tree")
-```
-
-これはおそらく私たちが望んでいたものではないでしょう。代わりに、木の画像が生成されることがより可能性が高いです。
-特定のツールを使用するようエージェントを誘導するために、ツールの名前や説明に含まれている重要なキーワードを使用することは非常に役立ちます。さて、詳しく見てみましょう。
-
-```py
-agent.toolbox["image_generator"].description
-```
-
-```text
-'This is a tool that creates an image according to a prompt, which is a text description. It takes an input named `prompt` which contains the image description and outputs an image.
-```
-
-名前と説明文には、キーワード「画像」、「プロンプト」、「作成」、および「生成」が使用されています。これらの言葉を使用することで、ここでの動作がより効果的になる可能性が高いです。プロンプトを少し詳細に調整しましょう。
-
-```py
-agent.run("Create an image of a tree", return_code=True)
-```
-
-gives:
-```text
-==Explanation from the agent==
-I will use the following tool `image_generator` to generate an image of a tree.
-
-
-==Code generated by the agent==
-image = image_generator(prompt="tree")
-```
-
-簡単に言うと、エージェントがタスクを正確に適切なツールにマッピングできない場合は、ツールの名前や説明の最も関連性のあるキーワードを調べて、タスクリクエストをそれに合わせて洗練させてみてください。
-
-
-### Customizing the tool descriptions
-
-以前にも見たように、エージェントは各ツールの名前と説明にアクセスできます。ベースのツールは非常に正確な名前と説明を持っているはずですが、特定のユースケースに合わせてツールの説明や名前を変更することが役立つかもしれません。これは、非常に類似した複数のツールを追加した場合や、特定のドメイン（たとえば、画像生成や変換など）でエージェントを使用する場合に特に重要になるかもしれません。
-
-よくある問題は、エージェントが画像生成タスクに頻繁に使用される場合、画像生成と画像変換/修正を混同することです。
-
-例：
-
-```py
-agent.run("Make an image of a house and a car", return_code=True)
-```
-
-returns
-```text
-==Explanation from the agent== 
-I will use the following tools `image_generator` to generate an image of a house and `image_transformer` to transform the image of a car into the image of a house.
-
-==Code generated by the agent==
-house_image = image_generator(prompt="A house")
-car_image = image_generator(prompt="A car")
-house_car_image = image_transformer(image=car_image, prompt="A house")
-```
-
-これはおそらく私たちがここで望んでいる正確なものではないようです。エージェントは「image_generator」と「image_transformer」の違いを理解するのが難しいようで、しばしば両方を一緒に使用します。
-
-ここでエージェントをサポートするために、"image_transformer"のツール名と説明を変更して、少し"image"や"prompt"から切り離してみましょう。代わりにそれを「modifier」と呼びましょう：
-
-```py
-agent.toolbox["modifier"] = agent.toolbox.pop("image_transformer")
-agent.toolbox["modifier"].description = agent.toolbox["modifier"].description.replace(
-    "transforms an image according to a prompt", "modifies an image"
-)
-```
-
-「変更」は、上記のプロンプトに新しい画像プロセッサを使用する強力な手がかりです。それでは、もう一度実行してみましょう。
-
-
-```py
-agent.run("Make an image of a house and a car", return_code=True)
-```
-
-Now we're getting:
-```text
-==Explanation from the agent==
-I will use the following tools: `image_generator` to generate an image of a house, then `image_generator` to generate an image of a car.
-
-
-==Code generated by the agent==
-house_image = image_generator(prompt="A house")
-car_image = image_generator(prompt="A car")
-```
-
-これは、私たちが考えていたものに確実に近づいています！ただし、家と車を同じ画像に含めたいと考えています。タスクを単一の画像生成に向けることで、より適切な方向に進めるはずです：
-
-```py
-agent.run("Create image: 'A house and car'", return_code=True)
-```
-
-```text
-==Explanation from the agent==
-I will use the following tool: `image_generator` to generate an image.
-
-
-==Code generated by the agent==
-image = image_generator(prompt="A house and car")
-```
-
-<Tip warning={true}>
-
-エージェントは、特に複数のオブジェクトの画像を生成するなど、やや複雑なユースケースに関しては、まだ多くのユースケースに対して脆弱です。
-エージェント自体とその基礎となるプロンプトは、今後数ヶ月でさらに改善され、さまざまなユーザーの入力に対してエージェントがより頑健になるようになります。
-
-</Tip>
-
-### Customizing the whole project
-
-ユーザーに最大限の柔軟性を提供するために、[上記](#structure-of-the-prompt)で説明されたプロンプトテンプレート全体をユーザーが上書きできます。この場合、カスタムプロンプトには導入セクション、ツールセクション、例セクション、未完了の例セクションが含まれていることを確認してください。`run` プロンプトテンプレートを上書きしたい場合、以下のように行うことができます:
-
-
-```py
-template = """ [...] """
-
-agent = HfAgent(your_endpoint, run_prompt_template=template)
-```
-
-<Tip warning={true}>
-
-`<<all_tools>>` 文字列と `<<prompt>>` は、エージェントが使用できるツールを認識し、ユーザーのプロンプトを正しく挿入できるように、`template` のどこかに定義されていることを確認してください。
-
-</Tip>
-
-同様に、`chat` プロンプトテンプレートを上書きすることもできます。なお、`chat` モードでは常に以下の形式で交換が行われます：
-
-上記のテキストの上に日本語の翻訳を提供してください。Markdownコードとして書いてください。
-
-
-```text
-Human: <<task>>
-
-Assistant:
-```
-
-したがって、カスタム`chat`プロンプトテンプレートの例もこのフォーマットを使用することが重要です。以下のように、インスタンス化時に`chat`テンプレートを上書きできます。
-
-```
-template = """ [...] """
-
-agent = HfAgent(url_endpoint=your_endpoint, chat_prompt_template=template)
-```
-
-<Tip warning={true}>
-
-`<<all_tools>>` という文字列が `template` 内で定義されていることを確認してください。これにより、エージェントは使用可能なツールを把握できます。
-
-</Tip>
-
-両方の場合、プロンプトテンプレートの代わりに、コミュニティの誰かがホストしたテンプレートを使用したい場合は、リポジトリIDを渡すことができます。デフォルトのプロンプトは、[このリポジトリ](https://huggingface.co/datasets/huggingface-tools/default-prompts) にありますので、参考になります。
-
-カスタムプロンプトをHubのリポジトリにアップロードしてコミュニティと共有する場合は、次のことを確認してください：
-- データセットリポジトリを使用すること
-- `run` コマンド用のプロンプトテンプレートを `run_prompt_template.txt` という名前のファイルに配置すること
-- `chat` コマンド用のプロンプトテンプレートを `chat_prompt_template.txt` という名前のファイルに配置すること
-
-## Using custom tools
-
-このセクションでは、画像生成に特化した2つの既存のカスタムツールを利用します：
-
-- [huggingface-tools/image-transformation](https://huggingface.co/spaces/huggingface-tools/image-transformation) をより多くの画像変更を可能にするために [diffusers/controlnet-canny-tool](https://huggingface.co/spaces/diffusers/controlnet-canny-tool) に置き換えます。
-- 画像のアップスケーリング用の新しいツールをデフォルトのツールボックスに追加します：[diffusers/latent-upscaler-tool](https://huggingface.co/spaces/diffusers/latent-upscaler-tool) は既存の画像変換ツールを置き換えます。
-
-便利な [`load_tool`] 関数を使用してカスタムツールをロードします：
-
-```py
-from transformers import load_tool
-
-controlnet_transformer = load_tool("diffusers/controlnet-canny-tool")
-upscaler = load_tool("diffusers/latent-upscaler-tool")
-```
-
-エージェントにカスタムツールを追加すると、ツールの説明と名前がエージェントのプロンプトに自動的に含まれます。したがって、エージェントがカスタムツールの使用方法を理解できるように、カスタムツールには適切に記述された説明と名前が必要です。
-
-`controlnet_transformer`の説明と名前を見てみましょう。
-
-最初に、便利な[`load_tool`]関数を使用してカスタムツールをロードします。
-
-```py
-print(f"Description: '{controlnet_transformer.description}'")
-print(f"Name: '{controlnet_transformer.name}'")
-```
-
-gives 
-```text
-Description: 'This is a tool that transforms an image with ControlNet according to a prompt. 
-It takes two inputs: `image`, which should be the image to transform, and `prompt`, which should be the prompt to use to change it. It returns the modified image.'
-Name: 'image_transformer'
-```
-
-名前と説明は正確であり、[厳選されたツール](./transformers_agents#a-curated-set-of-tools)のスタイルに合っています。
-
-次に、`controlnet_transformer`と`upscaler`を使ってエージェントをインスタンス化します。
-
-```py
-tools = [controlnet_transformer, upscaler]
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder", additional_tools=tools)
-
-```
-
-以下のコマンドは、以下の情報を提供します：
-
-
-```text
-image_transformer has been replaced by <transformers_modules.diffusers.controlnet-canny-tool.bd76182c7777eba9612fc03c0
-8718a60c0aa6312.image_transformation.ControlNetTransformationTool object at 0x7f1d3bfa3a00> as provided in `additional_tools`
-```
-
-一連の厳選されたツールにはすでに `image_transformer` ツールがあり、これをカスタムツールで置き換えます。
-
-<Tip>
-
-既存のツールを上書きすることは、特定のタスクに既存のツールをまったく同じ目的で使用したい場合に有益であることがあります。
-なぜなら、エージェントはその特定のタスクの使用方法に精通しているからです。この場合、カスタムツールは既存のツールとまったく同じAPIに従うか、そのツールを使用するすべての例が更新されるようにプロンプトテンプレートを適応させる必要があります。
-
-</Tip>
-
-アップスケーラーツールには `image_upscaler` という名前が付けられ、これはデフォルトのツールボックスにはまだ存在しないため、単にツールのリストに追加されます。
-エージェントが現在使用可能なツールボックスを確認するには、`agent.toolbox` 属性を使用できます。
-
-```py
-print("\n".join([f"- {a}" for a in agent.toolbox.keys()]))
-```
-
-```text
-- document_qa
-- image_captioner
-- image_qa
-- image_segmenter
-- transcriber
-- summarizer
-- text_classifier
-- text_qa
-- text_reader
-- translator
-- image_transformer
-- text_downloader
-- image_generator
-- video_generator
-- image_upscaler
-```
-
-注意: `image_upscaler` がエージェントのツールボックスの一部となったことに注目してください。
-
-それでは、新しいツールを試してみましょう！[Transformers Agents Quickstart](./transformers_agents#single-execution-run) で生成した画像を再利用します。
-
-
-```py
-from diffusers.utils import load_image
-
-image = load_image(
-    "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png"
-)
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200> 
-
-美しい冬の風景にこの画像を変身させましょう：
-
-```py
-image = agent.run("Transform the image: 'A frozen lake and snowy forest'", image=image)
-```
-
-```text
-==Explanation from the agent==
-I will use the following tool: `image_transformer` to transform the image.
-
-
-==Code generated by the agent==
-image = image_transformer(image, prompt="A frozen lake and snowy forest")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_winter.png" width=200> 
-
-新しい画像処理ツールは、非常に強力な画像の変更を行うことができるControlNetに基づいています。
-デフォルトでは、画像処理ツールはサイズが512x512ピクセルの画像を返します。それを拡大できるか見てみましょう。
-
-
-```py
-image = agent.run("Upscale the image", image)
-```
-
-```text
-==Explanation from the agent==
-I will use the following tool: `image_upscaler` to upscale the image.
-
-
-==Code generated by the agent==
-upscaled_image = image_upscaler(image)
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_winter_upscale.png" width=400> 
-
-
-エージェントは、プロンプト「画像の拡大」を、その説明とツールの名前だけを基に、新たに追加されたアップスケーリングツールに自動的にマッピングし、正しく実行できました。
-
-次に、新しいカスタムツールを作成する方法を見てみましょう。
-
-### Adding new tools
-
-このセクションでは、エージェントに追加できる新しいツールの作成方法を示します。
-
-#### Creating a new tool
-
-まず、ツールの作成から始めましょう。次のコードで、特定のタスクに関してHugging Face Hubで最もダウンロードされたモデルを取得する、あまり役立たないけれども楽しいタスクを追加します。
-
-以下のコードでそれを行うことができます：
-
-
-```python
-from huggingface_hub import list_models
-
-task = "text-classification"
-
-model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-print(model.id)
-```
-
-タスク `text-classification` の場合、これは `'facebook/bart-large-mnli'` を返します。`translation` の場合、`'t5-base'` を返します。
-
-これをエージェントが利用できるツールに変換する方法は何でしょうか？すべてのツールは、主要な属性を保持するスーパークラス `Tool` に依存しています。私たちは、それを継承したクラスを作成します:
-
-
-```python
-from transformers import Tool
-
-
-class HFModelDownloadsTool(Tool):
-    pass
-```
-
-このクラスにはいくつかの必要な要素があります：
-- `name` 属性：これはツール自体の名前に対応し、他のツールと調和するために `model_download_counter` と名付けます。
-- `description` 属性：これはエージェントのプロンプトを埋めるために使用されます。
-- `inputs` と `outputs` 属性：これらを定義することで、Python インタープリターが型に関する賢明な選択を行うのに役立ち、ツールをHubにプッシュする際にgradio-demoを生成できるようになります。これらは、予想される値のリストであり、`text`、`image`、または`audio`になることがあります。
-- `__call__` メソッド：これには推論コードが含まれています。これは上記で試したコードです！
-
-こちらが現在のクラスの外観です：
-
-
-```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 takes the name of the category (such as text-classification, depth-estimation, etc), and "
-        "returns the name of the checkpoint."
-    )
-
-    inputs = ["text"]
-    outputs = ["text"]
-
-    def __call__(self, task: str):
-        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-        return model.id
-```
-
-さて、今度はツールが使えるようになりました。このツールをファイルに保存し、メインスクリプトからインポートしましょう。このファイルを `model_downloads.py` という名前にし、結果のインポートコードは次のようになります：
-
-以下は、現在のクラスの外観です：
-
-
-```python
-from model_downloads import HFModelDownloadsTool
-
-tool = HFModelDownloadsTool()
-```
-
-他の人々に利益をもたらし、より簡単な初期化のために、それをHubにあなたの名前空間でプッシュすることをお勧めします。これを行うには、`tool` 変数で `push_to_hub` を呼び出すだけです：
-
-```python
-tool.push_to_hub("hf-model-downloads")
-```
-
-エージェントがツールを使用する方法について、最終ステップを見てみましょう。
-
-#### Having the agent use the tool
-
-Hubにあるツールがあります。これは次のようにインスタンス化できます（ユーザー名をツールに合わせて変更してください）:
-
-```python
-from transformers import load_tool
-
-tool = load_tool("lysandre/hf-model-downloads")
-```
-
-エージェントで使用するためには、エージェントの初期化メソッドの `additional_tools` パラメータにそれを渡すだけです：
-
-
-```python
-from transformers import HfAgent
-
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder", additional_tools=[tool])
-
-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?"
-)
-```
-which outputs the following:
-```text
-==Code generated by the agent==
-model = model_download_counter(task="text-to-video")
-print(f"The model with the most downloads is {model}.")
-audio_model = text_reader(model)
-
-
-==Result==
-The model with the most downloads is damo-vilab/text-to-video-ms-1.7b.
-```
-
-以下のテキストは、次のオーディオを生成します。
-
-
-
-**Audio**                                                                                                                                            |
-|------------------------------------------------------------------------------------------------------------------------------------------------------|
-| <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/damo.wav" type="audio/wav"/> |
-
-
-<Tip>
-
-特定のLLMに依存することがあり、うまく機能させるためには非常に正確なプロンプトが必要なものもあります。ツールの名前と説明を明確に定義することは、エージェントによって活用されるために非常に重要です。
-
-</Tip>
-
-### Replacing existing tools
-
-既存のツールを置き換えるには、新しいアイテムをエージェントのツールボックスに割り当てるだけで行うことができます。以下はその方法です:
-
-```python
-from transformers import HfAgent, load_tool
-
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
-agent.toolbox["image-transformation"] = load_tool("diffusers/controlnet-canny-tool")
-```
-
-<Tip>
-
-他のツールでツールを置き換える際には注意が必要です！これにより、エージェントのプロンプトも調整されます。これは、タスクに適したより良いプロンプトを持っている場合には良いことですが、他のツールが選択される確率が高くなり、定義したツールの代わりに他のツールが選択されることもあるかもしれません。
-
-</Tip>
-
-## Leveraging gradio-tools
-
-[gradio-tools](https://github.com/freddyaboulton/gradio-tools)は、Hugging Face Spacesをツールとして使用することを可能にする強力なライブラリです。既存の多くのSpacesおよびカスタムSpacesを設計することもサポートしています。
-
-我々は、`gradio_tools`を使用して`StableDiffusionPromptGeneratorTool`ツールを活用したいと考えています。このツールは`gradio-tools`ツールキットで提供されており、プロンプトを改善し、より良い画像を生成するために使用します。
-
-まず、`gradio_tools`からツールをインポートし、それをインスタンス化します:
-
-```python
-from gradio_tools import StableDiffusionPromptGeneratorTool
-
-gradio_tool = StableDiffusionPromptGeneratorTool()
-```
-
-そのインスタンスを `Tool.from_gradio` メソッドに渡します：
-
-
-```python
-from transformers import Tool
-
-tool = Tool.from_gradio(gradio_tool)
-```
-
-これからは、通常のカスタムツールと同じようにそれを管理できます。私たちはプロンプトを改善するためにそれを活用します。
-` a rabbit wearing a space suit`:
-
-```python
-from transformers import HfAgent
-
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder", additional_tools=[tool])
-
-agent.run("Generate an image of the `prompt` after improving it.", prompt="A rabbit wearing a space suit")
-```
-
-The model adequately leverages the tool:
-```text
-==Explanation from the agent==
-I will use the following  tools: `StableDiffusionPromptGenerator` to improve the prompt, then `image_generator` to generate an image according to the improved prompt.
-
-
-==Code generated by the agent==
-improved_prompt = StableDiffusionPromptGenerator(prompt)
-print(f"The improved prompt is {improved_prompt}.")
-image = image_generator(improved_prompt)
-```
-
-最終的に画像を生成する前に：
-
-![画像](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png)
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png">
-
-<Tip warning={true}>
-
-gradio-toolsは、さまざまなモダリティを使用する場合でも、*テキスト*の入力と出力が必要です。この実装は画像と音声オブジェクトと連携します。現時点では、これら2つは互換性がありませんが、サポートを向上させるために取り組んでおり、迅速に互換性が向上するでしょう。
-
-</Tip>
-
-## Future compatibility with Langchain
-
-私たちはLangchainを愛しており、非常に魅力的なツールのスイートを持っていると考えています。これらのツールを扱うために、Langchainはさまざまなモダリティで作業する場合でも、*テキスト*の入出力が必要です。これは、オブジェクトのシリアル化バージョン（つまり、ディスクに保存されたバージョン）であることが多いです。
-
-この違いにより、transformers-agentsとlangchain間ではマルチモダリティが処理されていません。
-この制限は将来のバージョンで解決されることを目指しており、熱心なlangchainユーザーからの任意の支援を歓迎します。
-
-私たちはより良いサポートを提供したいと考えています。お手伝いいただける場合は、ぜひ[問題を開いて](https://github.com/huggingface/transformers/issues/new)、お考えのことを共有してください。
-
-

docs/source/ja/fast_tokenizers.md

@@ -1,73 +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.
-
--->
-
-# Use tokenizers from 🤗 Tokenizers
-
-[`PreTrainedTokenizerFast`]は[🤗 Tokenizers](https://huggingface.co/docs/tokenizers)ライブラリに依存しています。🤗 Tokenizersライブラリから取得したトークナイザーは、非常に簡単に🤗 Transformersにロードできます。
-
-具体的な内容に入る前に、まずはいくつかの行でダミーのトークナイザーを作成することから始めましょう：
-
-
-```python
->>> from tokenizers import Tokenizer
->>> from tokenizers.models import BPE
->>> from tokenizers.trainers import BpeTrainer
->>> from tokenizers.pre_tokenizers import Whitespace
-
->>> tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
->>> trainer = BpeTrainer(special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
-
->>> tokenizer.pre_tokenizer = Whitespace()
->>> files = [...]
->>> tokenizer.train(files, trainer)
-```
-
-私たちは今、定義したファイルにトレーニングされたトークナイザーを持っています。これをランタイムで引き続き使用するか、
-将来の再利用のためにJSONファイルに保存することができます。
-
-## Loading directly from the tokenizer object
-
-🤗 Transformersライブラリでこのトークナイザーオブジェクトをどのように活用できるかを見てみましょう。[`PreTrainedTokenizerFast`]クラスは、
-*tokenizer*オブジェクトを引数として受け入れ、簡単にインスタンス化できるようにします。
-
-
-```python
->>> from transformers import PreTrainedTokenizerFast
-
->>> fast_tokenizer = PreTrainedTokenizerFast(tokenizer_object=tokenizer)
-```
-
-このオブジェクトは、🤗 Transformers トークナイザーが共有するすべてのメソッドと一緒に使用できます！詳細については、[トークナイザーページ](main_classes/tokenizer)をご覧ください。
-
-## Loading from a JSON file
-
-JSONファイルからトークナイザーを読み込むには、まずトークナイザーを保存することから始めましょう：
-
-```python
->>> tokenizer.save("tokenizer.json")
-```
-
-このファイルを保存したパスは、`PreTrainedTokenizerFast` の初期化メソッドに `tokenizer_file` パラメータを使用して渡すことができます：
-
-
-```python
->>> from transformers import PreTrainedTokenizerFast
-
->>> fast_tokenizer = PreTrainedTokenizerFast(tokenizer_file="tokenizer.json")
-```
-
-このオブジェクトは、🤗 Transformers トークナイザーが共有するすべてのメソッドと一緒に使用できるようになりました！詳細については、[トークナイザーページ](main_classes/tokenizer)をご覧ください。
-

docs/source/ja/generation_strategies.md

@@ -1,345 +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.
-
--->
-
-# Text generation strategies
-
-テキスト生成は、オープンエンドのテキスト生成、要約、翻訳など、多くの自然言語処理タスクに不可欠です。また、テキストを出力とするさまざまな混在モダリティアプリケーションにも影響を与えており、例えば音声からテキストへの変換や画像からテキストへの変換などがあります。テキストを生成できるいくつかのモデルには、GPT2、XLNet、OpenAI GPT、CTRL、TransformerXL、XLM、Bart、T5、GIT、Whisperが含まれます。
-
-[`~transformers.generation_utils.GenerationMixin.generate`] メソッドを使用して、異なるタスクのテキスト出力を生成するいくつかの例をご紹介します：
-* [テキスト要約](./tasks/summarization#inference)
-* [画像のキャプション](./model_doc/git#transformers.GitForCausalLM.forward.example)
-* [音声の転記](./model_doc/whisper#transformers.WhisperForConditionalGeneration.forward.example)
-
-generateメソッドへの入力は、モデルのモダリティに依存します。これらの入力は、AutoTokenizerやAutoProcessorなどのモデルのプリプロセッサクラスによって返されます。モデルのプリプロセッサが複数の種類の入力を生成する場合は、すべての入力をgenerate()に渡します。各モデルのプリプロセッサについての詳細は、対応するモデルのドキュメンテーションで確認できます。
-
-テキストを生成するためのトークンの選択プロセスはデコーディングとして知られ、`generate()`メソッドが使用するデコーディング戦略をカスタマイズできます。デコーディング戦略を変更することは、訓練可能なパラメータの値を変更しませんが、生成されるテキストの品質に顕著な影響を与えることがあります。これにより、テキスト内の繰り返しを減少させ、より一貫性のあるテキストを生成するのに役立ちます。
-
-このガイドでは以下の内容が説明されています：
-* デフォルトのテキスト生成設定
-* 一般的なデコーディング戦略とその主要なパラメータ
-* 🤗 Hubのあなたのファインチューンモデルとカスタム生成設定の保存と共有
-
-## Default text generation configuration
-
-モデルのデコーディング戦略は、その生成設定で定義されています。[`pipeline`] 内で推論に事前訓練モデルを使用する際には、モデルはデフォルトの生成設定を内部で適用する `PreTrainedModel.generate()` メソッドを呼び出します。デフォルトの設定は、モデルにカスタム設定が保存されていない場合にも使用されます。
-
-モデルを明示的に読み込む場合、それに付属する生成設定を `model.generation_config` を介して確認できます。
-
-```python
->>> from transformers import AutoModelForCausalLM
-
->>> model = AutoModelForCausalLM.from_pretrained("distilgpt2")
->>> model.generation_config
-GenerationConfig {
-    "bos_token_id": 50256,
-    "eos_token_id": 50256,
-}
-```
-
-`model.generation_config` を出力すると、デフォルトの生成設定から異なる値のみが表示され、デフォルトの値はリストされません。
-
-デフォルトの生成設定では、出力のサイズは入力プロンプトとの組み合わせで最大20トークンに制限されており、リソース制限に達しないようにしています。デフォルトのデコーディング戦略は貪欲探索で、最も確率の高いトークンを次のトークンとして選択する最も単純なデコーディング戦略です。多くのタスクや小さな出力サイズの場合、これはうまく機能します。ただし、長い出力を生成するために使用される場合、貪欲探索は高度に繰り返される結果を生成し始めることがあります。
-
-## Customize text generation
-
-`generate` メソッドに直接パラメータとその値を渡すことで、`generation_config` を上書きできます。
-
-```python
->>> my_model.generate(**inputs, num_beams=4, do_sample=True)  # doctest: +SKIP
-```
-
-デフォルトのデコーディング戦略がほとんどのタスクでうまく機能する場合でも、いくつかの設定を微調整できます。一般的に調整されるパラメータには次のものがあります：
-
-- `max_new_tokens`: 生成するトークンの最大数。つまり、出力シーケンスのサイズであり、プロンプト内のトークンは含まれません。
-- `num_beams`: 1よりも大きなビーム数を指定することで、貪欲検索からビームサーチに切り替えることができます。この戦略では、各時間ステップでいくつかの仮説を評価し、最終的に全体のシーケンスに対する最も高い確率を持つ仮説を選択します。これにより、初期の確率が低いトークンで始まる高確率のシーケンスが貪欲検索によって無視されることがなくなります。
-- `do_sample`: このパラメータを`True`に設定すると、多項分布サンプリング、ビームサーチ多項分布サンプリング、Top-Kサンプリング、Top-pサンプリングなどのデコーディング戦略が有効になります。これらの戦略は、各戦略固有の調整を含む単語彙全体の確率分布から次のトークンを選択します。
-- `num_return_sequences`: 各入力に対して返すシーケンス候補の数。これは、複数のシーケンス候補をサポートするデコーディング戦略（ビームサーチやサンプリングのバリエーションなど）にのみ適用されます。貪欲検索や対照的な検索など、単一の出力シーケンスを返すデコーディング戦略では使用できません。
-
-## Save a custom decoding strategy with your model
-
-特定の生成構成で調整したモデルを共有したい場合、以下の手順を実行できます：
-* [`GenerationConfig`] クラスのインスタンスを作成する
-* デコーディング戦略のパラメータを指定する
-* [`GenerationConfig.save_pretrained`] を使用して生成構成を保存し、`config_file_name` 引数を空にすることを忘れないでください
-* `push_to_hub` を `True` に設定して、構成をモデルのリポジトリにアップロードします
-
-```python
->>> from transformers import AutoModelForCausalLM, GenerationConfig
-
->>> model = AutoModelForCausalLM.from_pretrained("my_account/my_model")  # doctest: +SKIP
->>> generation_config = GenerationConfig(
-...     max_new_tokens=50, do_sample=True, top_k=50, eos_token_id=model.config.eos_token_id
-... )
->>> generation_config.save_pretrained("my_account/my_model", push_to_hub=True)  # doctest: +SKIP
-```
-
-1つのディレクトリに複数の生成設定を保存することもでき、[`GenerationConfig.save_pretrained`] の `config_file_name`
-引数を使用します。後で [`GenerationConfig.from_pretrained`] でこれらをインスタンス化できます。これは、1つのモデルに対して複数の生成設定を保存したい場合に便利です
-（例：サンプリングを使用したクリエイティブなテキスト生成用の1つと、ビームサーチを使用した要約用の1つ）。モデルに設定ファイルを追加するには、適切な Hub 権限が必要です。
-
-
-```python
->>> from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, GenerationConfig
-
->>> tokenizer = AutoTokenizer.from_pretrained("t5-small")
->>> model = AutoModelForSeq2SeqLM.from_pretrained("t5-small")
-
->>> translation_generation_config = GenerationConfig(
-...     num_beams=4,
-...     early_stopping=True,
-...     decoder_start_token_id=0,
-...     eos_token_id=model.config.eos_token_id,
-...     pad_token=model.config.pad_token_id,
-... )
-
->>> # Tip: add `push_to_hub=True` to push to the Hub
->>> translation_generation_config.save_pretrained("/tmp", "translation_generation_config.json")
-
->>> # You could then use the named generation config file to parameterize generation
->>> generation_config = GenerationConfig.from_pretrained("/tmp", "translation_generation_config.json")
->>> inputs = tokenizer("translate English to French: Configuration files are easy to use!", return_tensors="pt")
->>> outputs = model.generate(**inputs, generation_config=generation_config)
->>> print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
-['Les fichiers de configuration sont faciles à utiliser!']
-```
-
-## Streaming
-
-`generate()` は、その `streamer` 入力を介してストリーミングをサポートしています。`streamer` 入力は、次のメソッドを持つクラスのインスタンスと互換性があります：`put()` と `end()`。内部的には、`put()` は新しいトークンをプッシュするために使用され、`end()` はテキスト生成の終了をフラグ付けするために使用されます。
-
-<Tip warning={true}>
-
-ストリーマークラスのAPIはまだ開発中であり、将来変更される可能性があります。
-
-</Tip>
-
-実際には、さまざまな目的に対して独自のストリーミングクラスを作成できます！また、使用できる基本的なストリーミングクラスも用意されています。例えば、[`TextStreamer`] クラスを使用して、`generate()` の出力を画面に単語ごとにストリームすることができます：
-
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
-
->>> tok = AutoTokenizer.from_pretrained("gpt2")
->>> model = AutoModelForCausalLM.from_pretrained("gpt2")
->>> inputs = tok(["An increasing sequence: one,"], return_tensors="pt")
->>> streamer = TextStreamer(tok)
-
->>> # Despite returning the usual output, the streamer will also print the generated text to stdout.
->>> _ = model.generate(**inputs, streamer=streamer, max_new_tokens=20)
-An increasing sequence: one, two, three, four, five, six, seven, eight, nine, ten, eleven,
-```
-
-## Decoding strategies
-
-特定の `generate()` パラメータの組み合わせ、そして最終的に `generation_config` は、特定のデコーディング戦略を有効にするために使用できます。このコンセプトが新しい場合、[このブログポスト](https://huggingface.co/blog/how-to-generate)を読むことをお勧めします。このブログポストでは、一般的なデコーディング戦略がどのように動作するかが説明されています。
-
-ここでは、デコーディング戦略を制御するいくつかのパラメータを示し、それらをどのように使用できるかを説明します。
-
-### Greedy Search
-
-[`generate`] はデフォルトで貪欲探索デコーディングを使用するため、有効にするためにパラメータを渡す必要はありません。これは、パラメータ `num_beams` が 1 に設定され、`do_sample=False` であることを意味します。
-
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
-
->>> prompt = "I look forward to"
->>> checkpoint = "distilgpt2"
-
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
->>> outputs = model.generate(**inputs)
->>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
-['I look forward to seeing you all again!\n\n\n\n\n\n\n\n\n\n\n']
-```
-
-### Contrastive search
-
-コントラスティブ検索デコーディング戦略は、2022年の論文[A Contrastive Framework for Neural Text Generation](https://arxiv.org/abs/2202.06417)で提案されました。
-これは、非反復的でありながら一貫性のある長い出力を生成するために優れた結果を示しています。コントラスティブ検索の動作原理を学ぶには、[このブログポスト](https://huggingface.co/blog/introducing-csearch)をご覧ください。
-コントラスティブ検索の動作を有効にし、制御する2つの主要なパラメータは「penalty_alpha」と「top_k」です：
-
-```python
->>> from transformers import AutoTokenizer, AutoModelForCausalLM
-
->>> checkpoint = "gpt2-large"
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
-
->>> prompt = "Hugging Face Company is"
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> outputs = model.generate(**inputs, penalty_alpha=0.6, top_k=4, max_new_tokens=100)
->>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
-['Hugging Face Company is a family owned and operated business. We pride ourselves on being the best
-in the business and our customer service is second to none.\n\nIf you have any questions about our
-products or services, feel free to contact us at any time. We look forward to hearing from you!']
-```
-
-### Multinomial sampling
-
-常に最高確率のトークンを次のトークンとして選択する貪欲検索とは異なり、多項分布サンプリング（または祖先サンプリングとも呼ばれます）はモデルによって提供される語彙全体の確率分布に基づいて次のトークンをランダムに選択します。ゼロ以外の確率を持つすべてのトークンには選択される可能性があり、これにより繰り返しのリスクが減少します。
-
-多項分布サンプリングを有効にするには、`do_sample=True` および `num_beams=1` を設定します。
-
-```python
->>> from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
->>> set_seed(0)  # For reproducibility
-
->>> checkpoint = "gpt2-large"
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
-
->>> prompt = "Today was an amazing day because"
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> outputs = model.generate(**inputs, do_sample=True, num_beams=1, max_new_tokens=100)
->>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
-['Today was an amazing day because when you go to the World Cup and you don\'t, or when you don\'t get invited,
-that\'s a terrible feeling."']
-```
-
-### Beam-search decoding
-
-貪欲探索とは異なり、ビームサーチデコーディングは各時間ステップでいくつかの仮説を保持し、最終的にシーケンス全体で最も確率が高い仮説を選択します。これにより、貪欲探索では無視されてしまう初期トークンの確率が低い高確率のシーケンスを特定する利点があります。
-
-このデコーディング戦略を有効にするには、`num_beams`（追跡する仮説の数）を1よりも大きな値に指定します。
-
-希望されるテキストの翻訳がお手伝いできて嬉しいです！もしさらなる質問やサポートが必要な場合は、お気軽にお知らせください。
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
-
->>> prompt = "It is astonishing how one can"
->>> checkpoint = "gpt2-medium"
-
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
-
->>> outputs = model.generate(**inputs, num_beams=5, max_new_tokens=50)
->>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
-['It is astonishing how one can have such a profound impact on the lives of so many people in such a short period of
-time."\n\nHe added: "I am very proud of the work I have been able to do in the last few years.\n\n"I have']
-```
-
-### Beam-search multinomial sampling
-
-その名前からもわかるように、このデコーディング戦略はビームサーチと多項サンプリングを組み合わせています。このデコーディング戦略を使用するには、`num_beams` を1より大きな値に設定し、`do_sample=True` を設定する必要があります。
-
-```python
->>> from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, set_seed
->>> set_seed(0)  # For reproducibility
-
->>> prompt = "translate English to German: The house is wonderful."
->>> checkpoint = "t5-small"
-
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
-
->>> outputs = model.generate(**inputs, num_beams=5, do_sample=True)
->>> tokenizer.decode(outputs[0], skip_special_tokens=True)
-'Das Haus ist wunderbar.'
-```
-
-### Diverse beam search decoding
-
-多様なビームサーチデコーディング戦略は、ビームサーチ戦略の拡張であり、選択肢からより多様なビームシーケンスを生成できるようにします。この仕組みの詳細については、[Diverse Beam Search: Decoding Diverse Solutions from Neural Sequence Models](https://arxiv.org/pdf/1610.02424.pdf) をご参照ください。このアプローチには、`num_beams`、`num_beam_groups`、および `diversity_penalty` という3つの主要なパラメータがあります。多様性ペナルティは、出力がグループごとに異なることを保証し、ビームサーチは各グループ内で使用されます。
-
-
-```python
->>> from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-
->>> checkpoint = "google/pegasus-xsum"
->>> prompt = (
-...     "The Permaculture Design Principles are a set of universal design principles "
-...     "that can be applied to any location, climate and culture, and they allow us to design "
-...     "the most efficient and sustainable human habitation and food production systems. "
-...     "Permaculture is a design system that encompasses a wide variety of disciplines, such "
-...     "as ecology, landscape design, environmental science and energy conservation, and the "
-...     "Permaculture design principles are drawn from these various disciplines. Each individual "
-...     "design principle itself embodies a complete conceptual framework based on sound "
-...     "scientific principles. When we bring all these separate  principles together, we can "
-...     "create a design system that both looks at whole systems, the parts that these systems "
-...     "consist of, and how those parts interact with each other to create a complex, dynamic, "
-...     "living system. Each design principle serves as a tool that allows us to integrate all "
-...     "the separate parts of a design, referred to as elements, into a functional, synergistic, "
-...     "whole system, where the elements harmoniously interact and work together in the most "
-...     "efficient way possible."
-... )
-
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
-
->>> outputs = model.generate(**inputs, num_beams=5, num_beam_groups=5, max_new_tokens=30, diversity_penalty=1.0)
->>> tokenizer.decode(outputs[0], skip_special_tokens=True)
-'The Design Principles are a set of universal design principles that can be applied to any location, climate and
-culture, and they allow us to design the'
-```
-
-### Assisted Decoding
-
-アシストデコーディングは、上記のデコーディング戦略を変更したもので、同じトークナイザー（理想的にははるかに小さなモデル）を使用して、いくつかの候補トークンを貪欲に生成するアシスタントモデルを使用します。その後、主要なモデルは候補トークンを1つの前向きパスで検証し、デコーディングプロセスを高速化します。現在、アシストデコーディングでは貪欲検索とサンプリングのみがサポートされており、バッチ入力はサポートされていません。アシストデコーディングの詳細については、[このブログ記事](https://huggingface.co/blog/assisted-generation) をご覧ください。
-
-アシストデコーディングを有効にするには、`assistant_model` 引数をモデルで設定します。
-
-このガイドは、さまざまなデコーディング戦略を可能にする主要なパラメーターを説明しています。さらに高度なパラメーターは [`generate`] メソッドに存在し、[`generate`] メソッドの動作をさらに制御できます。使用可能なパラメーターの完全なリストについては、[APIドキュメント](./main_classes/text_generation.md) を参照してください。
-
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
-
->>> prompt = "Alice and Bob"
->>> checkpoint = "EleutherAI/pythia-1.4b-deduped"
->>> assistant_checkpoint = "EleutherAI/pythia-160m-deduped"
-
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
->>> assistant_model = AutoModelForCausalLM.from_pretrained(assistant_checkpoint)
->>> outputs = model.generate(**inputs, assistant_model=assistant_model)
->>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
-['Alice and Bob are sitting in a bar. Alice is drinking a beer and Bob is drinking a']
-```
-
-サンプリング方法を使用する場合、アシストデコーディングでは `temperature` 引数を使用して、多項サンプリングと同様にランダム性を制御できます。ただし、アシストデコーディングでは、温度を低くすることで遅延の改善に役立ちます。
-
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
->>> set_seed(42)  # For reproducibility
-
->>> prompt = "Alice and Bob"
->>> checkpoint = "EleutherAI/pythia-1.4b-deduped"
->>> assistant_checkpoint = "EleutherAI/pythia-160m-deduped"
-
->>> tokenizer = AutoTokenizer.from_pretrained(checkpoint)
->>> inputs = tokenizer(prompt, return_tensors="pt")
-
->>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
->>> assistant_model = AutoModelForCausalLM.from_pretrained(assistant_checkpoint)
->>> outputs = model.generate(**inputs, assistant_model=assistant_model, do_sample=True, temperature=0.5)
->>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
-['Alice and Bob are going to the same party. It is a small party, in a small']
-```

docs/source/ja/glossary.md

@@ -1,444 +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.
-
--->
-
-# Glossary
-
-この用語集は、一般的な機械学習と 🤗 トランスフォーマーの用語を定義し、ドキュメンテーションをより理解するのに役立ちます。
-
-## A
-
-### attention mask
-
-アテンション マスクは、シーケンスをバッチ処理する際に使用されるオプションの引数です。
-
-<Youtube id="M6adb1j2jPI"/>
-
-この引数は、モデルにどのトークンを注視すべきか、どのトークンを注視しないかを示します。
-
-例えば、次の2つのシーケンスを考えてみてください：
-
-```python
->>> from transformers import BertTokenizer
-
->>> tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
-
->>> sequence_a = "This is a short sequence."
->>> sequence_b = "This is a rather long sequence. It is at least longer than the sequence A."
-
->>> encoded_sequence_a = tokenizer(sequence_a)["input_ids"]
->>> encoded_sequence_b = tokenizer(sequence_b)["input_ids"]
-```
-
-The encoded versions have different lengths:
-
-```python
->>> len(encoded_sequence_a), len(encoded_sequence_b)
-(8, 19)
-```
-
-したがって、これらのシーケンスをそのまま同じテンソルに配置することはできません。最初のシーケンスは、
-2番目のシーケンスの長さに合わせてパディングする必要があります。または、2番目のシーケンスは、最初のシーケンスの
-長さに切り詰める必要があります。
-
-最初の場合、IDのリストはパディングインデックスで拡張されます。トークナイザにリストを渡し、次のようにパディングするように
-依頼できます:
-
-
-```python
->>> padded_sequences = tokenizer([sequence_a, sequence_b], padding=True)
-```
-
-0sが追加されて、最初の文が2番目の文と同じ長さになるのがわかります：
-
-```python
->>> padded_sequences["input_ids"]
-[[101, 1188, 1110, 170, 1603, 4954, 119, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [101, 1188, 1110, 170, 1897, 1263, 4954, 119, 1135, 1110, 1120, 1655, 2039, 1190, 1103, 4954, 138, 119, 102]]
-```
-
-これは、PyTorchまたはTensorFlowでテンソルに変換できます。注意マスクは、モデルがそれらに注意を払わないように、埋め込まれたインデックスの位置を示すバイナリテンソルです。[`BertTokenizer`]では、`1`は注意を払う必要がある値を示し、`0`は埋め込まれた値を示します。この注意マスクは、トークナイザが返す辞書のキー「attention_mask」の下にあります。
-
-
-
-```python
->>> padded_sequences["attention_mask"]
-[[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]
-```
-
-### autoencoding models
-
-[エンコーダーモデル](#encoder-models) および [マスク言語モデリング](#masked-language-modeling-mlm) を参照してください。
-
-### autoregressive models
-
-[因果言語モデリング](#causal-language-modeling) および [デコーダーモデル](#decoder-models) を参照してください。
-
-## B
-
-### backbone
-
-バックボーンは、生の隠れた状態や特徴を出力するネットワーク（埋め込みと層）です。通常、特徴を入力として受け取るために [ヘッド](#head) に接続されており、予測を行います。たとえば、[`ViTModel`] は特定のヘッドが上にないバックボーンです。他のモデルも [`VitModel`] をバックボーンとして使用できます、例えば [DPT](model_doc/dpt) です。
-
-## C
-
-### causal language modeling
-
-モデルがテキストを順番に読み、次の単語を予測する事前トレーニングタスクです。通常、モデルは文全体を読み取りますが、特定のタイムステップで未来のトークンを隠すためにモデル内でマスクを使用します。
-
-### channel
-
-カラー画像は、赤、緑、青（RGB）の3つのチャネルの値の組み合わせから成り立っており、グレースケール画像は1つのチャネルしか持ちません。🤗 Transformers では、チャネルは画像のテンソルの最初または最後の次元になることがあります：[`n_channels`, `height`, `width`] または [`height`, `width`, `n_channels`]。
-
-### connectionist temporal classification (CTC)
-
-入力と出力が正確にどのように整列するかを正確に知らなくてもモデルを学習させるアルゴリズム。CTC は、特定の入力に対してすべての可能な出力の分布を計算し、その中から最も可能性の高い出力を選択します。CTC は、スピーカーの異なる発話速度など、さまざまな理由で音声がトランスクリプトと完全に整合しない場合に、音声認識タスクで一般的に使用されます。
-
-### convolution
-
-ニューラルネットワークの一種で、入力行列が要素ごとに小さな行列（カーネルまたはフィルター）と乗算され、値が新しい行列に合計されるレイヤーのタイプ。これは入力行列全体に対して繰り返される畳み込み操作として知られ、各操作は入力行列の異なるセグメントに適用されます。畳み込みニューラルネットワーク（CNN）は、コンピュータビジョンで一般的に使用されています。
-
-## D
-
-### decoder input IDs
-
-この入力はエンコーダーデコーダーモデルに特有であり、デコーダーに供給される入力IDを含みます。これらの入力は、翻訳や要約などのシーケンスツーシーケンスタスクに使用され、通常、各モデルに固有の方法で構築されます。
-
-ほとんどのエンコーダーデコーダーモデル（BART、T5）は、`labels` から独自に `decoder_input_ids` を作成します。このようなモデルでは、`labels` を渡すことがトレーニングを処理する優れた方法です。
-
-シーケンスツーシーケンストレーニングにおけるこれらの入力IDの処理方法を確認するために、各モデルのドキュメントを確認してください。
-
-### decoder models
-
-オートリグレッションモデルとも呼ばれ、モデルがテキストを順番に読み、次の単語を予測する事前トレーニングタスク（因果言語モデリング）に関与します。通常、モデルは文全体を読み取り、特定のタイムステップで未来のトークンを隠すマスクを使用して行われます。
-
-<Youtube id="d_ixlCubqQw"/>
-
-
-### deep learning (DL)
-
-ニューラルネットワークを使用する機械学習アルゴリズムで、複数の層を持っています。
-
-## E
-
-### encoder models
-
-オートエンコーディングモデルとしても知られており、エンコーダーモデルは入力（テキストや画像など）を、埋め込みと呼ばれる簡略化された数値表現に変換します。エンコーダーモデルは、しばしば[マスクされた言語モデリング（#masked-language-modeling-mlm）](#masked-language-modeling-mlm)などの技術を使用して事前にトレーニングされ、入力シーケンスの一部をマスクし、モデルにより意味のある表現を作成することが強制されます。
-
-<Youtube id="H39Z_720T5s"/>
-
-## F
-
-### feature extraction
-
-生データをより情報豊かで機械学習アルゴリズムにとって有用な特徴のセットに選択および変換するプロセス。特徴抽出の例には、生のテキストを単語埋め込みに変換したり、画像/ビデオデータからエッジや形状などの重要な特徴を抽出したりすることが含まれます。
-
-### feed forward chunking
-
-トランスフォーマー内の各残差注意ブロックでは、通常、自己注意層の後に2つのフィードフォワード層が続きます。
-フィードフォワード層の中間埋め込みサイズは、モデルの隠れたサイズよりも大きいことがよくあります（たとえば、`bert-base-uncased`の場合）。
-
-入力サイズが `[batch_size、sequence_length]` の場合、中間フィードフォワード埋め込み `[batch_size、sequence_length、config.intermediate_size]` を保存するために必要なメモリは、メモリの大部分を占めることがあります。[Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451)の著者は、計算が `sequence_length` 次元に依存しないため、両方のフィードフォワード層の出力埋め込み `[batch_size、config.hidden_size]_0、...、[batch_size、config.hidden_size]_n` を個別に計算し、後で `[batch_size、sequence_length、config.hidden_size]` に連結することは数学的に等価であると気付きました。これにより、増加した計算時間とメモリ使用量のトレードオフが生じますが、数学的に等価な結果が得られます。
-
-[`apply_chunking_to_forward`] 関数を使用するモデルの場合、`chunk_size` は並列に計算される出力埋め込みの数を定義し、メモリと時間の複雑さのトレードオフを定義します。`chunk_size` が 0 に設定されている場合、フィードフォワードのチャンキングは行われません。
-
-### finetuned models
-
-ファインチューニングは、事前にトレーニングされたモデルを取り、その重みを固定し、新しく追加された[model head](#head)で出力レイヤーを置き換える形式の転移学習です。モデルヘッドは対象のデータセットでトレーニングされます。
-
-詳細については、[Fine-tune a pretrained model](https://huggingface.co/docs/transformers/training) チュートリアルを参照して、🤗 Transformersを使用したモデルのファインチューニング方法を学びましょう。
-
-## H
-
-### head
-
-モデルヘッドは、ニューラルネットワークの最後のレイヤーを指し、生の隠れた状態を受け入れて異なる次元に射影します。各タスクに対して異なるモデルヘッドがあります。例えば：
-
-  * [`GPT2ForSequenceClassification`] は、ベースの[`GPT2Model`]の上にあるシーケンス分類ヘッド（線形層）です。
-  * [`ViTForImageClassification`] は、ベースの[`ViTModel`]の`CLS`トークンの最終隠れた状態の上にある画像分類ヘッド（線形層）です。
-  * [`Wav2Vec2ForCTC`] は、[CTC](#connectionist-temporal-classification-(CTC))を持つベースの[`Wav2Vec2Model`]の言語モデリングヘッドです。
-
-## I
-
-### image patch
-
-ビジョンベースのトランスフォーマーモデルは、画像をより小さなパッチに分割し、それらを線形に埋め込み、モデルにシーケンスとして渡します。モデルの
-
-### inference
-
-推論は、トレーニングが完了した後に新しいデータでモデルを評価するプロセスです。 🤗 Transformers を使用して推論を実行する方法については、[推論のパイプライン](https://huggingface.co/docs/transformers/pipeline_tutorial) チュートリアルを参照してください。
-
-### input IDs
-
-入力IDは、モデルへの入力として渡す必要があるパラメーターの中で最も一般的なものです。これらはトークンのインデックスであり、モデルによって入力として使用されるシーケンスを構築するトークンの数値表現です。
-
-<Youtube id="VFp38yj8h3A"/>
-
-各トークナイザーは異なる方法で動作しますが、基本的なメカニズムは同じです。以下はBERTトークナイザーを使用した例です。BERTトークナイザーは[WordPiece](https://arxiv.org/pdf/1609.08144.pdf)トークナイザーです。
-
-
-```python
->>> from transformers import BertTokenizer
-
->>> tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
-
->>> sequence = "A Titan RTX has 24GB of VRAM"
-```
-
-トークナイザーは、シーケンスをトークナイザー語彙で使用可能なトークンに分割します。
-
-```python
->>> tokenized_sequence = tokenizer.tokenize(sequence)
-```
-
-トークンは単語またはサブワードです。 たとえば、ここでは "VRAM" はモデルの語彙に含まれていなかったため、"V"、"RA"、"M" に分割されました。
-これらのトークンが別々の単語ではなく、同じ単語の一部であることを示すために、"RA" と "M" にはダブルハッシュのプレフィックスが追加されます。
-
-```python
->>> print(tokenized_sequence)
-['A', 'Titan', 'R', '##T', '##X', 'has', '24', '##GB', 'of', 'V', '##RA', '##M']
-```
-
-これらのトークンは、モデルが理解できるようにIDに変換できます。これは、文をトークナイザーに直接供給して行うことができます。トークナイザーは、パフォーマンスの向上のために[🤗 Tokenizers](https://github.com/huggingface/tokenizers)のRust実装を活用しています。
-
-```python
->>> inputs = tokenizer(sequence)
-```
-
-トークナイザーは、対応するモデルが正しく動作するために必要なすべての引数を含む辞書を返します。トークンのインデックスは、キー `input_ids` の下にあります。
-
-```python
->>> encoded_sequence = inputs["input_ids"]
->>> print(encoded_sequence)
-[101, 138, 18696, 155, 1942, 3190, 1144, 1572, 13745, 1104, 159, 9664, 2107, 102]
-```
-
-注意：トークナイザは、関連するモデルがそれらを必要とする場合に自動的に「特別なトークン」を追加します。これらは、モデルが時折使用する特別なIDです。
-
-前のIDシーケンスをデコードする場合、
-
-
-```python
->>> decoded_sequence = tokenizer.decode(encoded_sequence)
-```
-
-私たちは見ます
-
-```python
->>> print(decoded_sequence)
-[CLS] A Titan RTX has 24GB of VRAM [SEP]
-```
-
-これは[`BertModel`]がその入力を期待する方法です。
-
-
-## L
-
-### Labels
-
-ラベルは、モデルが損失を計算するために渡すことができるオプションの引数です。これらのラベルは、モデルの予測の期待値であるべきです。モデルは、通常の損失を使用して、その予測と期待値（ラベル）との間の損失を計算します。
-
-これらのラベルはモデルのヘッドに応じて異なります。たとえば：
-
-- シーケンス分類モデル（[`BertForSequenceClassification`]）の場合、モデルは次元が `(batch_size)` のテンソルを期待し、バッチ内の各値がシーケンス全体の予測ラベルに対応します。
-- トークン分類モデル（[`BertForTokenClassification`]）の場合、モデルは次元が `(batch_size, seq_length)` のテンソルを期待し、各値が各個々のトークンの予測ラベルに対応します。
-- マスク言語モデリングの場合（[`BertForMaskedLM`]）、モデルは次元が `(batch_size, seq_length)` のテンソルを期待し、各値が各個々のトークンの予測ラベルに対応します。ここでのラベルはマスクされたトークンのトークンIDであり、他のトークンには通常 -100 などの値が設定されます。
-- シーケンス間のタスクの場合（[`BartForConditionalGeneration`]、[`MBartForConditionalGeneration`]）、モデルは次元が `(batch_size, tgt_seq_length)` のテンソルを期待し、各値が各入力シーケンスに関連付けられたターゲットシーケンスに対応します。トレーニング中、BARTとT5の両方は適切な `decoder_input_ids` とデコーダーのアテンションマスクを内部で生成します。通常、これらを提供する必要はありません。これはエンコーダーデコーダーフレームワークを利用するモデルには適用されません。
-- 画像分類モデルの場合（[`ViTForImageClassification`]）、モデルは次元が `(batch_size)` のテンソルを期待し、バッチ内の各値が各個々の画像の予測ラベルに対応します。
-- セマンティックセグメンテーションモデルの場合（[`SegformerForSemanticSegmentation`]）、モデルは次元が `(batch_size, height, width)` のテンソルを期待し、バッチ内の各値が各個々のピクセルの予測ラベルに対応します。
-- 物体検出モデルの場合（[`DetrForObjectDetection`]）、モデルは各個々の画像の予測ラベルと境界ボックスの数に対応する `class_labels` と `boxes` キーを持つ辞書のリストを期待します。
-- 自動音声認識モデルの場合（[`Wav2Vec2ForCTC`]）、モデルは次元が `(batch_size, target_length)` のテンソルを期待し、各値が各個々のトークンの予測ラベルに対応します。
-
-<Tip>
-
-各モデルのラベルは異なる場合があるため、常に各モデルのドキュメントを確認して、それらの特定のラベルに関する詳細情報を確認してください！
-
-</Tip>
-
-ベースモデル（[`BertModel`]）はラベルを受け入れません。これらはベースのトランスフォーマーモデルであり、単に特徴を出力します。
-
-
-### large language models (LLM)
-
-大量のデータでトレーニングされた変換器言語モデル（GPT-3、BLOOM、OPT）を指す一般的な用語です。これらのモデルは通常、多くの学習可能なパラメータを持っています（たとえば、GPT-3の場合、1750億個）。
-
-## M
-
-### masked language modeling (MLM)
-
-モデルはテキストの破損バージョンを見る事前トレーニングタスクで、通常はランダムに一部のトークンをマスキングして元のテキストを予測する必要があります。
-
-### multimodal
-
-テキストと別の種類の入力（たとえば画像）を組み合わせるタスクです。
-
-## N
-
-### Natural language generation (NLG)
-
-テキストを生成する関連するすべてのタスク（たとえば、[Transformersで書く](https://transformer.huggingface.co/)、翻訳など）。
-
-### Natural language processing (NLP)
-
-テキストを扱う方法を一般的に表現したものです。
-
-### Natural language understanding (NLU)
-
-テキスト内に何があるかを理解する関連するすべてのタスク（たとえば、テキスト全体の分類、個々の単語の分類など）。
-
-## P
-
-### pipeline
-
-🤗 Transformersのパイプラインは、データの前処理と変換を特定の順序で実行してデータを処理し、モデルから予測を返す一連のステップを指す抽象化です。パイプラインに見られるいくつかのステージの例には、データの前処理、特徴抽出、正規化などがあります。
-
-詳細については、[推論のためのパイプライン](https://huggingface.co/docs/transformers/pipeline_tutorial)を参照してください。
-
-### pixel values
-
-モデルに渡される画像の数値表現のテンソルです。ピクセル値は、形状が [`バッチサイズ`, `チャネル数`, `高さ`, `幅`] の行列で、画像プロセッサから生成されます。
-
-### pooling
-
-行列を小さな行列に縮小する操作で、プール対象の次元の最大値または平均値を取ることが一般的です。プーリングレイヤーは一般的に畳み込みレイヤーの間に見られ、特徴表現をダウンサンプリングします。
-
-### position IDs
-
-トークンごとの位置が埋め込まれているRNNとは異なり、トランスフォーマーは各トークンの位置を把握していません。したがって、モデルはトークンの位置を識別するために位置ID（`position_ids`）を使用します。
-
-これはオプションのパラメータです。モデルに `position_ids` が渡されない場合、IDは自動的に絶対的な位置埋め込みとして作成されます。
-
-絶対的な位置埋め込みは範囲 `[0、config.max_position_embeddings - 1]` から選択されます。一部のモデルは、正弦波位置埋め込みや相対位置埋め込みなど、他のタイプの位置埋め込みを使用することがあります。
-
-
-### preprocessing
-
-生データを機械学習モデルで簡単に処理できる形式に準備するタスクです。例えば、テキストは通常、トークン化によって前処理されます。他の入力タイプに対する前処理の具体的な方法を知りたい場合は、[Preprocess](https://huggingface.co/docs/transformers/preprocessing) チュートリアルをご覧ください。
-
-### pretrained model
-
-あるデータ（たとえば、Wikipedia全体など）で事前に学習されたモデルです。事前学習の方法には、自己教師ありの目的が含まれ、テキストを読み取り、次の単語を予測しようとするもの（[因果言語モデリング](#因果言語モデリング)を参照）や、一部の単語をマスクし、それらを予測しようとするもの（[マスク言語モデリング](#マスク言語モデリング-mlm)を参照）があります。
-
-音声とビジョンモデルには独自の事前学習の目的があります。たとえば、Wav2Vec2は音声モデルで、モデルに対して「真の」音声表現を偽の音声表現のセットから識別する必要がある対比的なタスクで事前学習されています。一方、BEiTはビジョンモデルで、一部の画像パッチをマスクし、モデルにマスクされたパッチを予測させるタスク（マスク言語モデリングの目的と似ています）で事前学習されています。
-
-## R
-
-### recurrent neural network (RNN)
-
-テキストを処理するために層をループさせるモデルの一種です。
-
-### representation learning
-
-生データの意味のある表現を学習する機械学習のサブフィールドです。表現学習の技術の一部には単語埋め込み、オートエンコーダー、Generative Adversarial Networks（GANs）などがあります。
-
-## S
-
-### sampling rate
-
-秒ごとに取られるサンプル（オーディオ信号など）の数をヘルツ単位で測定したものです。サンプリングレートは音声などの連続信号を離散化する結果です。
-
-### self-attention
-
-入力の各要素は、どの他の要素に注意を払うべきかを検出します。
-
-### self-supervised learning 
-
-モデルがラベルのないデータから自分自身の学習目標を作成する機械学習技術のカテゴリです。これは[教師なし学習](#教師なし学習)や[教師あり学習](#教師あり学習)とは異なり、学習プロセスはユーザーからは明示的には監督されていない点が異なります。
-
-自己教師あり学習の1つの例は[マスク言語モデリング](#マスク言語モデリング-mlm)で、モデルには一部のトークンが削除された文が与えられ、欠落したトークンを予測するように学習します。
-
-### semi-supervised learning
-
-ラベル付きデータの少量とラベルのないデータの大量を組み合わせてモデルの精度を向上させる広範な機械学習トレーニング技術のカテゴリです。[教師あり学習](#教師あり学習)や[教師なし学習](#教師なし学習)とは異なり、半教師あり学習のアプローチの1つは「セルフトレーニング」であり、モデルはラベル付きデータでトレーニングされ、次にラベルのないデータで予測を行います。モデルが最も自信を持って予測する部分がラベル付きデータセットに追加され、モデルの再トレーニングに使用されます。
-
-### sequence-to-sequence (seq2seq)
-
-入力から新しいシーケンスを生成するモデルです。翻訳モデルや要約モデル（[Bart](model_doc/bart)や[T5](model_doc/t5)など）などがこれに該当します。
-
-### stride
-
-[畳み込み](#畳み込み)または[プーリング](#プーリング)において、ストライドはカーネルが行列上で移動する距離を指します。ストライドが1の場合、カーネルは1ピクセルずつ移動し、ストライドが2の場合、カーネルは2ピクセルずつ移動します。
-
-### supervised learning
-
-モデルのトレーニング方法の一つで、直接ラベル付きデータを使用してモデルの性能を修正し指導します。データがトレーニングされているモデルに供給され、その予測が既知のラベルと比較されます。モデルは予測がどれだけ誤っていたかに基づいて重みを更新し、プロセスはモデルの性能を最適化するために繰り返されます。
-
-## T
-
-### token
-
-文の一部であり、通常は単語ですが、サブワード（一般的でない単語はしばしばサブワードに分割されることがあります）または句読点の記号であることもあります。
-
-### token Type IDs
-
-一部のモデルは、文のペアの分類や質問応答を行うことを目的としています。
-
-<Youtube id="0u3ioSwev3s"/>
-
-これには異なる2つのシーケンスを単一の「input_ids」エントリに結合する必要があり、通常は分類子（`[CLS]`）や区切り記号（`[SEP]`）などの特別なトークンの助けを借りて実行されます。例えば、BERTモデルは次のように2つのシーケンス入力を構築します：
-
-日本語訳を提供していただきたいです。Markdown形式で記述してください。
-
-
-```python
->>> # [CLS] SEQUENCE_A [SEP] SEQUENCE_B [SEP]
-```
-
-我々は、前述のように、2つのシーケンスを2つの引数として `tokenizer` に渡すことで、このような文を自動的に生成することができます（以前のようにリストではなく）。以下のように：
-
-```python
->>> from transformers import BertTokenizer
-
->>> tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
->>> sequence_a = "HuggingFace is based in NYC"
->>> sequence_b = "Where is HuggingFace based?"
-
->>> encoded_dict = tokenizer(sequence_a, sequence_b)
->>> decoded = tokenizer.decode(encoded_dict["input_ids"])
-```
-
-これに対応するコードは以下です：
-
-```python
->>> print(decoded)
-[CLS] HuggingFace is based in NYC [SEP] Where is HuggingFace based? [SEP]
-```
-
-一部のモデルでは、1つのシーケンスがどこで終わり、別のシーケンスがどこで始まるかを理解するのに十分な情報が備わっています。ただし、BERTなどの他のモデルでは、トークンタイプID（セグメントIDとも呼ばれる）も使用されています。これは、モデル内の2つのシーケンスを識別するバイナリマスクとして表されます。
-
-トークナイザは、このマスクを「token_type_ids」として返します。
-
-```python
->>> encoded_dict["token_type_ids"]
-[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1]
-```
-
-最初のシーケンス、つまり質問のために使用される「コンテキスト」は、すべてのトークンが「0」で表されています。一方、2番目のシーケンス、質問に対応するものは、すべてのトークンが「1」で表されています。
-
-一部のモデル、例えば [`XLNetModel`] のように、追加のトークンが「2」で表されます。
-
-
-### transfer learning
-
-事前に学習されたモデルを取り、それをタスク固有のデータセットに適応させる技術。ゼロからモデルを訓練する代わりに、既存のモデルから得た知識を出発点として活用できます。これにより学習プロセスが加速し、必要な訓練データの量が減少します。
-
-### transformer
-
-自己注意ベースの深層学習モデルアーキテクチャ。
-
-## U
-
-### unsupervised learning
-
-モデルに提供されるデータがラベル付けされていないモデルトレーニングの形態。教師なし学習の技術は、タスクに役立つパターンを見つけるためにデータ分布の統計情報を活用します。

docs/source/ja/hpo_train.md

@@ -1,150 +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
-
-⚠️ 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.
-
--->
-
-# Hyperparameter Search using Trainer API
-
-🤗 Transformersは、🤗 Transformersモデルのトレーニングを最適化する[`Trainer`]クラスを提供し、独自のトレーニングループを手動で記述せずにトレーニングを開始するのが簡単になります。[`Trainer`]はハイパーパラメーター検索のAPIも提供しています。このドキュメントでは、それを例示します。
-
-## Hyperparameter Search backend
-
-[`Trainer`]は現在、4つのハイパーパラメーター検索バックエンドをサポートしています：
-[optuna](https://optuna.org/)、[sigopt](https://sigopt.com/)、[raytune](https://docs.ray.io/en/latest/tune/index.html)、および[wandb](https://wandb.ai/site/sweeps)。
-
-これらを使用する前に、ハイパーパラメーター検索バックエンドをインストールする必要があります。
-```bash
-pip install optuna/sigopt/wandb/ray[tune] 
-```
-
-## How to enable Hyperparameter search in example
-
-ハイパーパラメータの検索スペースを定義し、異なるバックエンドには異なるフォーマットが必要です。
-
-Sigoptの場合、sigopt [object_parameter](https://docs.sigopt.com/ai-module-api-references/api_reference/objects/object_parameter) を参照してください。それは以下のようなものです：
-```py
->>> def sigopt_hp_space(trial):
-...     return [
-...         {"bounds": {"min": 1e-6, "max": 1e-4}, "name": "learning_rate", "type": "double"},
-...         {
-...             "categorical_values": ["16", "32", "64", "128"],
-...             "name": "per_device_train_batch_size",
-...             "type": "categorical",
-...         },
-...     ]
-```
-
-
-Optunaに関しては、[object_parameter](https://optuna.readthedocs.io/en/stable/tutorial/10_key_features/002_configurations.html#sphx-glr-tutorial-10-key-features-002-configurations-py)をご覧ください。以下のようになります：
-
-
-```py
->>> def optuna_hp_space(trial):
-...     return {
-...         "learning_rate": trial.suggest_float("learning_rate", 1e-6, 1e-4, log=True),
-...         "per_device_train_batch_size": trial.suggest_categorical("per_device_train_batch_size", [16, 32, 64, 128]),
-...     }
-```
-
-Optunaは、多目的のハイパーパラメータ最適化（HPO）を提供しています。 `hyperparameter_search` で `direction` を渡し、複数の目的関数値を返すための独自の `compute_objective` を定義することができます。 Pareto Front（`List[BestRun]`）は `hyperparameter_search` で返され、[test_trainer](https://github.com/huggingface/transformers/blob/main/tests/trainer/test_trainer.py) のテストケース `TrainerHyperParameterMultiObjectOptunaIntegrationTest` を参照する必要があります。これは以下のようになります。
-
-
-```py
->>> best_trials = trainer.hyperparameter_search(
-...     direction=["minimize", "maximize"],
-...     backend="optuna",
-...     hp_space=optuna_hp_space,
-...     n_trials=20,
-...     compute_objective=compute_objective,
-... )
-```
-
-Ray Tuneに関して、[object_parameter](https://docs.ray.io/en/latest/tune/api/search_space.html)を参照してください。以下のようになります：
-
-
-```py
->>> def ray_hp_space(trial):
-...     return {
-...         "learning_rate": tune.loguniform(1e-6, 1e-4),
-...         "per_device_train_batch_size": tune.choice([16, 32, 64, 128]),
-...     }
-```
-
-Wandbについては、[object_parameter](https://docs.wandb.ai/guides/sweeps/configuration)をご覧ください。これは以下のようになります：
-
-```py
->>> def wandb_hp_space(trial):
-...     return {
-...         "method": "random",
-...         "metric": {"name": "objective", "goal": "minimize"},
-...         "parameters": {
-...             "learning_rate": {"distribution": "uniform", "min": 1e-6, "max": 1e-4},
-...             "per_device_train_batch_size": {"values": [16, 32, 64, 128]},
-...         },
-...     }
-```
-
-`model_init` 関数を定義し、それを [`Trainer`] に渡す例を示します：
-
-
-```py
->>> def model_init(trial):
-...     return AutoModelForSequenceClassification.from_pretrained(
-...         model_args.model_name_or_path,
-...         from_tf=bool(".ckpt" in model_args.model_name_or_path),
-...         config=config,
-...         cache_dir=model_args.cache_dir,
-...         revision=model_args.model_revision,
-...         use_auth_token=True if model_args.use_auth_token else None,
-...     )
-```
-
-[`Trainer`] を `model_init` 関数、トレーニング引数、トレーニングデータセット、テストデータセット、および評価関数と共に作成してください:
-
-
-```py
->>> trainer = Trainer(
-...     model=None,
-...     args=training_args,
-...     train_dataset=small_train_dataset,
-...     eval_dataset=small_eval_dataset,
-...     compute_metrics=compute_metrics,
-...     tokenizer=tokenizer,
-...     model_init=model_init,
-...     data_collator=data_collator,
-... )
-```
-
-ハイパーパラメーターの探索を呼び出し、最良のトライアル パラメーターを取得します。バックエンドは `"optuna"` / `"sigopt"` / `"wandb"` / `"ray"` である可能性があります。方向は `"minimize"` または `"maximize"` であり、目標をより大きくするか小さくするかを示します。
-
-`compute_objective` 関数を独自に定義することもできます。定義されていない場合、デフォルトの `compute_objective` が呼び出され、F1などの評価メトリックの合計が目標値として返されます。
-
-
-```py
->>> best_trial = trainer.hyperparameter_search(
-...     direction="maximize",
-...     backend="optuna",
-...     hp_space=optuna_hp_space,
-...     n_trials=20,
-...     compute_objective=compute_objective,
-... )
-```
-
-## Hyperparameter search For DDP finetune
-現在、DDP（Distributed Data Parallel）のためのハイパーパラメーター検索は、Optuna と SigOpt に対して有効になっています。ランクゼロプロセスのみが検索トライアルを生成し、他のランクに引数を渡します。
-
-
-
-
-
-

docs/source/ja/internal/audio_utils.md

@@ -1,39 +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.
-
--->
-
-# `FeatureExtractor` 用のユーティリティ
-
-このページには、*短時間フーリエ変換* や *ログ メル スペクトログラム* などの一般的なアルゴリズムを使用して生のオーディオから特別な特徴を計算するために、オーディオ [`FeatureExtractor`] で使用できるすべてのユーティリティ関数がリストされています。
-
-これらのほとんどは、ライブラリ内のオーディオ プロセッサのコードを学習する場合にのみ役に立ちます。
-
-## オーディオ変換
-
-[[autodoc]] audio_utils.hertz_to_mel
-
-[[autodoc]] audio_utils.mel_to_hertz
-
-[[autodoc]] audio_utils.mel_filter_bank
-
-[[autodoc]] audio_utils.optimal_fft_length
-
-[[autodoc]] audio_utils.window_function
-
-[[autodoc]] audio_utils.spectrogram
-
-[[autodoc]] audio_utils.power_to_db
-
-[[autodoc]] audio_utils.amplitude_to_db

docs/source/ja/internal/file_utils.md

@@ -1,49 +0,0 @@
-<!--Copyright 2021 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ 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.
-
--->
-
-# 一般的なユーティリティ
-
-このページには、ファイル `utils.py` にある Transformers の一般的なユーティリティ関数がすべてリストされています。
-
-これらのほとんどは、ライブラリで一般的なコードを学習する場合にのみ役に立ちます。
-
-## 列挙型と名前付きタプル
-
-[[autodoc]] utils.ExplicitEnum
-
-[[autodoc]] utils.PaddingStrategy
-
-[[autodoc]] utils.TensorType
-
-## 特別なデコレーター
-
-[[autodoc]] utils.add_start_docstrings
-
-[[autodoc]] utils.add_start_docstrings_to_model_forward
-
-[[autodoc]] utils.add_end_docstrings
-
-[[autodoc]] utils.add_code_sample_docstrings
-
-[[autodoc]] utils.replace_return_docstrings
-
-## 特殊なプロパティ
-
-[[autodoc]] utils.cached_property
-
-## その他のユーティリティ
-
-[[autodoc]] utils._LazyModule

docs/source/ja/internal/generation_utils.md

@@ -1,369 +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.
-
-⚠️ 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.
-
--->
-
-# 発電用ユーティリティ
-
-このページには、[`~generation.GenerationMixin.generate`] で使用されるすべてのユーティリティ関数がリストされています。
-[`~generation.GenerationMixin.greedy_search`],
-[`~generation.GenerationMixin.contrastive_search`],
-[`~generation.GenerationMixin.sample`],
-[`~generation.GenerationMixin.beam_search`],
-[`~generation.GenerationMixin.beam_sample`],
-[`~generation.GenerationMixin.group_beam_search`]、および
-[`~generation.GenerationMixin.constrained_beam_search`]。
-
-これらのほとんどは、ライブラリ内の生成メソッドのコードを学習する場合にのみ役に立ちます。
-
-## 出力を生成する
-
-[`~generation.GenerationMixin.generate`] の出力は、次のサブクラスのインスタンスです。
-[`~utils.ModelOutput`]。この出力は、返されたすべての情報を含むデータ構造です。
-[`~generation.GenerationMixin.generate`] によって作成されますが、タプルまたは辞書としても使用できます。
-
-以下に例を示します。
-
-```python
-from transformers import GPT2Tokenizer, GPT2LMHeadModel
-
-tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
-model = GPT2LMHeadModel.from_pretrained("gpt2")
-
-inputs = tokenizer("Hello, my dog is cute and ", return_tensors="pt")
-generation_output = model.generate(**inputs, return_dict_in_generate=True, output_scores=True)
-```
-
-`generation_output` オブジェクトは、できる限り [`~generation.GreedySearchDecoderOnlyOutput`] です。
-以下のそのクラスのドキュメントを参照してください。これは、次の属性があることを意味します。
-
-- `sequences`: 生成されたトークンのシーケンス
-- `scores` (オプション): 各生成ステップの言語モデリング ヘッドの予測スコア
-- `hidden_​​states` (オプション): 生成ステップごとのモデルの隠れた状態
-- `attentions` (オプション): 生成ステップごとのモデルのアテンションの重み
-
-ここでは、`output_scores=True`を渡したので `scores` がありますが、`hidden_​​states` はありません。
-`attentions` は、`output_hidden_​​states=True`または`output_attentions=True`を渡さなかったためです。
-
-通常と同じように各属性にアクセスできます。その属性がモデルから返されなかった場合は、
-は「なし」を取得します。ここで、たとえば`generation_output.scores`は、生成されたすべての予測スコアです。
-言語モデリングのヘッドであり、`generation_output.attentions`は`None`です。
-
-`generation_output` オブジェクトをタプルとして使用する場合、`None` 値を持たない属性のみが保持されます。
-たとえば、ここには 2 つの要素、`loss`、次に`logits`があります。
-
-```python
-generation_output[:2]
-```
-
-たとえば、タプル `(generation_output.sequences,generation_output.scores)` を返します。
-
-`generation_output` オブジェクトを辞書として使用する場合、`None` を持たない属性のみが保持されます。
-ここでは、たとえば、`sequences`と`scores`という 2 つのキーがあります。
-
-ここではすべての出力タイプを文書化します。
-
-### PyTorch
-
-[[autodoc]] generation.GreedySearchEncoderDecoderOutput
-
-[[autodoc]] generation.GreedySearchDecoderOnlyOutput
-
-[[autodoc]] generation.SampleEncoderDecoderOutput
-
-[[autodoc]] generation.SampleDecoderOnlyOutput
-
-[[autodoc]] generation.BeamSearchEncoderDecoderOutput
-
-[[autodoc]] generation.BeamSearchDecoderOnlyOutput
-
-[[autodoc]] generation.BeamSampleEncoderDecoderOutput
-
-[[autodoc]] generation.BeamSampleDecoderOnlyOutput
-
-[[autodoc]] generation.ContrastiveSearchEncoderDecoderOutput
-
-[[autodoc]] generation.ContrastiveSearchDecoderOnlyOutput
-
-### TensorFlow
-
-[[autodoc]] generation.TFGreedySearchEncoderDecoderOutput
-
-[[autodoc]] generation.TFGreedySearchDecoderOnlyOutput
-
-[[autodoc]] generation.TFSampleEncoderDecoderOutput
-
-[[autodoc]] generation.TFSampleDecoderOnlyOutput
-
-[[autodoc]] generation.TFBeamSearchEncoderDecoderOutput
-
-[[autodoc]] generation.TFBeamSearchDecoderOnlyOutput
-
-[[autodoc]] generation.TFBeamSampleEncoderDecoderOutput
-
-[[autodoc]] generation.TFBeamSampleDecoderOnlyOutput
-
-[[autodoc]] generation.TFContrastiveSearchEncoderDecoderOutput
-
-[[autodoc]] generation.TFContrastiveSearchDecoderOnlyOutput
-
-### FLAX
-
-[[autodoc]] generation.FlaxSampleOutput
-
-[[autodoc]] generation.FlaxGreedySearchOutput
-
-[[autodoc]] generation.FlaxBeamSearchOutput
-
-## LogitsProcessor
-
-[`LogitsProcessor`] を使用して、言語モデルのヘッドの予測スコアを変更できます。
-世代。
-
-### PyTorch
-
-[[autodoc]] AlternatingCodebooksLogitsProcessor
-    - __call__
-
-[[autodoc]] ClassifierFreeGuidanceLogitsProcessor
-    - __call__
-
-[[autodoc]] EncoderNoRepeatNGramLogitsProcessor
-    - __call__
-
-[[autodoc]] EncoderRepetitionPenaltyLogitsProcessor
-    - __call__
-
-[[autodoc]] EpsilonLogitsWarper
-    - __call__
-
-[[autodoc]] EtaLogitsWarper
-    - __call__
-
-[[autodoc]] ExponentialDecayLengthPenalty
-    - __call__
-
-[[autodoc]] ForcedBOSTokenLogitsProcessor
-    - __call__
-
-[[autodoc]] ForcedEOSTokenLogitsProcessor
-    - __call__
-
-[[autodoc]] ForceTokensLogitsProcessor
-    - __call__
-
-[[autodoc]] HammingDiversityLogitsProcessor
-    - __call__
-
-[[autodoc]] InfNanRemoveLogitsProcessor
-    - __call__
-
-[[autodoc]] LogitNormalization
-    - __call__
-
-[[autodoc]] LogitsProcessor
-    - __call__
-
-[[autodoc]] LogitsProcessorList
-    - __call__
-
-[[autodoc]] LogitsWarper
-    - __call__
-
-[[autodoc]] MinLengthLogitsProcessor
-    - __call__
-
-[[autodoc]] MinNewTokensLengthLogitsProcessor
-    - __call__
-
-[[autodoc]] NoBadWordsLogitsProcessor
-    - __call__
-
-[[autodoc]] NoRepeatNGramLogitsProcessor
-    - __call__
-
-[[autodoc]] PrefixConstrainedLogitsProcessor
-    - __call__
-
-[[autodoc]] RepetitionPenaltyLogitsProcessor
-    - __call__
-
-[[autodoc]] SequenceBiasLogitsProcessor
-    - __call__
-
-[[autodoc]] SuppressTokensAtBeginLogitsProcessor
-    - __call__
-
-[[autodoc]] SuppressTokensLogitsProcessor
-    - __call__
-
-[[autodoc]] TemperatureLogitsWarper
-    - __call__
-
-[[autodoc]] TopKLogitsWarper
-    - __call__
-
-[[autodoc]] TopPLogitsWarper
-    - __call__
-
-[[autodoc]] TypicalLogitsWarper
-    - __call__
-
-[[autodoc]] UnbatchedClassifierFreeGuidanceLogitsProcessor
-    - __call__
-
-[[autodoc]] WhisperTimeStampLogitsProcessor
-    - __call__
-
-### TensorFlow
-
-[[autodoc]] TFForcedBOSTokenLogitsProcessor
-    - __call__
-
-[[autodoc]] TFForcedEOSTokenLogitsProcessor
-    - __call__
-
-[[autodoc]] TFForceTokensLogitsProcessor
-    - __call__
-
-[[autodoc]] TFLogitsProcessor
-    - __call__
-
-[[autodoc]] TFLogitsProcessorList
-    - __call__
-
-[[autodoc]] TFLogitsWarper
-    - __call__
-
-[[autodoc]] TFMinLengthLogitsProcessor
-    - __call__
-
-[[autodoc]] TFNoBadWordsLogitsProcessor
-    - __call__
-
-[[autodoc]] TFNoRepeatNGramLogitsProcessor
-    - __call__
-
-[[autodoc]] TFRepetitionPenaltyLogitsProcessor
-    - __call__
-
-[[autodoc]] TFSuppressTokensAtBeginLogitsProcessor
-    - __call__
-
-[[autodoc]] TFSuppressTokensLogitsProcessor
-    - __call__
-
-[[autodoc]] TFTemperatureLogitsWarper
-    - __call__
-
-[[autodoc]] TFTopKLogitsWarper
-    - __call__
-
-[[autodoc]] TFTopPLogitsWarper
-    - __call__
-
-### FLAX
-
-[[autodoc]] FlaxForcedBOSTokenLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxForcedEOSTokenLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxForceTokensLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxLogitsProcessorList
-    - __call__
-
-[[autodoc]] FlaxLogitsWarper
-    - __call__
-
-[[autodoc]] FlaxMinLengthLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxSuppressTokensAtBeginLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxSuppressTokensLogitsProcessor
-    - __call__
-
-[[autodoc]] FlaxTemperatureLogitsWarper
-    - __call__
-
-[[autodoc]] FlaxTopKLogitsWarper
-    - __call__
-
-[[autodoc]] FlaxTopPLogitsWarper
-    - __call__
-
-[[autodoc]] FlaxWhisperTimeStampLogitsProcessor
-    - __call__
-
-## StoppingCriteria
-
-[`StoppingCriteria`] を使用して、(EOS トークン以外の) 生成を停止するタイミングを変更できます。これは PyTorch 実装でのみ利用可能であることに注意してください。
-
-[[autodoc]] StoppingCriteria
-    - __call__
-
-[[autodoc]] StoppingCriteriaList
-    - __call__
-
-[[autodoc]] MaxLengthCriteria
-    - __call__
-
-[[autodoc]] MaxTimeCriteria
-    - __call__
-
-## Constraints
-
-[`Constraint`] を使用すると、生成時に出力に特定のトークンまたはシーケンスが含まれるように強制できます。これは PyTorch 実装でのみ利用可能であることに注意してください。
-
-[[autodoc]] Constraint
-
-[[autodoc]] PhrasalConstraint
-
-[[autodoc]] DisjunctiveConstraint
-
-[[autodoc]] ConstraintListState
-
-## BeamSearch
-
-[[autodoc]] BeamScorer
-    - process
-    - finalize
-
-[[autodoc]] BeamSearchScorer
-    - process
-    - finalize
-
-[[autodoc]] ConstrainedBeamSearchScorer
-    - process
-    - finalize
-
-## Utilities
-
-[[autodoc]] top_k_top_p_filtering
-
-[[autodoc]] tf_top_k_top_p_filtering
-
-## Streamers
-
-[[autodoc]] TextStreamer
-
-[[autodoc]] TextIteratorStreamer

docs/source/ja/internal/image_processing_utils.md

@@ -1,48 +0,0 @@
-!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# 画像プロセッサ用ユーティリティ
-
-このページには、画像プロセッサーで使用されるすべてのユーティリティー関数がリストされています。主に機能的なものです。
-画像を処理するために使用される変換。
-
-これらのほとんどは、ライブラリ内の画像プロセッサのコードを学習する場合にのみ役に立ちます。
-
-## Image Transformations
-
-[[autodoc]] image_transforms.center_crop
-
-[[autodoc]] image_transforms.center_to_corners_format
-
-[[autodoc]] image_transforms.corners_to_center_format
-
-[[autodoc]] image_transforms.id_to_rgb
-
-[[autodoc]] image_transforms.normalize
-
-[[autodoc]] image_transforms.pad
-
-[[autodoc]] image_transforms.rgb_to_id
-
-[[autodoc]] image_transforms.rescale
-
-[[autodoc]] image_transforms.resize
-
-[[autodoc]] image_transforms.to_pil_image
-
-## ImageProcessingMixin
-
-[[autodoc]] image_processing_utils.ImageProcessingMixin

docs/source/ja/internal/modeling_utils.md

@@ -1,83 +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.
-
-⚠️ 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.
-
--->
-
-# カスタムレイヤーとユーティリティ
-
-このページには、ライブラリで使用されるすべてのカスタム レイヤーと、モデリングに提供されるユーティリティ関数がリストされます。
-
-これらのほとんどは、ライブラリ内のモデルのコードを研究する場合にのみ役に立ちます。
-
-
-## Pytorch custom modules
-
-[[autodoc]] pytorch_utils.Conv1D
-
-[[autodoc]] modeling_utils.PoolerStartLogits
-    - forward
-
-[[autodoc]] modeling_utils.PoolerEndLogits
-    - forward
-
-[[autodoc]] modeling_utils.PoolerAnswerClass
-    - forward
-
-[[autodoc]] modeling_utils.SquadHeadOutput
-
-[[autodoc]] modeling_utils.SQuADHead
-    - forward
-
-[[autodoc]] modeling_utils.SequenceSummary
-    - forward
-
-## PyTorch Helper Functions
-
-[[autodoc]] pytorch_utils.apply_chunking_to_forward
-
-[[autodoc]] pytorch_utils.find_pruneable_heads_and_indices
-
-[[autodoc]] pytorch_utils.prune_layer
-
-[[autodoc]] pytorch_utils.prune_conv1d_layer
-
-[[autodoc]] pytorch_utils.prune_linear_layer
-
-## TensorFlow custom layers
-
-[[autodoc]] modeling_tf_utils.TFConv1D
-
-[[autodoc]] modeling_tf_utils.TFSequenceSummary
-
-## TensorFlow loss functions
-
-[[autodoc]] modeling_tf_utils.TFCausalLanguageModelingLoss
-
-[[autodoc]] modeling_tf_utils.TFMaskedLanguageModelingLoss
-
-[[autodoc]] modeling_tf_utils.TFMultipleChoiceLoss
-
-[[autodoc]] modeling_tf_utils.TFQuestionAnsweringLoss
-
-[[autodoc]] modeling_tf_utils.TFSequenceClassificationLoss
-
-[[autodoc]] modeling_tf_utils.TFTokenClassificationLoss
-
-## TensorFlow Helper Functions
-
-[[autodoc]] modeling_tf_utils.get_initializer
-
-[[autodoc]] modeling_tf_utils.keras_serializable
-
-[[autodoc]] modeling_tf_utils.shape_list

docs/source/ja/internal/pipelines_utils.md

@@ -1,44 +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.
-
-⚠️ 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.
-
--->
-
-# パイプライン用のユーティリティ
-
-このページには、ライブラリがパイプラインに提供するすべてのユーティリティ関数がリストされます。
-
-これらのほとんどは、ライブラリ内のモデルのコードを研究する場合にのみ役に立ちます。
-
-
-## Argument handling
-
-[[autodoc]] pipelines.ArgumentHandler
-
-[[autodoc]] pipelines.ZeroShotClassificationArgumentHandler
-
-[[autodoc]] pipelines.QuestionAnsweringArgumentHandler
-
-## Data format
-
-[[autodoc]] pipelines.PipelineDataFormat
-
-[[autodoc]] pipelines.CsvPipelineDataFormat
-
-[[autodoc]] pipelines.JsonPipelineDataFormat
-
-[[autodoc]] pipelines.PipedPipelineDataFormat
-
-## Utilities
-
-[[autodoc]] pipelines.PipelineException

docs/source/ja/internal/time_series_utils.md

@@ -1,29 +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.
-
--->
-
-# 時系列ユーティリティ
-
-このページには、時系列ベースのモデルに使用できるすべてのユーティリティ関数とクラスがリストされます。
-
-これらのほとんどは、時系列モデルのコードを研究している場合、または分散出力クラスのコレクションに追加したい場合にのみ役立ちます。
-
-## Distributional Output
-
-[[autodoc]] time_series_utils.NormalOutput
-
-[[autodoc]] time_series_utils.StudentTOutput
-
-[[autodoc]] time_series_utils.NegativeBinomialOutput

docs/source/ja/internal/tokenization_utils.md

@@ -1,42 +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.
-
-⚠️ 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.
-
--->
-
-# Utilities for Tokenizers
-
-このページには、トークナイザーによって使用されるすべてのユーティリティ関数 (主にクラス) がリストされます。
-[`~tokenization_utils_base.PreTrainedTokenizerBase`] 間の共通メソッドを実装します。
-[`PreTrainedTokenizer`] と [`PreTrainedTokenizerFast`] およびミックスイン
-[`~tokenization_utils_base.SpecialTokensMixin`]。
-
-これらのほとんどは、ライブラリ内のトークナイザーのコードを学習する場合にのみ役に立ちます。
-
-## PreTrainedTokenizerBase
-
-[[autodoc]] tokenization_utils_base.PreTrainedTokenizerBase
-    - __call__
-    - all
-
-## SpecialTokensMixin
-
-[[autodoc]] tokenization_utils_base.SpecialTokensMixin
-
-## Enums and namedtuples
-
-[[autodoc]] tokenization_utils_base.TruncationStrategy
-
-[[autodoc]] tokenization_utils_base.CharSpan
-
-[[autodoc]] tokenization_utils_base.TokenSpan

docs/source/ja/internal/trainer_utils.md

@@ -1,49 +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.
-
-⚠️ 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.
-
--->
-
-# トレーナー用ユーティリティ
-
-このページには、[`Trainer`] で使用されるすべてのユーティリティ関数がリストされています。
-
-これらのほとんどは、ライブラリ内のトレーナーのコードを学習する場合にのみ役に立ちます。
-
-## Utilities
-
-[[autodoc]] EvalPrediction
-
-[[autodoc]] IntervalStrategy
-
-[[autodoc]] enable_full_determinism
-
-[[autodoc]] set_seed
-
-[[autodoc]] torch_distributed_zero_first
-
-## Callbacks internals
-
-[[autodoc]] trainer_callback.CallbackHandler
-
-## Distributed Evaluation
-
-[[autodoc]] trainer_pt_utils.DistributedTensorGatherer
-
-## Distributed Evaluation
-
-[[autodoc]] HfArgumentParser
-
-## Debug Utilities
-
-[[autodoc]] debug_utils.DebugUnderflowOverflow

docs/source/ja/llm_tutorial.md

@@ -1,223 +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.
-
--->
-
-# Generation with LLMs
-
-[[open-in-colab]]
-
-LLM、またはLarge Language Models（大規模言語モデル）は、テキスト生成の鍵となる要素です。要するに、これらは大規模な事前訓練済みトランスフォーマーモデルで、与えられた入力テキストに基づいて次の単語（または、より正確にはトークン）を予測するように訓練されています。トークンを1つずつ予測するため、モデルを呼び出すだけでは新しい文を生成するために何かより精巧なことをする必要があります。自己回帰生成を行う必要があります。
-
-自己回帰生成は、推論時の手続きで、いくつかの初期入力を与えた状態で、モデルを反復的に呼び出す手法です。🤗 Transformersでは、これは[`~generation.GenerationMixin.generate`]メソッドによって処理され、これは生成能力を持つすべてのモデルで利用可能です。
-
-このチュートリアルでは、以下のことを示します：
-
-* LLMを使用してテキストを生成する方法
-* 一般的な落とし穴を回避する方法
-* LLMを最大限に活用するための次のステップ
-
-始める前に、必要なライブラリがすべてインストールされていることを確認してください：
-
-
-```bash
-pip install transformers bitsandbytes>=0.39.0 -q
-```
-
-## Generate text
-
-[因果言語モデリング](tasks/language_modeling)のためにトレーニングされた言語モデルは、テキストトークンのシーケンスを入力として受け取り、次のトークンの確率分布を返します。
-
-<!-- [GIF 1 -- FWD PASS] -->
-<figure class="image table text-center m-0 w-full">
-    <video
-        style="max-width: 90%; margin: auto;"
-        autoplay loop muted playsinline
-        src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/assisted-generation/gif_1_1080p.mov"
-    ></video>
-    <figcaption>"Forward pass of an LLM"</figcaption>
-</figure>
-
-
-LLM（Language Model）による自己回帰生成の重要な側面の1つは、この確率分布から次のトークンを選択する方法です。このステップでは、次のイテレーションのためのトークンが得られる限り、何でも可能です。これは、確率分布から最も可能性の高いトークンを選択するだけのシンプルな方法から、結果の分布からサンプリングする前に数々の変換を適用するほど複雑な方法まで、あらゆる方法が考えられます。
-
-
-<!-- [GIF 2 -- TEXT GENERATION] -->
-<figure class="image table text-center m-0 w-full">
-    <video
-        style="max-width: 90%; margin: auto;"
-        autoplay loop muted playsinline
-        src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/assisted-generation/gif_2_1080p.mov"
-    ></video>
-    <figcaption>"Autoregressive generation iteratively selects the next token from a probability distribution to generate text"</figcaption>
-</figure>
-
-上記のプロセスは、ある停止条件が満たされるまで反復的に繰り返されます。理想的には、停止条件はモデルによって指示され、モデルは終了シーケンス（`EOS`）トークンを出力するタイミングを学習すべきです。これがそうでない場合、生成はあらかじめ定義された最大長に達したときに停止します。
-
-トークン選択ステップと停止条件を適切に設定することは、モデルがタスクで期待どおりに振る舞うために重要です。それが、各モデルに関連付けられた [`~generation.GenerationConfig`] ファイルがある理由であり、これには優れたデフォルトの生成パラメータ化が含まれ、モデルと一緒に読み込まれます。
-
-コードについて話しましょう！
-
-<Tip>
-
-基本的なLLMの使用に興味がある場合、高レベルの [`Pipeline`](pipeline_tutorial) インターフェースが良い出発点です。ただし、LLMはしばしば量子化やトークン選択ステップの細かい制御などの高度な機能が必要であり、これは [`~generation.GenerationMixin.generate`] を介して最良に行われます。LLMとの自己回帰生成はリソースが多く必要であり、適切なスループットのためにGPUで実行する必要があります。
-
-</Tip>
-
-<!-- TODO: llama 2（またはより新しい一般的なベースライン）が利用可能になったら、例を更新する -->
-まず、モデルを読み込む必要があります。
-
-
-```py
->>> from transformers import AutoModelForCausalLM
-
->>> model = AutoModelForCausalLM.from_pretrained(
-...     "openlm-research/open_llama_7b", device_map="auto", load_in_4bit=True
-... )
-```
-
-`from_pretrained` 呼び出しで2つのフラグがあることに注意してください：
-
-- `device_map` はモデルをあなたのGPUに移動させます
-- `load_in_4bit` は[4ビットの動的量子化](main_classes/quantization)を適用してリソース要件を大幅に削減します
-
-モデルを初期化する他の方法もありますが、これはLLMを始めるための良い基準です。
-
-次に、[トークナイザ](tokenizer_summary)を使用してテキスト入力を前処理する必要があります。
-
-
-```py
->>> from transformers import AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
->>> model_inputs = tokenizer(["A list of colors: red, blue"], return_tensors="pt").to("cuda")
-```
-
-
-`model_inputs` 変数は、トークン化されたテキスト入力とアテンションマスクを保持しています。 [`~generation.GenerationMixin.generate`] は、アテンションマスクが渡されていない場合でも、最善の努力をしてそれを推測しようとしますが、できる限り渡すことをお勧めします。最適な結果を得るためです。
-
-最後に、[`~generation.GenerationMixin.generate`] メソッドを呼び出して生成されたトークンを取得し、それを表示する前にテキストに変換する必要があります。
-
-
-```py
->>> generated_ids = model.generate(**model_inputs)
->>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'A list of colors: red, blue, green, yellow, black, white, and brown'
-```
-
-これで完了です！わずかなコード行数で、LLM（Large Language Model）のパワーを活用できます。
-
-## Common pitfalls
-
-[生成戦略](generation_strategies)はたくさんあり、デフォルトの値があなたのユースケースに適していないことがあります。出力が期待通りでない場合、最も一般的な落とし穴とその回避方法のリストを作成しました。
-
-```py
->>> from transformers import AutoModelForCausalLM, AutoTokenizer
-
->>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
->>> tokenizer.pad_token = tokenizer.eos_token  # Llama has no pad token by default
->>> model = AutoModelForCausalLM.from_pretrained(
-...     "openlm-research/open_llama_7b", device_map="auto", load_in_4bit=True
-... )
-```
-
-### Generated output is too short/long
-
-[`~generation.GenerationConfig`] ファイルで指定されていない場合、`generate` はデフォルトで最大で 20 トークンまで返します。我々は `generate` コールで `max_new_tokens` を手動で設定することを強くお勧めします。これにより、返される新しいトークンの最大数を制御できます。LLM（正確には、[デコーダー専用モデル](https://huggingface.co/learn/nlp-course/chapter1/6?fw=pt)）も出力の一部として入力プロンプトを返すことに注意してください。
-
-```py
->>> model_inputs = tokenizer(["A sequence of numbers: 1, 2"], return_tensors="pt").to("cuda")
-
->>> # By default, the output will contain up to 20 tokens
->>> generated_ids = model.generate(**model_inputs)
->>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'A sequence of numbers: 1, 2, 3, 4, 5'
-
->>> # Setting `max_new_tokens` allows you to control the maximum length
->>> generated_ids = model.generate(**model_inputs, max_new_tokens=50)
->>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'A sequence of numbers: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,'
-```
-
-### Incorrect generation mode
-
-デフォルトでは、 [`~generation.GenerationConfig`] ファイルで指定されていない限り、`generate` は各イテレーションで最も可能性の高いトークンを選択します（貪欲デコーディング）。タスクに応じて、これは望ましくないことがあります。チャットボットやエッセイのような創造的なタスクでは、サンプリングが有益です。一方、音声の転写や翻訳のような入力に基づくタスクでは、貪欲デコーディングが有益です。`do_sample=True` でサンプリングを有効にできます。このトピックについての詳細は、この[ブログポスト](https://huggingface.co/blog/how-to-generate)で学ぶことができます。
-
-```py
->>> # Set seed or reproducibility -- you don't need this unless you want full reproducibility
->>> from transformers import set_seed
->>> set_seed(0)
-
->>> model_inputs = tokenizer(["I am a cat."], return_tensors="pt").to("cuda")
-
->>> # LLM + greedy decoding = repetitive, boring output
->>> generated_ids = model.generate(**model_inputs)
->>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'I am a cat. I am a cat. I am a cat. I am a cat'
-
->>> # With sampling, the output becomes more creative!
->>> generated_ids = model.generate(**model_inputs, do_sample=True)
->>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'I am a cat.\nI just need to be. I am always.\nEvery time'
-```
-
-### Wrong padding side
-
-LLM（Large Language Models）は[デコーダー専用](https://huggingface.co/learn/nlp-course/chapter1/6?fw=pt)のアーキテクチャであり、入力プロンプトを繰り返し処理することを意味します。入力が同じ長さでない場合、それらをパディングする必要があります。LLMはパッドトークンからの続きを学習していないため、入力は左パディングする必要があります。また、生成に対して注目マスクを渡し忘れないようにしてください！
-
-
-```py
->>> # The tokenizer initialized above has right-padding active by default: the 1st sequence,
->>> # which is shorter, has padding on the right side. Generation fails.
->>> model_inputs = tokenizer(
-...     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
-... ).to("cuda")
->>> generated_ids = model.generate(**model_inputs)
->>> tokenizer.batch_decode(generated_ids[0], skip_special_tokens=True)[0]
-''
-
->>> # With left-padding, it works as expected!
->>> tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b", padding_side="left")
->>> tokenizer.pad_token = tokenizer.eos_token  # Llama has no pad token by default
->>> model_inputs = tokenizer(
-...     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
-... ).to("cuda")
->>> generated_ids = model.generate(**model_inputs)
->>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-'1, 2, 3, 4, 5, 6,'
-```
-
-## Further resources
-
-オートリグレッシブ生成プロセスは比較的簡単ですが、LLMを最大限に活用することは多くの要素が絡むため、挑戦的な試みとなります。LLMの使用と理解をさらに深めるための次のステップについては以下のリソースをご覧ください。
-
-<!-- TODO: 新しいガイドで完了 -->
-### Advanced generate usage
-
-1. [ガイド](generation_strategies)：異なる生成方法を制御する方法、生成構成ファイルの設定方法、出力のストリーミング方法についてのガイド;
-2. [`~generation.GenerationConfig`]、[`~generation.GenerationMixin.generate`]、および[生成関連クラス](internal/generation_utils)に関するAPIリファレンス。
-
-### LLM leaderboards
-
-1. [Open LLM リーダーボード](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)：オープンソースモデルの品質に焦点を当てたリーダーボード;
-2. [Open LLM-Perf リーダーボード](https://huggingface.co/spaces/optimum/llm-perf-leaderboard)：LLMのスループットに焦点を当てたリーダーボード。
-
-### Latency and throughput
-
-1. [ガイド](main_classes/quantization)：ダイナミッククオンタイズに関するガイド。これによりメモリ要件を劇的に削減する方法が示されています。
-
-### Related libraries
-
-1. [`text-generation-inference`](https://github.com/huggingface/text-generation-inference)：LLM用の本番向けサーバー;
-2. [`optimum`](https://github.com/huggingface/optimum)：特定のハードウェアデバイス向けに最適化された🤗 Transformersの拡張。