Merge branch 'main' into videollama3

Author: lkhl

.circleci/create_circleci_config.py

@@ -29,6 +29,7 @@
     "RUN_PIPELINE_TESTS": False,
     # will be adjust in `CircleCIJob.to_dict`.
     "RUN_FLAKY": True,
+    "DISABLE_SAFETENSORS_CONVERSION": True,
 }
 # Disable the use of {"s": None} as the output is way too long, causing the navigation on CircleCI impractical
 COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "vvv": None, "rsfE":None}
@@ -129,6 +130,12 @@ def __post_init__(self):
 
     def to_dict(self):
         env = COMMON_ENV_VARIABLES.copy()
+        if self.job_name != "tests_hub":
+            # fmt: off
+            # not critical
+            env.update({"HF_TOKEN": "".join(["h", "f", "_", "H", "o", "d", "V", "u", "M", "q", "b", "R", "m", "t", "b", "z", "F", "Q", "O", "Q", "A", "J", "G", "D", "l", "V", "Q", "r", "R", "N", "w", "D", "M", "V", "C", "s", "d"])})
+            # fmt: on
+
         # Do not run tests decorated by @is_flaky on pull requests
         env['RUN_FLAKY'] = os.environ.get("CIRCLE_PULL_REQUEST", "") == ""
         env.update(self.additional_env)

.github/PULL_REQUEST_TEMPLATE.md

@@ -39,41 +39,40 @@ members/contributors who may be interested in your PR.
 
 Models:
 
-- text models: @ArthurZucker
-- vision models: @amyeroberts, @qubvel
-- speech models: @eustlb
+- text models: @ArthurZucker @Cyrilvallez
+- vision models: @yonigozlan @molbap
+- audio models: @eustlb @ebezzam @vasqu
+- multimodal models: @zucchini-nlp
 - graph models: @clefourrier
 
 Library:
 
-- flax: @gante and @Rocketknight1
 - generate: @zucchini-nlp (visual-language models) or @gante (all others)
+- continuous batching: @remi-or @ArthurZucker @McPatate
 - pipelines: @Rocketknight1
-- tensorflow: @gante and @Rocketknight1
-- tokenizers: @ArthurZucker
-- trainer: @zach-huggingface, @SunMarc and @qgallouedec
-- chat templates: @Rocketknight1
+- tokenizers: @ArthurZucker and @itazap
+- trainer: @zach-huggingface @SunMarc
+- attention: @vasqu @ArthurZucker @CyrilVallez
+- model loading (from pretrained, etc): @CyrilVallez
+- distributed: @3outeille @ArthurZucker @S1ro1
+- CIs: @ydshieh
 
 Integrations:
 
 - deepspeed: HF Trainer/Accelerate: @SunMarc @zach-huggingface
 - ray/raytune: @richardliaw, @amogkam
 - Big Model Inference: @SunMarc
 - quantization (bitsandbytes, autogpt): @SunMarc @MekkCyber
+- kernels: @MekkCyber @drbh
 
-Documentation: @stevhliu
-
-HF projects:
+Devices/Backends:
 
-- accelerate: [different repo](https://github.com/huggingface/accelerate)
-- datasets: [different repo](https://github.com/huggingface/datasets)
-- diffusers: [different repo](https://github.com/huggingface/diffusers)
-- rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
+- AMD ROCm: @ivarflakstad
+- Intel XPU: @IlyasMoutawwakil
+- Ascend NPU: @ivarflakstad 
 
-Maintained examples (not research project or legacy):
+Documentation: @stevhliu
 
-- Flax: @Rocketknight1
-- PyTorch: See Models above and tag the person corresponding to the modality of the example.
-- TensorFlow: @Rocketknight1
+Research projects are not maintained and should be taken as is.
 
  -->

.github/scripts/codeowners_for_review_action

@@ -7,8 +7,8 @@ docs/ @stevhliu
 /docker/ @ydshieh @ArthurZucker
 
 # More high-level globs catch cases when specific rules later don't apply
-/src/transformers/models/*/processing* @molbap @yonigozlan @qubvel
-/src/transformers/models/*/image_processing* @qubvel
+/src/transformers/models/*/processing* @molbap @yonigozlan
+/src/transformers/models/*/image_processing* @yonigozlan
 /src/transformers/models/*/image_processing_*_fast* @yonigozlan
 
 # Owners of subsections of the library
@@ -186,65 +186,65 @@ trainer_utils.py @zach-huggingface @SunMarc
 /src/transformers/models/zamba/mod*_zamba* @ArthurZucker
 
 # Vision models
-/src/transformers/models/beit/mod*_beit* @amyeroberts @qubvel
-/src/transformers/models/bit/mod*_bit* @amyeroberts @qubvel
-/src/transformers/models/conditional_detr/mod*_conditional_detr* @amyeroberts @qubvel
-/src/transformers/models/convnext/mod*_convnext* @amyeroberts @qubvel
-/src/transformers/models/convnextv2/mod*_convnextv2* @amyeroberts @qubvel
-/src/transformers/models/cvt/mod*_cvt* @amyeroberts @qubvel
-/src/transformers/models/deformable_detr/mod*_deformable_detr* @amyeroberts @qubvel
-/src/transformers/models/deit/mod*_deit* @amyeroberts @qubvel
-/src/transformers/models/depth_anything/mod*_depth_anything* @amyeroberts @qubvel
-/src/transformers/models/depth_anything_v2/mod*_depth_anything_v2* @amyeroberts @qubvel
-/src/transformers/models/deta/mod*_deta* @amyeroberts @qubvel
-/src/transformers/models/detr/mod*_detr* @amyeroberts @qubvel
-/src/transformers/models/dinat/mod*_dinat* @amyeroberts @qubvel
-/src/transformers/models/dinov2/mod*_dinov2* @amyeroberts @qubvel
-/src/transformers/models/dinov2_with_registers/mod*_dinov2_with_registers* @amyeroberts @qubvel
-/src/transformers/models/dit/mod*_dit* @amyeroberts @qubvel
-/src/transformers/models/dpt/mod*_dpt* @amyeroberts @qubvel
-/src/transformers/models/efficientformer/mod*_efficientformer* @amyeroberts @qubvel
-/src/transformers/models/efficientnet/mod*_efficientnet* @amyeroberts @qubvel
-/src/transformers/models/focalnet/mod*_focalnet* @amyeroberts @qubvel
-/src/transformers/models/glpn/mod*_glpn* @amyeroberts @qubvel
-/src/transformers/models/hiera/mod*_hiera* @amyeroberts @qubvel
-/src/transformers/models/ijepa/mod*_ijepa* @amyeroberts @qubvel
-/src/transformers/models/imagegpt/mod*_imagegpt* @amyeroberts @qubvel
-/src/transformers/models/levit/mod*_levit* @amyeroberts @qubvel
-/src/transformers/models/mask2former/mod*_mask2former* @amyeroberts @qubvel
-/src/transformers/models/maskformer/mod*_maskformer* @amyeroberts @qubvel
-/src/transformers/models/mobilenet_v1/mod*_mobilenet_v1* @amyeroberts @qubvel
-/src/transformers/models/mobilenet_v2/mod*_mobilenet_v2* @amyeroberts @qubvel
-/src/transformers/models/mobilevit/mod*_mobilevit* @amyeroberts @qubvel
-/src/transformers/models/mobilevitv2/mod*_mobilevitv2* @amyeroberts @qubvel
-/src/transformers/models/nat/mod*_nat* @amyeroberts @qubvel
-/src/transformers/models/poolformer/mod*_poolformer* @amyeroberts @qubvel
-/src/transformers/models/pvt/mod*_pvt* @amyeroberts @qubvel
-/src/transformers/models/pvt_v2/mod*_pvt_v2* @amyeroberts @qubvel
-/src/transformers/models/regnet/mod*_regnet* @amyeroberts @qubvel
-/src/transformers/models/resnet/mod*_resnet* @amyeroberts @qubvel
-/src/transformers/models/rt_detr/mod*_rt_detr* @amyeroberts @qubvel
-/src/transformers/models/segformer/mod*_segformer* @amyeroberts @qubvel
-/src/transformers/models/seggpt/mod*_seggpt* @amyeroberts @qubvel
-/src/transformers/models/superpoint/mod*_superpoint* @amyeroberts @qubvel
-/src/transformers/models/swiftformer/mod*_swiftformer* @amyeroberts @qubvel
-/src/transformers/models/swin/mod*_swin* @amyeroberts @qubvel
-/src/transformers/models/swinv2/mod*_swinv2* @amyeroberts @qubvel
-/src/transformers/models/swin2sr/mod*_swin2sr* @amyeroberts @qubvel
-/src/transformers/models/table_transformer/mod*_table_transformer* @amyeroberts @qubvel
-/src/transformers/models/textnet/mod*_textnet* @amyeroberts @qubvel
-/src/transformers/models/timm_wrapper/mod*_timm_wrapper* @amyeroberts @qubvel
-/src/transformers/models/upernet/mod*_upernet* @amyeroberts @qubvel
-/src/transformers/models/van/mod*_van* @amyeroberts @qubvel
-/src/transformers/models/vit/mod*_vit* @amyeroberts @qubvel
-/src/transformers/models/vit_hybrid/mod*_vit_hybrid* @amyeroberts @qubvel
-/src/transformers/models/vitdet/mod*_vitdet* @amyeroberts @qubvel
-/src/transformers/models/vit_mae/mod*_vit_mae* @amyeroberts @qubvel
-/src/transformers/models/vitmatte/mod*_vitmatte* @amyeroberts @qubvel
-/src/transformers/models/vit_msn/mod*_vit_msn* @amyeroberts @qubvel
-/src/transformers/models/vitpose/mod*_vitpose* @amyeroberts @qubvel
-/src/transformers/models/yolos/mod*_yolos* @amyeroberts @qubvel
-/src/transformers/models/zoedepth/mod*_zoedepth* @amyeroberts @qubvel
+/src/transformers/models/beit/mod*_beit* @yonigozlan @molbap
+/src/transformers/models/bit/mod*_bit* @yonigozlan @molbap
+/src/transformers/models/conditional_detr/mod*_conditional_detr* @yonigozlan @molbap
+/src/transformers/models/convnext/mod*_convnext* @yonigozlan @molbap
+/src/transformers/models/convnextv2/mod*_convnextv2* @yonigozlan @molbap
+/src/transformers/models/cvt/mod*_cvt* @yonigozlan @molbap
+/src/transformers/models/deformable_detr/mod*_deformable_detr* @yonigozlan @molbap
+/src/transformers/models/deit/mod*_deit* @yonigozlan @molbap
+/src/transformers/models/depth_anything/mod*_depth_anything* @yonigozlan @molbap
+/src/transformers/models/depth_anything_v2/mod*_depth_anything_v2* @yonigozlan @molbap
+/src/transformers/models/deta/mod*_deta* @yonigozlan @molbap
+/src/transformers/models/detr/mod*_detr* @yonigozlan @molbap
+/src/transformers/models/dinat/mod*_dinat* @yonigozlan @molbap
+/src/transformers/models/dinov2/mod*_dinov2* @yonigozlan @molbap
+/src/transformers/models/dinov2_with_registers/mod*_dinov2_with_registers* @yonigozlan @molbap
+/src/transformers/models/dit/mod*_dit* @yonigozlan @molbap
+/src/transformers/models/dpt/mod*_dpt* @yonigozlan @molbap
+/src/transformers/models/efficientformer/mod*_efficientformer* @yonigozlan @molbap
+/src/transformers/models/efficientnet/mod*_efficientnet* @yonigozlan @molbap
+/src/transformers/models/focalnet/mod*_focalnet* @yonigozlan @molbap
+/src/transformers/models/glpn/mod*_glpn* @yonigozlan @molbap
+/src/transformers/models/hiera/mod*_hiera* @yonigozlan @molbap
+/src/transformers/models/ijepa/mod*_ijepa* @yonigozlan @molbap
+/src/transformers/models/imagegpt/mod*_imagegpt* @yonigozlan @molbap
+/src/transformers/models/levit/mod*_levit* @yonigozlan @molbap
+/src/transformers/models/mask2former/mod*_mask2former* @yonigozlan @molbap
+/src/transformers/models/maskformer/mod*_maskformer* @yonigozlan @molbap
+/src/transformers/models/mobilenet_v1/mod*_mobilenet_v1* @yonigozlan @molbap
+/src/transformers/models/mobilenet_v2/mod*_mobilenet_v2* @yonigozlan @molbap
+/src/transformers/models/mobilevit/mod*_mobilevit* @yonigozlan @molbap
+/src/transformers/models/mobilevitv2/mod*_mobilevitv2* @yonigozlan @molbap
+/src/transformers/models/nat/mod*_nat* @yonigozlan @molbap
+/src/transformers/models/poolformer/mod*_poolformer* @yonigozlan @molbap
+/src/transformers/models/pvt/mod*_pvt* @yonigozlan @molbap
+/src/transformers/models/pvt_v2/mod*_pvt_v2* @yonigozlan @molbap
+/src/transformers/models/regnet/mod*_regnet* @yonigozlan @molbap
+/src/transformers/models/resnet/mod*_resnet* @yonigozlan @molbap
+/src/transformers/models/rt_detr/mod*_rt_detr* @yonigozlan @molbap
+/src/transformers/models/segformer/mod*_segformer* @yonigozlan @molbap
+/src/transformers/models/seggpt/mod*_seggpt* @yonigozlan @molbap
+/src/transformers/models/superpoint/mod*_superpoint* @yonigozlan @molbap
+/src/transformers/models/swiftformer/mod*_swiftformer* @yonigozlan @molbap
+/src/transformers/models/swin/mod*_swin* @yonigozlan @molbap
+/src/transformers/models/swinv2/mod*_swinv2* @yonigozlan @molbap
+/src/transformers/models/swin2sr/mod*_swin2sr* @yonigozlan @molbap
+/src/transformers/models/table_transformer/mod*_table_transformer* @yonigozlan @molbap
+/src/transformers/models/textnet/mod*_textnet* @yonigozlan @molbap
+/src/transformers/models/timm_wrapper/mod*_timm_wrapper* @yonigozlan @molbap
+/src/transformers/models/upernet/mod*_upernet* @yonigozlan @molbap
+/src/transformers/models/van/mod*_van* @yonigozlan @molbap
+/src/transformers/models/vit/mod*_vit* @yonigozlan @molbap
+/src/transformers/models/vit_hybrid/mod*_vit_hybrid* @yonigozlan @molbap
+/src/transformers/models/vitdet/mod*_vitdet* @yonigozlan @molbap
+/src/transformers/models/vit_mae/mod*_vit_mae* @yonigozlan @molbap
+/src/transformers/models/vitmatte/mod*_vitmatte* @yonigozlan @molbap
+/src/transformers/models/vit_msn/mod*_vit_msn* @yonigozlan @molbap
+/src/transformers/models/vitpose/mod*_vitpose* @yonigozlan @molbap
+/src/transformers/models/yolos/mod*_yolos* @yonigozlan @molbap
+/src/transformers/models/zoedepth/mod*_zoedepth* @yonigozlan @molbap
 
 # Audio models
 /src/transformers/models/audio_spectrogram_transformer/mod*_audio_spectrogram_transformer* @eustlb
@@ -304,7 +304,7 @@ trainer_utils.py @zach-huggingface @SunMarc
 /src/transformers/models/donut/mod*_donut* @zucchini-nlp
 /src/transformers/models/flava/mod*_flava* @zucchini-nlp
 /src/transformers/models/git/mod*_git* @zucchini-nlp
-/src/transformers/models/grounding_dino/mod*_grounding_dino* @qubvel
+/src/transformers/models/grounding_dino/mod*_grounding_dino* @yonigozlan
 /src/transformers/models/groupvit/mod*_groupvit* @zucchini-nlp
 /src/transformers/models/idefics/mod*_idefics* @zucchini-nlp
 /src/transformers/models/idefics2/mod*_idefics2* @zucchini-nlp
@@ -326,10 +326,10 @@ trainer_utils.py @zach-huggingface @SunMarc
 /src/transformers/models/mgp_str/mod*_mgp_str* @zucchini-nlp
 /src/transformers/models/mllama/mod*_mllama* @zucchini-nlp
 /src/transformers/models/nougat/mod*_nougat* @NielsRogge
-/src/transformers/models/omdet_turbo/mod*_omdet_turbo* @qubvel @yonigozlan
+/src/transformers/models/omdet_turbo/mod*_omdet_turbo* @yonigozlan
 /src/transformers/models/oneformer/mod*_oneformer* @zucchini-nlp
-/src/transformers/models/owlvit/mod*_owlvit* @qubvel
-/src/transformers/models/owlv2/mod*_owlv2* @qubvel
+/src/transformers/models/owlvit/mod*_owlvit* @yonigozlan
+/src/transformers/models/owlv2/mod*_owlv2* @yonigozlan
 /src/transformers/models/paligemma/mod*_paligemma* @zucchini-nlp @molbap
 /src/transformers/models/perceiver/mod*_perceiver* @zucchini-nlp
 /src/transformers/models/pix2struct/mod*_pix2struct* @zucchini-nlp

.github/workflows/benchmark_v2.yml

@@ -7,6 +7,14 @@ on:
         description: 'GH Actions runner group to use'
         required: true
         type: string
+      container_image:
+        description: 'Docker image to use'
+        required: true
+        type: string
+      container_options:
+        description: 'Container options to use'
+        required: true
+        type: string
       commit_sha:
         description: 'Commit SHA to benchmark'
         required: false
@@ -38,8 +46,8 @@ jobs:
       (github.event_name == 'pull_request' && contains( github.event.pull_request.labels.*.name, 'run-benchmark')) ||
       (github.event_name == 'schedule')
     container:
-      image: huggingface/transformers-pytorch-gpu
-      options: --gpus all --privileged --ipc host --shm-size "16gb"
+      image: ${{ inputs.container_image }}
+      options: ${{ inputs.container_options }}
     steps:
       - name: Get repo
         uses: actions/checkout@v4

.github/workflows/benchmark_v2_a10_caller.yml

@@ -13,6 +13,8 @@ jobs:
     uses: ./.github/workflows/benchmark_v2.yml
     with:
       runner: aws-g5-4xlarge-cache-use1-public-80
+      container_image: huggingface/transformers-pytorch-gpu
+      container_options: --gpus all --privileged --ipc host --shm-size "16gb"
       commit_sha: ${{ github.sha }}
       run_id: ${{ github.run_id }}
       benchmark_repo_id: hf-internal-testing/transformers-daily-benchmarks

.github/workflows/benchmark_v2_mi325_caller.yml

@@ -13,6 +13,8 @@ jobs:
     uses: ./.github/workflows/benchmark_v2.yml
     with:
       runner: amd-mi325-ci-1gpu
+      container_image: huggingface/transformers-pytorch-amd-gpu
+      container_options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache
       commit_sha: ${{ github.sha }}
       run_id: ${{ github.run_id }}
       benchmark_repo_id: hf-internal-testing/transformers-daily-benchmarks

.github/workflows/self-scheduled-amd-mi325-caller.yml

@@ -20,7 +20,7 @@ jobs:
     with:
       job: run_models_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi325-ci
+      runner_group: amd-mi325
       docker: huggingface/transformers-pytorch-amd-gpu
       ci_event: Scheduled CI (AMD) - mi325
       report_repo_id: optimum-amd/transformers_daily_ci
@@ -33,7 +33,7 @@ jobs:
     with:
       job: run_pipelines_torch_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi325-ci
+      runner_group: amd-mi325
       docker: huggingface/transformers-pytorch-amd-gpu
       ci_event: Scheduled CI (AMD) - mi325
       report_repo_id: optimum-amd/transformers_daily_ci
@@ -46,7 +46,7 @@ jobs:
     with:
       job: run_examples_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi325-ci
+      runner_group: amd-mi325
       docker: huggingface/transformers-pytorch-amd-gpu
       ci_event: Scheduled CI (AMD) - mi325
       report_repo_id: optimum-amd/transformers_daily_ci
@@ -59,7 +59,7 @@ jobs:
     with:
       job: run_torch_cuda_extensions_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi325-ci
+      runner_group: amd-mi325
       docker: huggingface/transformers-pytorch-deepspeed-amd-gpu
       ci_event: Scheduled CI (AMD) - mi325
       report_repo_id: optimum-amd/transformers_daily_ci

.github/workflows/self-scheduled-amd-mi355-caller.yml

@@ -3,7 +3,7 @@ name: Self-hosted runner scale set (AMD mi355 scheduled CI caller)
 # Note: For every job in this workflow, the name of the runner scale set is finalized in the runner yaml i.e. huggingface/hf-workflows/.github/workflows/transformers_amd_ci_scheduled_arc_scale_set.yaml
 # For example, 1gpu : amd-mi355-ci-1gpu
 #              2gpu : amd-mi355-ci-2gpu
-
+ 
 on:
   workflow_run:
     workflows: ["Self-hosted runner (AMD scheduled CI caller)"]
@@ -20,7 +20,7 @@ jobs:
     with:
       job: run_models_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi355-ci
+      runner_group: hfc-amd-mi355
       docker: huggingface/testing-rocm7.0-preview
       ci_event: Scheduled CI (AMD) - mi355
       report_repo_id: hf-transformers-bot/transformers-ci-dummy
@@ -32,7 +32,7 @@ jobs:
     with:
       job: run_pipelines_torch_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi355-ci
+      runner_group: hfc-amd-mi355
       docker: huggingface/testing-rocm7.0-preview
       ci_event: Scheduled CI (AMD) - mi355
       report_repo_id: hf-transformers-bot/transformers-ci-dummy
@@ -44,7 +44,7 @@ jobs:
     with:
       job: run_examples_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi355-ci
+      runner_group: hfc-amd-mi355
       docker: huggingface/testing-rocm7.0-preview
       ci_event: Scheduled CI (AMD) - mi355
       report_repo_id: hf-transformers-bot/transformers-ci-dummy
@@ -53,10 +53,10 @@ jobs:
   deepspeed-ci:
     name: DeepSpeed CI
     uses: huggingface/hf-workflows/.github/workflows/transformers_amd_ci_scheduled_arc_scale_set.yaml@main
-    with:
+    with:  
       job: run_torch_cuda_extensions_gpu
       slack_report_channel: "#amd-hf-ci"
-      runner_scale_set: amd-mi355-ci
+      runner_group: hfc-amd-mi355
       docker: huggingface/testing-rocm7.0-preview
       ci_event: Scheduled CI (AMD) - mi355
       report_repo_id: hf-transformers-bot/transformers-ci-dummy

CONTRIBUTING.md

@@ -278,13 +278,14 @@ are working on it).<br>
 useful to avoid duplicated work, and to differentiate it from PRs ready to be merged.<br>
 ☐ Make sure existing tests pass.<br>
 ☐ If adding a new feature, also add tests for it.<br>
-   - If you are adding a new model, make sure you use
+
+- If you are adding a new model, make sure you use
      `ModelTester.all_model_classes = (MyModel, MyModelWithLMHead,...)` to trigger the common tests.
-   - If you are adding new `@slow` tests, make sure they pass using
+- If you are adding new `@slow` tests, make sure they pass using
      `RUN_SLOW=1 python -m pytest tests/models/my_new_model/test_my_new_model.py`.
-   - If you are adding a new tokenizer, write tests and make sure
+- If you are adding a new tokenizer, write tests and make sure
      `RUN_SLOW=1 python -m pytest tests/models/{your_model_name}/test_tokenization_{your_model_name}.py` passes.
-   - CircleCI does not run the slow tests, but GitHub Actions does every night!<br>
+- CircleCI does not run the slow tests, but GitHub Actions does every night!<br>
 
 ☐ All public methods must have informative docstrings (see
 [`modeling_bert.py`](https://github.com/huggingface/transformers/blob/main/src/transformers/models/bert/modeling_bert.py)
@@ -340,6 +341,7 @@ RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/t
 ```
 
 Like the slow tests, there are other environment variables available which are not enabled by default during testing:
+
 - `RUN_CUSTOM_TOKENIZERS`: Enables tests for custom tokenizers.
 
 More environment variables and additional information can be found in the [testing_utils.py](https://github.com/huggingface/transformers/blob/main/src/transformers/testing_utils.py).