Merge branch 'main' into trainer_seq2seq-compute_loss_func

.circleci/create_circleci_config.py

@@ -55,6 +55,7 @@ def to_dict(self):
 
         return {
             "docker": copy.deepcopy(DEFAULT_DOCKER_IMAGE),
+            "resource_class": "small",
             "steps": steps,
         }
 
@@ -67,9 +68,9 @@ class CircleCIJob:
     install_steps: List[str] = None
     marker: Optional[str] = None
     parallelism: Optional[int] = 0
-    pytest_num_workers: int = 12
+    pytest_num_workers: int = 8
     pytest_options: Dict[str, Any] = None
-    resource_class: Optional[str] = "2xlarge"
+    resource_class: Optional[str] = "xlarge"
     tests_to_run: Optional[List[str]] = None
     num_test_files_per_worker: Optional[int] = 10
     # This should be only used for doctest job!
@@ -198,44 +199,40 @@ def job_name(self):
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     marker="not generate",
     parallelism=6,
-    pytest_num_workers=8
 )
 
 generate_job = CircleCIJob(
     "generate",
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     marker="generate",
     parallelism=6,
-    pytest_num_workers=8
 )
 
 tokenization_job = CircleCIJob(
     "tokenization",
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     parallelism=8,
-    pytest_num_workers=16
 )
 
 processor_job = CircleCIJob(
     "processors",
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     parallelism=8,
-    pytest_num_workers=6
 )
 
 tf_job = CircleCIJob(
     "tf",
     docker_image=[{"image":"huggingface/transformers-tf-light"}],
     parallelism=6,
-    pytest_num_workers=16,
 )
 
 
 flax_job = CircleCIJob(
     "flax",
     docker_image=[{"image":"huggingface/transformers-jax-light"}],
     parallelism=6,
-    pytest_num_workers=16
+    pytest_num_workers=16,
+    resource_class="2xlarge",
 )
 
 
@@ -244,7 +241,7 @@ def job_name(self):
     additional_env={"RUN_PIPELINE_TESTS": True},
     docker_image=[{"image":"huggingface/transformers-torch-light"}],
     marker="is_pipeline_test",
-    parallelism=4
+    parallelism=4,
 )
 
 
@@ -253,7 +250,7 @@ def job_name(self):
     additional_env={"RUN_PIPELINE_TESTS": True},
     docker_image=[{"image":"huggingface/transformers-tf-light"}],
     marker="is_pipeline_test",
-    parallelism=4
+    parallelism=4,
 )
 
 
@@ -270,15 +267,13 @@ def job_name(self):
     docker_image=[{"image":"huggingface/transformers-examples-torch"}],
     # TODO @ArthurZucker remove this once docker is easier to build
     install_steps=["uv venv && uv pip install . && uv pip install -r examples/pytorch/_tests_requirements.txt"],
-    pytest_num_workers=8,
 )
 
 
 examples_tensorflow_job = CircleCIJob(
     "examples_tensorflow",
     additional_env={"OMP_NUM_THREADS": 8},
     docker_image=[{"image":"huggingface/transformers-examples-tf"}],
-    pytest_num_workers=16,
 )
 
 
@@ -293,6 +288,7 @@ def job_name(self):
     ],
     marker="is_staging_test",
     pytest_num_workers=2,
+    resource_class="medium",
 )
 
 
@@ -305,13 +301,13 @@ def job_name(self):
     ],
     pytest_options={"k onnx": None},
     pytest_num_workers=1,
+    resource_class="small",
 )
 
 
 exotic_models_job = CircleCIJob(
     "exotic_models",
     docker_image=[{"image":"huggingface/transformers-exotic-models"}],
-    pytest_num_workers=12,
     parallelism=4,
     pytest_options={"durations": 100},
 )
@@ -330,7 +326,6 @@ def job_name(self):
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
     marker="not generate",
     parallelism=6,
-    pytest_num_workers=8,
 )
 
 

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

@@ -21,39 +21,6 @@ jobs:
           echo "$(python3 -c 'print(int(${{ github.run_number }}) % 10)')"
           echo "run_number=$(python3 -c 'print(int(${{ github.run_number }}) % 10)')" >> $GITHUB_OUTPUT
 
-  run_past_ci_pytorch_1-13:
-    name: PyTorch 1.13
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 0 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: pytorch
-      version: "1.13"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_pytorch_1-12:
-    name: PyTorch 1.12
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 1 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: pytorch
-      version: "1.12"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_pytorch_1-11:
-    name: PyTorch 1.11
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 2 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: pytorch
-      version: "1.11"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
   run_past_ci_tensorflow_2-11:
     name: TensorFlow 2.11
     needs: get_number

README.md

@@ -249,7 +249,7 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 
 ### With pip
 
-This repository is tested on Python 3.9+, Flax 0.4.1+, PyTorch 1.11+, and TensorFlow 2.6+.
+This repository is tested on Python 3.9+, Flax 0.4.1+, PyTorch 2.0+, 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/).
 

docker/transformers-quantization-latest-gpu/Dockerfile

@@ -50,6 +50,9 @@ RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/pef
 # Add aqlm for quantization testing
 RUN python3 -m pip install --no-cache-dir aqlm[gpu]==1.0.2
 
+# Add vptq for quantization testing
+RUN python3 -m pip install --no-cache-dir vptq
+
 # Add hqq for quantization testing
 RUN python3 -m pip install --no-cache-dir hqq
 

docs/source/ar/_toctree.yml

@@ -157,6 +157,8 @@
 #     title: AWQ
 #   - local: quantization/aqlm
 #     title: AQLM
+#   - local: quantization/vptq
+#     title: VPTQ
 #   - local: quantization/quanto
 #     title: Quanto
 #   - local: quantization/eetq

docs/source/en/_toctree.yml

@@ -167,6 +167,8 @@
     title: AWQ
   - local: quantization/aqlm
     title: AQLM
+  - local: quantization/vptq
+    title: VPTQ
   - local: quantization/quanto
     title: Quanto
   - local: quantization/eetq

docs/source/en/add_new_pipeline.md

@@ -184,7 +184,7 @@ class PairClassificationPipeline(Pipeline):
 ```
 
 The implementation is framework agnostic, and will work for PyTorch and TensorFlow models. If we have saved this in
-a file named `pair_classification.py`, we can then import it and register it like this. The [register_pipeline](https://github.com/huggingface/transformers/blob/9feae5fb0164e89d4998e5776897c16f7330d3df/src/transformers/pipelines/base.py#L1387) function registers the pipeline details (task type, pipeline class, supported backends) to a models `config.json` file.
+a file named `pair_classification.py`, we can then import it and register it like this.
 
 ```py
 from pair_classification import PairClassificationPipeline
@@ -199,6 +199,22 @@ PIPELINE_REGISTRY.register_pipeline(
 )
 ```
 
+The [register_pipeline](https://github.com/huggingface/transformers/blob/9feae5fb0164e89d4998e5776897c16f7330d3df/src/transformers/pipelines/base.py#L1387) function registers the pipeline details (task type, pipeline class, supported backends) to a models `config.json` file.
+
+```json
+  "custom_pipelines": {
+    "pair-classification": {
+      "impl": "pair_classification.PairClassificationPipeline",
+      "pt": [
+        "AutoModelForSequenceClassification"
+      ],
+      "tf": [
+        "TFAutoModelForSequenceClassification"
+      ],
+    }
+  },
+```
+
 Once this is done, we can use it with a pretrained model. For instance `sgugger/finetuned-bert-mrpc` has been
 fine-tuned on the MRPC dataset, which classifies pairs of sentences as paraphrases or not.
 

docs/source/en/internal/generation_utils.md

@@ -352,6 +352,8 @@ A [`Constraint`] can be used to force the generation to include specific tokens
 
 [[autodoc]] TextIteratorStreamer
 
+[[autodoc]] AsyncTextIteratorStreamer
+
 ## Caches
 
 [[autodoc]] Cache

docs/source/en/llm_optims.md

@@ -473,7 +473,7 @@ with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable
 Quantization reduces the size of the LLM weights by storing them in a lower precision. This translates to lower memory usage and makes loading LLMs for inference more accessible if you're constrained by your GPUs memory. If you aren't limited by your GPU, you don't necessarily need to quantize your model because it can incur a small latency cost (except for AWQ and fused AWQ modules) due to the extra step required to quantize and dequantize the weights.
 
 > [!TIP]
-> There are many quantization libraries (see the [Quantization](./quantization) guide for more details) available, such as Quanto, AQLM, AWQ, and AutoGPTQ. Feel free to try them out and see which one works best for your use case. We also recommend reading the [Overview of natively supported quantization schemes in 🤗 Transformers](https://hf.co/blog/overview-quantization-transformers) blog post which compares AutoGPTQ and bitsandbytes.
+> There are many quantization libraries (see the [Quantization](./quantization) guide for more details) available, such as Quanto, AQLM, VPTQ, AWQ, and AutoGPTQ. Feel free to try them out and see which one works best for your use case. We also recommend reading the [Overview of natively supported quantization schemes in 🤗 Transformers](https://hf.co/blog/overview-quantization-transformers) blog post which compares AutoGPTQ and bitsandbytes.
 
 Use the Model Memory Calculator below to estimate and compare how much memory is required to load a model. For example, try estimating how much memory it costs to load [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1).
 

docs/source/en/main_classes/quantization.md

@@ -34,6 +34,10 @@ Learn how to quantize models in the [Quantization](../quantization) guide.
 
 [[autodoc]] AqlmConfig
 
+## VptqConfig
+
+[[autodoc]] VptqConfig
+
 ## AwqConfig
 
 [[autodoc]] AwqConfig

docs/source/en/modular_transformers.md

@@ -22,6 +22,9 @@ etc. Model contribution PRs rarely add less than 3-5k lines of code, with much o
 This raises the bar for contributions, and with Modular Transformers, we're aiming to lower the bar to a much more
 acceptable point.
 
+If you plan to add a model to `transformers` make sure you read [How to add a model to 🤗 Transformers?](https://huggingface.co/docs/transformers/add_new_model).
+For any kind of contributions, see [CONTRIBUTING.md](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md).
+
 ## What is it?
 
 Modular Transformers introduces the concept of a "modular" file to a model folder. This modular file accepts code
@@ -43,6 +46,12 @@ be moved to the new Modular Transformers format in the coming months.
 
 ### Details 
 
+To generate a single file from the modular file, run the following command.
+
+```bash
+python utils/modular_model_converter.py --files-to-parse src/transformers/models/<your_model>/modular_<your_model>.py
+```
+
 The "linter", which unravels the inheritance and creates all single-files from the modular file, will flatten the 
 inheritance while trying to be invisible to Python users. At this time, the linter flattens a **single** level of
 inheritance.
@@ -59,7 +68,11 @@ file, and the corresponding files will be created for you.
 
 ### Enforcement
 
-[TODO] We are introducing a new test, that makes sure the generated content matches what is present in the `modular_xxxx.py`
+Run the command below to ensure the generated content matches `modular_<your_model>.py`
+
+```bash
+python utils/check_modular_conversion.py --files src/transformers/models/<your_model>/modular_<your_model>.py
+```
 
 ### Examples
 
@@ -194,4 +207,4 @@ We now also support special cases like
 class GemmaVisionModel(CLIPModel):                                 
     pass
 ```
-where the name of your class `GemmaVision` is not the same as the modular `Gemma`. This is super useful for composite models.
+where the name of your class `GemmaVision` is not the same as the modular `Gemma`. This is super useful for composite models.

docs/source/en/quantization/overview.md

@@ -58,6 +58,7 @@ Use the table below to help you decide which quantization method to use.
 | [optimum-quanto](./quanto)                              | 🟢                       | 🟢   | 🟢        | 🔴              | 🟢                     | 🔴         | 🟢                       | 2 / 4 / 8      | 🔴                                   | 🔴            | 🟢                      | https://github.com/huggingface/optimum-quanto       |
 | [FBGEMM_FP8](./fbgemm_fp8.md)                              | 🟢                       | 🔴    | 🟢        | 🔴              | 🔴                      | 🔴         | 🔴                        | 8      | 🔴                                   | 🟢            | 🟢                      | https://github.com/pytorch/FBGEMM       |
 | [torchao](./torchao.md)                              | 🟢                       |     | 🟢        | 🔴              | partial support (int4 weight only)       | 🔴         |                       | 4 / 8      |                                   | 🟢🔴           | 🟢                      | https://github.com/pytorch/ao       |
+| [VPTQ](./vptq)                      | 🔴                       |  🔴   |     🟢     | 🟡              | 🔴      | 🔴                | 🟢                      | 1 - 8          | 🔴                                   | 🟢            | 🟢                      | https://github.com/microsoft/VPTQ            |
 
 <Tip>
 
@@ -71,4 +72,4 @@ We value your feedback to help identify bugs before the full release! Check out
 
 \** bitsandbytes is seeking contributors to help develop and lead the Apple Silicon backend. Interested? Contact them directly via their repo. Stipends may be available through sponsorships.
 
-</Tip>
+</Tip>