debugdebug

Update README.md

Updated the tested versions

.circleci/create_circleci_config.py

@@ -230,6 +230,8 @@ torch_and_tf_job = CircleCIJob(
         "pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]",
         "pip install tensorflow_probability",
         "pip install git+https://github.com/huggingface/accelerate",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     marker="is_pt_tf_cross_test",
     pytest_options={"rA": None, "durations": 0},
@@ -244,6 +246,8 @@ torch_and_flax_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]",
         "pip install git+https://github.com/huggingface/accelerate",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     marker="is_pt_flax_cross_test",
     pytest_options={"rA": None, "durations": 0},
@@ -257,6 +261,8 @@ torch_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]",
         "pip install git+https://github.com/huggingface/accelerate",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     parallelism=1,
     pytest_num_workers=3,
@@ -270,6 +276,8 @@ tf_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]",
         "pip install tensorflow_probability",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     parallelism=1,
     pytest_num_workers=6,
@@ -283,6 +291,8 @@ flax_job = CircleCIJob(
         "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
         "pip install --upgrade pip",
         "pip install .[flax,testing,sentencepiece,flax-speech,vision]",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     parallelism=1,
     pytest_options={"rA": None},
@@ -296,6 +306,8 @@ pipelines_torch_job = CircleCIJob(
         "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
         "pip install --upgrade pip",
         "pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm,video]",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     pytest_options={"rA": None},
     marker="is_pipeline_test",
@@ -310,6 +322,8 @@ pipelines_tf_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[sklearn,tf-cpu,testing,sentencepiece,vision]",
         "pip install tensorflow_probability",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     pytest_options={"rA": None},
     marker="is_pipeline_test",
@@ -334,6 +348,8 @@ custom_tokenizers_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[ja,testing,sentencepiece,jieba,spacy,ftfy,rjieba]",
         "python -m unidic download",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     parallelism=None,
     resource_class=None,
@@ -353,6 +369,8 @@ examples_torch_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[sklearn,torch,sentencepiece,testing,torch-speech]",
         "pip install -r examples/pytorch/_tests_requirements.txt",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
 )
 
@@ -365,6 +383,8 @@ examples_tensorflow_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[sklearn,tensorflow,sentencepiece,testing]",
         "pip install -r examples/tensorflow/_tests_requirements.txt",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
 )
 
@@ -376,6 +396,8 @@ examples_flax_job = CircleCIJob(
         "pip install --upgrade pip",
         "pip install .[flax,testing,sentencepiece]",
         "pip install -r examples/flax/_tests_requirements.txt",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
 )
 
@@ -388,6 +410,8 @@ hub_job = CircleCIJob(
         'git config --global user.name "ci"',
         "pip install --upgrade pip",
         "pip install .[torch,sentencepiece,testing]",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     marker="is_staging_test",
     pytest_num_workers=1,
@@ -400,6 +424,8 @@ onnx_job = CircleCIJob(
         "sudo apt-get -y update && sudo apt-get install -y cmake",
         "pip install --upgrade pip",
         "pip install .[torch,tf,testing,sentencepiece,onnxruntime,vision,rjieba]",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     pytest_options={"k onnx": None},
     pytest_num_workers=1,
@@ -418,6 +444,8 @@ exotic_models_job = CircleCIJob(
         "sudo apt install tesseract-ocr",
         "pip install pytesseract",
         "pip install natten",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     tests_to_run=[
         "tests/models/*layoutlmv*",
@@ -434,6 +462,8 @@ repo_utils_job = CircleCIJob(
     install_steps=[
         "pip install --upgrade pip",
         "pip install .[quality,testing,torch]",
+        "pip uninstall -y huggingface_hub",
+        "pip install git+https://github.com/huggingface/huggingface_hub.git@debugdebug",
     ],
     parallelism=None,
     pytest_num_workers=1,

CONTRIBUTING.md

@@ -275,7 +275,7 @@ You'll need **[Python 3.7]((https://github.com/huggingface/transformers/blob/mai
 request description to make sure they are linked (and people viewing the issue know you
 are working on it).<br>
 ☐ To indicate a work in progress please prefix the title with `[WIP]`. These are
-useful to avoid duplicated work, and to differentiate it from PRs ready to be merged.
+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
@@ -284,7 +284,7 @@ useful to avoid duplicated work, and to differentiate it from PRs ready to be me
      `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
      `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)

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 3.6+, Flax 0.3.2+, PyTorch 1.3.1+ and TensorFlow 2.3+.
+This repository is tested on Python 3.7+, Flax 0.4.1+, PyTorch 1.9+ and TensorFlow 2.4+.
 
 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/).
 
@@ -292,7 +292,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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/main/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. **[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. **[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.
@@ -346,6 +346,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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/main/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.
@@ -406,7 +407,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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/main/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. **[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. **[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. **[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.
@@ -448,7 +449,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 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/main/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. **[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.

README_es.md

@@ -267,7 +267,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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/main/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. **[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. **[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.
@@ -321,6 +321,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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/main/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.
@@ -381,7 +382,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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/main/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. **[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. **[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. **[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.
@@ -423,7 +424,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 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/main/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. **[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.

README_hd.md

@@ -239,7 +239,7 @@ conda install -c huggingface transformers
 1. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (Google Research से) Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig. द्वाराअनुसंधान पत्र [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) के साथ जारी किया गया
 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/main/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. **[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. **[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।
@@ -293,6 +293,7 @@ conda install -c huggingface transformers
 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)** (Google रिसर्च/स्टैनफोर्ड यूनिवर्सिटी से) साथ में दिया गया पेपर [इलेक्ट्रा: जेनरेटर के बजाय भेदभाव करने वाले के रूप में टेक्स्ट एन्कोडर्स का पूर्व-प्रशिक्षण] (https://arxiv.org/abs/2003.10555) केविन क्लार्क, मिन्ह-थांग लुओंग, क्वोक वी. ले, क्रिस्टोफर डी. मैनिंग द्वारा पोस्ट किया गया।
+1. **[EnCodec](https://huggingface.co/docs/transformers/main/model_doc/encodec)** (Meta AI से) Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi. द्वाराअनुसंधान पत्र [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) के साथ जारी किया गया
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (Google रिसर्च से) साथ में दिया गया 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. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)**(Baidu से) साथ देने वाला पेपर [ERNIE: एन्हांस्ड रिप्रेजेंटेशन थ्रू नॉलेज इंटीग्रेशन](https://arxiv.org/abs/1904.09223) यू सन, शुओहुआन वांग, युकुन ली, शिकुन फेंग, ज़ुई चेन, हान झांग, शिन तियान, डैनक्सियांग झू, हाओ तियान, हुआ वू द्वारा पोस्ट किया गया।
 1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (Baidu से) Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang. द्वाराअनुसंधान पत्र [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) के साथ जारी किया गया
@@ -353,7 +354,7 @@ conda install -c huggingface transformers
 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)** (Apple से) साथ में कागज [MobileViT: लाइट-वेट, जनरल-पर्पस, और मोबाइल-फ्रेंडली विजन ट्रांसफॉर्मर] (https://arxiv.org/abs/2110.02178) सचिन मेहता और मोहम्मद रस्तगरी द्वारा पोस्ट किया गया।
-1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (Apple से) Sachin Mehta and Mohammad Rastegari. द्वाराअनुसंधान पत्र [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) के साथ जारी किया गया
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (Apple से) Sachin Mehta and Mohammad Rastegari. द्वाराअनुसंधान पत्र [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) के साथ जारी किया गया
 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. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI से) साथ वाला पेपर [mT5: एक व्यापक बहुभाषी पूर्व-प्रशिक्षित टेक्स्ट-टू-टेक्स्ट ट्रांसफॉर्मर]( https://arxiv.org/abs/2010.11934) लिंटिंग ज़ू, नोआ कॉन्सटेंट, एडम रॉबर्ट्स, मिहिर काले, रामी अल-रफू, आदित्य सिद्धांत, आदित्य बरुआ, कॉलिन रैफेल द्वारा पोस्ट किया गया।
 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.
@@ -395,7 +396,7 @@ conda install -c huggingface transformers
 1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (फेसबुक से) साथ में पेपर [लार्ज-स्केल सेल्फ- एंड सेमी-सुपरवाइज्ड लर्निंग फॉर स्पीच ट्रांसलेशन](https://arxiv.org/abs/2104.06678) चांगहान वांग, ऐनी वू, जुआन पिनो, एलेक्सी बेवस्की, माइकल औली, एलेक्सिस द्वारा Conneau द्वारा पोस्ट किया गया।
 1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (तेल अवीव यूनिवर्सिटी से) साथ में पेपर [स्पैन सिलेक्शन को प्री-ट्रेनिंग करके कुछ-शॉट क्वेश्चन आंसरिंग](https:// arxiv.org/abs/2101.00438) ओरि राम, युवल कर्स्टन, जोनाथन बेरेंट, अमीर ग्लोबर्सन, ओमर लेवी द्वारा।
 1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (बर्कले से) कागज के साथ [SqueezeBERT: कुशल तंत्रिका नेटवर्क के बारे में NLP को कंप्यूटर विज़न क्या सिखा सकता है?](https: //arxiv.org/abs/2006.11316) फॉरेस्ट एन. इनडोला, अल्बर्ट ई. शॉ, रवि कृष्णा, और कर्ट डब्ल्यू. केटज़र द्वारा।
-1. **[SwiftFormer](https://huggingface.co/docs/transformers/main/model_doc/swiftformer)** (MBZUAI से) Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan. द्वाराअनुसंधान पत्र [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) के साथ जारी किया गया
+1. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (MBZUAI से) Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan. द्वाराअनुसंधान पत्र [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) के साथ जारी किया गया
 1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (माइक्रोसॉफ्ट से) साथ में कागज [स्वाइन ट्रांसफॉर्मर: शिफ्टेड विंडोज का उपयोग कर पदानुक्रमित विजन ट्रांसफॉर्मर](https://arxiv .org/abs/2103.14030) ज़ी लियू, युटोंग लिन, यू काओ, हान हू, यिक्सुआन वेई, झेंग झांग, स्टीफन लिन, बैनिंग गुओ द्वारा।
 1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (Microsoft से) साथ वाला पेपर [Swin Transformer V2: स्केलिंग अप कैपेसिटी एंड रेजोल्यूशन](https:// ज़ी लियू, हान हू, युटोंग लिन, ज़ुलिआंग याओ, ज़ेंडा ज़ी, यिक्सुआन वेई, जिया निंग, यू काओ, झेंग झांग, ली डोंग, फुरु वेई, बैनिंग गुओ द्वारा arxiv.org/abs/2111.09883।
 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.

README_ja.md

@@ -301,7 +301,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (Google Research から) Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig. から公開された研究論文 [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918)
 1. **[AltCLIP](https://huggingface.co/docs/transformers/model_doc/altclip)** (BAAI から) Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell から公開された研究論文: [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679)
 1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (MIT から) Yuan Gong, Yu-An Chung, James Glass から公開された研究論文: [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778)
-1. **[Autoformer](https://huggingface.co/docs/transformers/main/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. **[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. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (Facebook から) Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer から公開された研究論文: [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461)
 1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (École polytechnique から) Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis から公開された研究論文: [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321)
 1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (VinAI Research から) Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen から公開された研究論文: [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701)
@@ -355,6 +355,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[EfficientFormer](https://huggingface.co/docs/transformers/model_doc/efficientformer)** (Snap Research から) Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren. から公開された研究論文 [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191)
 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)** (Google Research/Stanford University から) Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning から公開された研究論文: [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555)
+1. **[EnCodec](https://huggingface.co/docs/transformers/main/model_doc/encodec)** (Meta AI から) Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi. から公開された研究論文 [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438)
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (Google Research から) Sascha Rothe, Shashi Narayan, Aliaksei Severyn から公開された研究論文: [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461)
 1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (Baidu から) Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu から公開された研究論文: [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223)
 1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (Baidu から) Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang. から公開された研究論文 [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674)
@@ -415,7 +416,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (Google Inc. から) Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam から公開された研究論文: [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861)
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (Google Inc. から) Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen から公開された研究論文: [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381)
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (Apple から) Sachin Mehta and Mohammad Rastegari から公開された研究論文: [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178)
-1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (Apple から) Sachin Mehta and Mohammad Rastegari. から公開された研究論文 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680)
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (Apple から) Sachin Mehta and Mohammad Rastegari. から公開された研究論文 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680)
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (Microsoft Research から) Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu から公開された研究論文: [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297)
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI から) Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel から公開された研究論文: [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934)
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (RUC AI Box から) Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen から公開された研究論文: [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131)
@@ -457,7 +458,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (Facebook から), Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau から公開された研究論文: [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678)
 1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (Tel Aviv University から), Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy から公開された研究論文: [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438)
 1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (Berkeley から) Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer から公開された研究論文: [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316)
-1. **[SwiftFormer](https://huggingface.co/docs/transformers/main/model_doc/swiftformer)** (MBZUAI から) Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan. から公開された研究論文 [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446)
+1. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (MBZUAI から) Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan. から公開された研究論文 [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446)
 1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (Microsoft から) Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo から公開された研究論文: [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030)
 1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (Microsoft から) Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo から公開された研究論文: [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883)
 1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (University of Würzburg から) Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte から公開された研究論文: [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345)

README_ko.md

@@ -216,7 +216,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (Google Research 에서 제공)은 Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.의 [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918)논문과 함께 발표했습니다.
 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/main/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. **[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. **[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/pdf/1910.13461.pdf) 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.
@@ -270,6 +270,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 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)** (Google Research/Stanford University 에서) Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning 의 [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) 논문과 함께 발표했습니다.
+1. **[EnCodec](https://huggingface.co/docs/transformers/main/model_doc/encodec)** (Meta AI 에서 제공)은 Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.의 [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438)논문과 함께 발표했습니다.
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (Google Research 에서) Sascha Rothe, Shashi Narayan, Aliaksei Severyn 의 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 논문과 함께 발표했습니다.
 1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (Baidu 에서) Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu 의 [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) 논문과 함께 발표했습니다.
 1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (Baidu 에서 제공)은 Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang.의 [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674)논문과 함께 발표했습니다.
@@ -330,7 +331,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (Google Inc. 에서) Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam 의 [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) 논문과 함께 발표했습니다.
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (Google Inc. 에서) Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen 의 [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) 논문과 함께 발표했습니다.
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (Apple 에서) Sachin Mehta and Mohammad Rastegari 의 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 논문과 함께 발표했습니다.
-1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (Apple 에서 제공)은 Sachin Mehta and Mohammad Rastegari.의 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680)논문과 함께 발표했습니다.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (Apple 에서 제공)은 Sachin Mehta and Mohammad Rastegari.의 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680)논문과 함께 발표했습니다.
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (Microsoft Research 에서) Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 의 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 논문과 함께 발표했습니다.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI 에서) Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 의 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 논문과 함께 발표했습니다.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (RUC AI Box 에서) Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 의 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 논문과 함께 발표했습니다.
@@ -372,7 +373,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (Facebook 에서) Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau 의 [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) 논문과 함께 발표했습니다.
 1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (Tel Aviv University 에서) Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy 의 [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) 논문과 함께 발표했습니다.
 1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (Berkeley 에서) Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer 의 [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) 논문과 함께 발표했습니다.
-1. **[SwiftFormer](https://huggingface.co/docs/transformers/main/model_doc/swiftformer)** (MBZUAI 에서 제공)은 Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.의 [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446)논문과 함께 발표했습니다.
+1. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (MBZUAI 에서 제공)은 Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.의 [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446)논문과 함께 발표했습니다.
 1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (Microsoft 에서) Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo 의 [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) 논문과 함께 발표했습니다.
 1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (Microsoft 에서) Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo 의 [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) 논문과 함께 발표했습니다.
 1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (University of Würzburg 에서) Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte 의 [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) 논문과 함께 발표했습니다.

README_zh-hans.md

@@ -240,7 +240,7 @@ conda install -c huggingface transformers
 1. **[ALIGN](https://huggingface.co/docs/transformers/model_doc/align)** (来自 Google Research) 伴随论文 [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://arxiv.org/abs/2102.05918) 由 Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig 发布。
 1. **[AltCLIP](https://huggingface.co/docs/transformers/model_doc/altclip)** (来自 BAAI) 伴随论文 [AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities](https://arxiv.org/abs/2211.06679) 由 Chen, Zhongzhi and Liu, Guang and Zhang, Bo-Wen and Ye, Fulong and Yang, Qinghong and Wu, Ledell 发布。
 1. **[Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer)** (来自 MIT) 伴随论文 [AST: Audio Spectrogram Transformer](https://arxiv.org/abs/2104.01778) 由 Yuan Gong, Yu-An Chung, James Glass 发布。
-1. **[Autoformer](https://huggingface.co/docs/transformers/main/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. **[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. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (来自 Facebook) 伴随论文 [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) 由 Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer 发布。
 1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (来自 École polytechnique) 伴随论文 [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) 由 Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis 发布。
 1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (来自 VinAI Research) 伴随论文 [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) 由 Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen 发布。
@@ -294,6 +294,7 @@ conda install -c huggingface transformers
 1. **[EfficientFormer](https://huggingface.co/docs/transformers/model_doc/efficientformer)** (来自 Snap Research) 伴随论文 [EfficientFormer: Vision Transformers at MobileNetSpeed](https://arxiv.org/abs/2206.01191) 由 Yanyu Li, Geng Yuan, Yang Wen, Ju Hu, Georgios Evangelidis, Sergey Tulyakov, Yanzhi Wang, Jian Ren 发布。
 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)** (来自 Google Research/Stanford University) 伴随论文 [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) 由 Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning 发布。
+1. **[EnCodec](https://huggingface.co/docs/transformers/main/model_doc/encodec)** (来自 Meta AI) 伴随论文 [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) 由 Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi 发布。
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (来自 Google Research) 伴随论文 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 由 Sascha Rothe, Shashi Narayan, Aliaksei Severyn 发布。
 1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (来自 Baidu) 伴随论文 [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 发布。
 1. **[ErnieM](https://huggingface.co/docs/transformers/model_doc/ernie_m)** (来自 Baidu) 伴随论文 [ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora](https://arxiv.org/abs/2012.15674) 由 Xuan Ouyang, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang 发布。
@@ -354,7 +355,7 @@ conda install -c huggingface transformers
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (来自 Google Inc.) 伴随论文 [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) 由 Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam 发布。
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (来自 Google Inc.) 伴随论文 [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) 由 Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen 发布。
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (来自 Apple) 伴随论文 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 由 Sachin Mehta and Mohammad Rastegari 发布。
-1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (来自 Apple) 伴随论文 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) 由 Sachin Mehta and Mohammad Rastegari 发布。
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/model_doc/mobilevitv2)** (来自 Apple) 伴随论文 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) 由 Sachin Mehta and Mohammad Rastegari 发布。
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (来自 Microsoft Research) 伴随论文 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 由 Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 发布。
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (来自 Google AI) 伴随论文 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 由 Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 发布。
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。
@@ -396,7 +397,7 @@ conda install -c huggingface transformers
 1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (来自 Facebook) 伴随论文 [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) 由 Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau 发布。
 1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (来自 Tel Aviv University) 伴随论文 [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) 由 Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy 发布。
 1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (来自 Berkeley) 伴随论文 [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) 由 Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer 发布。
-1. **[SwiftFormer](https://huggingface.co/docs/transformers/main/model_doc/swiftformer)** (来自 MBZUAI) 伴随论文 [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) 由 Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan 发布。
+1. **[SwiftFormer](https://huggingface.co/docs/transformers/model_doc/swiftformer)** (来自 MBZUAI) 伴随论文 [SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications](https://arxiv.org/abs/2303.15446) 由 Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan 发布。
 1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (来自 Microsoft) 伴随论文 [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) 由 Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo 发布。
 1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (来自 Microsoft) 伴随论文 [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) 由 Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo 发布。
 1. **[Swin2SR](https://huggingface.co/docs/transformers/model_doc/swin2sr)** (来自 University of Würzburg) 伴随论文 [Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration](https://arxiv.org/abs/2209.11345) 由 Marcos V. Conde, Ui-Jin Choi, Maxime Burchi, Radu Timofte 发布。
@@ -453,7 +454,7 @@ conda install -c huggingface transformers
 
 | 章节 | 描述 |
 |-|-|
-| [文档](https://huggingface.co/transformers/) | 完整的 API 文档和教程 |
+| [文档](https://huggingface.co/docs/transformers/) | 完整的 API 文档和教程 |
 | [任务总结](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers 支持的任务 |
 | [预处理教程](https://huggingface.co/docs/transformers/preprocessing) | 使用 `Tokenizer` 来为模型准备数据 |
 | [训练和微调](https://huggingface.co/docs/transformers/training) | 在 PyTorch/TensorFlow 的训练循环或 `Trainer` API 中使用 🤗 Transformers 提供的模型 |

README_zh-hant.md

@@ -252,7 +252,7 @@ conda install -c huggingface transformers
 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/main/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. **[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. **[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/pdf/1910.13461.pdf) 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.
@@ -306,6 +306,7 @@ conda install -c huggingface transformers
 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/main/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.
@@ -366,7 +367,7 @@ conda install -c huggingface transformers
 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/main/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. **[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. **[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. **[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.
@@ -408,7 +409,7 @@ conda install -c huggingface transformers
 1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook) released 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 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/main/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. **[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.

docs/source/en/_toctree.yml

@@ -541,6 +541,8 @@
         title: Audio Spectrogram Transformer
       - local: model_doc/clap
         title: CLAP
+      - local: model_doc/encodec
+        title: EnCodec
       - local: model_doc/hubert
         title: Hubert
       - local: model_doc/mctct

docs/source/en/index.mdx

@@ -107,6 +107,7 @@ The documentation is organized into five sections:
 1. **[EfficientFormer](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](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](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](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](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](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](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.
@@ -318,6 +319,7 @@ Flax), PyTorch, and/or TensorFlow.
 |        EfficientFormer        |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |         EfficientNet          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |            ELECTRA            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|            EnCodec            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |        Encoder decoder        |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
 |             ERNIE             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |            ErnieM             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/en/main_classes/agent.mdx

@@ -70,3 +70,32 @@ We provide three types of agents: [`HfAgent`] uses inference endpoints for opens
 ### launch_gradio_demo
 
 [[autodoc]] launch_gradio_demo
+
+## Agent Types
+
+Agents can handle any type of object in-between tools; tools, being completely multimodal, can accept and return
+text, image, audio, video, among other types. In order to increase compatibility between tools, as well as to 
+correctly render these returns in ipython (jupyter, colab, ipython notebooks, ...), we implement wrapper classes
+around these types.
+
+The wrapped objects should continue behaving as initially; a text object should still behave as a string, an image
+object should still behave as a `PIL.Image`.
+
+These types have three specific purposes:
+
+- Calling `to_raw` on the type should return the underlying object
+- Calling `to_string` on the type should return the object as a string: that can be the string in case of an `AgentText`
+  but will be the path of the serialized version of the object in other instances
+- Displaying it in an ipython kernel should display the object correctly
+
+### AgentText
+
+[[autodoc]] transformers.tools.agent_types.AgentText
+
+### AgentImage
+
+[[autodoc]] transformers.tools.agent_types.AgentImage
+
+### AgentAudio
+
+[[autodoc]] transformers.tools.agent_types.AgentAudio

docs/source/en/main_classes/deepspeed.mdx

@@ -760,7 +760,7 @@ time. "reuse distance" is a metric we are using to figure out when will a parame
 use the `stage3_max_reuse_distance` to decide whether to throw away the parameter or to keep it. If a parameter is
 going to be used again in near future (less than `stage3_max_reuse_distance`) then we keep it to reduce communication
 overhead. This is super helpful when you have activation checkpointing enabled, where we do a forward recompute and
-backward passes a a single layer granularity and want to keep the parameter in the forward recompute till the backward
+backward passes a single layer granularity and want to keep the parameter in the forward recompute till the backward
 
 The following configuration values depend on the model's hidden size:
 

docs/source/en/main_classes/trainer.mdx

@@ -656,7 +656,8 @@ Therefore, improving end-to-end performance.
 please follow this nice medium article [GPU-Acceleration Comes to PyTorch on M1 Macs](https://medium.com/towards-data-science/gpu-acceleration-comes-to-pytorch-on-m1-macs-195c399efcc1).
 
 **Usage**:
-User has to just pass `--use_mps_device` argument. 
+`mps` device will be used by default if available similar to the way `cuda` device is used.
+Therefore, no action from user is required. 
 For example, you can run the official Glue text classififcation task (from the root folder) using Apple Silicon GPU with below command:
 
 ```bash
@@ -672,7 +673,6 @@ python examples/pytorch/text-classification/run_glue.py \
   --learning_rate 2e-5 \
   --num_train_epochs 3 \
   --output_dir /tmp/$TASK_NAME/ \
-  --use_mps_device \
   --overwrite_output_dir
 ```
 
@@ -688,6 +688,156 @@ Finally, please, remember that, 🤗 `Trainer` only integrates MPS backend, ther
 have any problems or questions with regards to MPS backend usage, please, 
 file an issue with [PyTorch GitHub](https://github.com/pytorch/pytorch/issues).
 
+
+## Using Accelerate Launcher with Trainer
+
+Accelerate now powers Trainer. In terms of what users should expect:
+- They can keep using the Trainer ingterations such as FSDP, DeepSpeed vis trainer arguments without any changes on their part.
+- They can now use Accelerate Launcher with Trainer (recommended).
+
+Steps to use Accelerate Launcher with Trainer:
+1. Make sure 🤗 Accelerate is installed, you can't use the `Trainer` without it anyway. If not `pip install accelerate`. You may also need to update your version of Accelerate: `pip install accelerate --upgrade`
+2. Run `accelerate config` and fill the questionnaire. Below are example accelerate configs:
+  a. DDP Multi-node Multi-GPU config:
+    ```yaml
+    compute_environment: LOCAL_MACHINE                                                                                             
+    distributed_type: MULTI_GPU                                                                                                    
+    downcast_bf16: 'no'
+    gpu_ids: all
+    machine_rank: 0 #change rank as per the node
+    main_process_ip: 192.168.20.1
+    main_process_port: 9898
+    main_training_function: main
+    mixed_precision: fp16
+    num_machines: 2
+    num_processes: 8
+    rdzv_backend: static
+    same_network: true
+    tpu_env: []
+    tpu_use_cluster: false
+    tpu_use_sudo: false
+    use_cpu: false
+    ```
+
+  b. FSDP config:
+    ```yaml
+    compute_environment: LOCAL_MACHINE
+    distributed_type: FSDP
+    downcast_bf16: 'no'
+    fsdp_config:
+      fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+      fsdp_backward_prefetch_policy: BACKWARD_PRE
+      fsdp_forward_prefetch: true
+      fsdp_offload_params: false
+      fsdp_sharding_strategy: 1
+      fsdp_state_dict_type: FULL_STATE_DICT
+      fsdp_sync_module_states: true
+      fsdp_transformer_layer_cls_to_wrap: BertLayer
+      fsdp_use_orig_params: true
+    machine_rank: 0
+    main_training_function: main
+    mixed_precision: bf16
+    num_machines: 1
+    num_processes: 2
+    rdzv_backend: static
+    same_network: true
+    tpu_env: []
+    tpu_use_cluster: false
+    tpu_use_sudo: false
+    use_cpu: false
+    ```
+  c. DeepSpeed config pointing to a file:
+    ```yaml
+    compute_environment: LOCAL_MACHINE
+    deepspeed_config:
+      deepspeed_config_file: /home/user/configs/ds_zero3_config.json
+      zero3_init_flag: true
+    distributed_type: DEEPSPEED
+    downcast_bf16: 'no'
+    machine_rank: 0
+    main_training_function: main
+    num_machines: 1
+    num_processes: 4
+    rdzv_backend: static
+    same_network: true
+    tpu_env: []
+    tpu_use_cluster: false
+    tpu_use_sudo: false
+    use_cpu: false
+    ```
+
+  d. DeepSpeed config using accelerate plugin:
+    ```yaml
+    compute_environment: LOCAL_MACHINE                                                                                             
+    deepspeed_config:                                                                                                              
+      gradient_accumulation_steps: 1
+      gradient_clipping: 0.7
+      offload_optimizer_device: cpu
+      offload_param_device: cpu
+      zero3_init_flag: true
+      zero_stage: 2
+    distributed_type: DEEPSPEED
+    downcast_bf16: 'no'
+    machine_rank: 0
+    main_training_function: main
+    mixed_precision: bf16
+    num_machines: 1
+    num_processes: 4
+    rdzv_backend: static
+    same_network: true
+    tpu_env: []
+    tpu_use_cluster: false
+    tpu_use_sudo: false
+    use_cpu: false
+    ```
+
+3. Run the Trainer script with args other than the ones handled above by accelerate config or launcher args.
+Below is an example to run `run_glue.py` using `accelerate launcher` with FSDP config from above. 
+
+```bash
+cd transformers
+
+accelerate launch \
+./examples/pytorch/text-classification/run_glue.py \
+--model_name_or_path bert-base-cased \
+--task_name $TASK_NAME \
+--do_train \
+--do_eval \
+--max_seq_length 128 \
+--per_device_train_batch_size 16 \
+--learning_rate 5e-5 \
+--num_train_epochs 3 \
+--output_dir /tmp/$TASK_NAME/ \
+--overwrite_output_dir
+```
+
+4. You can also directly use the cmd args for `accelerate launch`. Above example would map to:
+
+```bash
+cd transformers
+
+accelerate launch --num_processes=2 \
+--use_fsdp \
+--mixed_precision=bf16 \
+--fsdp_auto_wrap_policy=TRANSFORMER_BASED_WRAP  \
+--fsdp_transformer_layer_cls_to_wrap="BertLayer" \
+--fsdp_sharding_strategy=1 \
+--fsdp_state_dict_type=FULL_STATE_DICT \
+./examples/pytorch/text-classification/run_glue.py
+--model_name_or_path bert-base-cased \
+--task_name $TASK_NAME \
+--do_train \
+--do_eval \
+--max_seq_length 128 \
+--per_device_train_batch_size 16 \
+--learning_rate 5e-5 \
+--num_train_epochs 3 \
+--output_dir /tmp/$TASK_NAME/ \
+--overwrite_output_dir
+```
+
+For more information, please refer the 🤗 Accelerate CLI guide: [Launching your 🤗 Accelerate scripts](https://huggingface.co/docs/accelerate/basic_tutorials/launch).
+
 Sections that were moved:
 
 [ <a href="./deepspeed#deepspeed-trainer-integration">DeepSpeed</a><a id="deepspeed"></a>

docs/source/en/model_doc/encodec.mdx

@@ -0,0 +1,59 @@
+<!--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.
+-->
+
+# EnCodec
+
+## Overview
+
+The EnCodec neural codec model was proposed in [High Fidelity Neural Audio Compression](https://arxiv.org/abs/2210.13438) by Alexandre Défossez, Jade Copet, Gabriel Synnaeve, Yossi Adi.
+
+The abstract from the paper is the following:
+
+*We introduce a state-of-the-art real-time, high-fidelity, audio codec leveraging neural networks. It consists in a streaming encoder-decoder architecture with quantized latent space trained in an end-to-end fashion. We simplify and speed-up the training by using a single multiscale spectrogram adversary that efficiently reduces artifacts and produce high-quality samples. We introduce a novel loss balancer mechanism to stabilize training: the weight of a loss now defines the fraction of the overall gradient it should represent, thus decoupling the choice of this hyper-parameter from the typical scale of the loss. Finally, we study how lightweight Transformer models can be used to further compress the obtained representation by up to 40%, while staying faster than real time. We provide a detailed description of the key design choices of the proposed model including: training objective, architectural changes and a study of various perceptual loss functions. We present an extensive subjective evaluation (MUSHRA tests) together with an ablation study for a range of bandwidths and audio domains, including speech, noisy-reverberant speech, and music. Our approach is superior to the baselines methods across all evaluated settings, considering both 24 kHz monophonic and 48 kHz stereophonic audio.*
+
+This model was contributed by [Matthijs](https://huggingface.co/Matthijs), [Patrick Von Platen](https://huggingface.co/patrickvonplaten) and [Arthur Zucker](https://huggingface.co/ArthurZ). 
+The original code can be found [here](https://github.com/facebookresearch/encodec).
+Here is a quick example of how to encode and decode an audio using this model:
+
+```python 
+>>> from datasets import load_dataset, Audio
+>>> from transformers import EncodecModel, AutoProcessor
+>>> librispeech_dummy = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
+
+>>> model = EncodecModel.from_pretrained("facebook/encodec_24khz")
+>>> processor = AutoProcessor.from_pretrained("facebook/encodec_24khz")
+>>> librispeech_dummy = librispeech_dummy.cast_column("audio", Audio(sampling_rate=processor.sampling_rate))
+>>> audio_sample = librispeech_dummy[-1]["audio"]["array"]
+>>> inputs = processor(raw_audio=audio_sample, sampling_rate=processor.sampling_rate, return_tensors="pt")
+
+>>> encoder_outputs = model.encode(inputs["input_values"], inputs["padding_mask"])
+>>> audio_values = model.decode(encoder_outputs.audio_codes, encoder_outputs.audio_scales, inputs["padding_mask"])[0]
+>>> # or the equivalent with a forward pass
+>>> audio_values = model(inputs["input_values"], inputs["padding_mask"]).audio_values
+```
+
+
+## EncodecConfig
+
+[[autodoc]] EncodecConfig
+
+## EncodecFeatureExtractor
+
+[[autodoc]] EncodecFeatureExtractor
+    - __call__
+
+## EncodecModel
+
+[[autodoc]] EncodecModel
+    - decode
+    - encode
+    - forward

docs/source/en/model_doc/llama.mdx

@@ -65,6 +65,7 @@ This model was contributed by [zphang](https://huggingface.co/zphang) with contr
     - build_inputs_with_special_tokens
     - get_special_tokens_mask
     - create_token_type_ids_from_sequences
+    - update_post_processor
     - save_vocabulary
 
 ## LlamaModel

docs/source/en/model_doc/mms.mdx

@@ -14,7 +14,7 @@ specific language governing permissions and limitations under the License.
 
 ## Overview
 
-The MMS model was proposed in [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2111.09296) 
+The MMS model was proposed in [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
 
 The abstract from the paper is the following:
@@ -44,11 +44,51 @@ MMS's architecture is based on the Wav2Vec2 model, so one can refer to [Wav2Vec2
 
 The original code can be found [here](https://github.com/facebookresearch/fairseq/tree/main/examples/mms).
 
+## Loading
+
+By default MMS loads adapter weights for English. If you want to load adapter weights of another language 
+make sure to specify `target_lang=<your-chosen-target-lang>` as well as `"ignore_mismatched_sizes=True`.
+The `ignore_mismatched_sizes=True` keyword has to be passed to allow the language model head to be resized according
+to the vocabulary of the specified language.
+Similarly, the processor should be loaded with the same target language
+
+```py
+from transformers import Wav2Vec2ForCTC, AutoProcessor
+
+model_id = "facebook/mms-1b-all"
+target_lang = "fra"
+
+processor = AutoProcessor.from_pretrained(model_id, target_lang=target_lang)
+model = Wav2Vec2ForCTC.from_pretrained(model_id, target_lang=target_lang, ignore_mismatched_sizes=True)
+```
+
+<Tip>
+
+You can safely ignore a warning such as:
+
+```text
+Some weights of Wav2Vec2ForCTC were not initialized from the model checkpoint at facebook/mms-1b-all and are newly initialized because the shapes did not match:
+- lm_head.bias: found shape torch.Size([154]) in the checkpoint and torch.Size([314]) in the model instantiated
+- lm_head.weight: found shape torch.Size([154, 1280]) in the checkpoint and torch.Size([314, 1280]) in the model instantiated
+You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
+```
+
+</Tip>
+
+If you want to use the ASR pipeline, you can load your chosen target language as such:
+
+```py
+from transformers import pipeline
+
+model_id = "facebook/mms-1b-all"
+target_lang = "fra"
+
+pipe = pipeline(model=model_id, model_kwargs={"target_lang": "fra", "ignore_mismatched_sizes": True})
+```
+
 ## Inference
 
-By default MMS loads adapter weights for English, but those can be easily switched out for another language.
-Let's look at an example.
-
+Next, let's look at how we can run MMS in inference and change adapter layers after having called [`~PretrainedModel.from_pretrained`]
 First, we load audio data in different languages using the [Datasets](https://github.com/huggingface/datasets).
 
 ```py

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

@@ -16,7 +16,7 @@ specific language governing permissions and limitations under the License.
 
 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 PLAM.
+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).

docs/source/en/tasks/image_classification.mdx

@@ -289,7 +289,7 @@ You're ready to start training your model now! Load ViT with [`AutoModelForImage
 
 At this point, only three steps remain:
 
-1. Define your training hyperparameters in [`TrainingArguments`]. It is important you don't remove unused columns because this'll drop the `image` column. Without the `image` column, you can't create `pixel_values`. Set `remove_unused_columns=False` to prevent this behavior! The only other required parameter is `output_dir` which specifies where to save your model. You'll push this model to the Hub by setting `push_to_hub=True` (you need to be signed in to Hugging Face to upload your model). At the end of each epoch, the [`Trainer`] will evaluate the accuracy and save the training checkpoint.
+1. Define your training hyperparameters in [`TrainingArguments`]. It is important you don't remove unused columns because that'll drop the `image` column. Without the `image` column, you can't create `pixel_values`. Set `remove_unused_columns=False` to prevent this behavior! The only other required parameter is `output_dir` which specifies where to save your model. You'll push this model to the Hub by setting `push_to_hub=True` (you need to be signed in to Hugging Face to upload your model). At the end of each epoch, the [`Trainer`] will evaluate the accuracy and save the training checkpoint.
 2. Pass the training arguments to [`Trainer`] along with the model, dataset, tokenizer, data collator, and `compute_metrics` function.
 3. Call [`~Trainer.train`] to finetune your model.
 
@@ -343,7 +343,7 @@ If you are unfamiliar with fine-tuning a model with Keras, check out the [basic
 
 To fine-tune a model in TensorFlow, follow these steps:
 1. Define the training hyperparameters, and set up an optimizer and a learning rate schedule.
-2. Instantiate a pre-treined model.
+2. Instantiate a pre-trained model.
 3. Convert a 🤗 Dataset to a `tf.data.Dataset`.
 4. Compile your model.
 5. Add callbacks and use the `fit()` method to run the training.

docs/source/en/tasks/question_answering.mdx

@@ -369,6 +369,7 @@ Tokenize the text and return PyTorch tensors:
 Pass your inputs to the model and return the `logits`:
 
 ```py
+>>> import torch
 >>> from transformers import AutoModelForQuestionAnswering
 
 >>> model = AutoModelForQuestionAnswering.from_pretrained("my_awesome_qa_model")

docs/source/en/tokenizer_summary.mdx

@@ -141,7 +141,7 @@ words. Pretokenization can be as simple as space tokenization, e.g. [GPT-2](mode
 [FlauBERT](model_doc/flaubert) which uses Moses for most languages, or [GPT](model_doc/gpt) which uses
 Spacy and ftfy, to count the frequency of each word in the training corpus.
 
-After pre-tokenization, a set of unique words has been created and the frequency of each word it occurred in the
+After pre-tokenization, a set of unique words has been created and the frequency with which each word occurred in the
 training data has been determined. Next, BPE creates a base vocabulary consisting of all symbols that occur in the set
 of unique words and learns merge rules to form a new symbol from two symbols of the base vocabulary. It does so until
 the vocabulary has attained the desired vocabulary size. Note that the desired vocabulary size is a hyperparameter to

docs/source/ko/_toctree.yml

@@ -156,16 +156,16 @@
     title: (번역중) Philosophy
   - local: in_translation
     title: (번역중) Glossary
-  - local: in_translation
-    title: (번역중) 🤗 What 🤗 Transformers can do
+  - local: task_summary
+    title: 🤗 Transformers로 할 수 있는 작업
   - local: tasks_explained
     title: 🤗 Transformers로 작업을 해결하는 방법
   - local: in_translation
     title: (번역중) The Transformer model family
   - local: in_translation
     title: (번역중) Summary of the tokenizers
-  - local: in_translation
-    title: (번역중) Attention mechanisms
+  - local: attention
+    title: 어텐션 매커니즘
   - local: pad_truncation
     title: 패딩과 잘라내기
   - local: bertology
@@ -673,4 +673,4 @@
     - local: in_translation
       title: (번역중) Utilities for Time Series
     title: (번역중) Internal Helpers
-  title: (번역중) API
+  title: (번역중) API

docs/source/ko/attention.mdx

@@ -0,0 +1,50 @@
+<!--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.
+-->
+
+# 어텐션 메커니즘[[attention_mechanisms]]
+
+대부분의 트랜스포머 모델은 정방행렬인 전체 어텐션을 사용합니다. 
+하지만 이는 긴 텍스트를 다룰 때는 큰 계산 병목 현상을 유발할 수 있습니다. 
+`Longformer`와 `Reformer`는 훈련 속도를 높이기 위해 어텐션 행렬의 희소 버전을 사용하여 효율을 높이려는 모델입니다.
+
+## LSH 어텐션[[lsh_attention]]
+
+
+[Reformer](#reformer)는 LSH(Locality Sensitive Hashing) 어텐션을 사용합니다. softmax(QK^t)에서는 행렬 QK^t의 (softmax 차원에서) 가장 큰 요소들만 유용한 기여를 할 것입니다. 
+따라서 각각의 쿼리 q에 대해, q와 가까운 키 k만 고려할 수 있습니다. 해시 함수는 q와 k가 가까운지 여부를 결정하는 데 사용됩니다. 
+어텐션 마스크는 현재 토큰을 마스킹하여 변경됩니다. 이 때 첫 번째 위치의 토큰은 제외합니다. 왜냐하면 쿼리와 키가 동일한 값을 갖게 되기 때문입니다(서로 매우 유사함). 
+해시는 약간의 무작위성을 가질 수 있으므로, 실제로는 여러 개의 해시 함수가 사용되고 (`n_rounds` 매개변수에 의해 결정됨) 그 후에 평균값을 취하게 됩니다.
+
+## 지역 어텐션[[local_attention]]
+
+[Longformer](#longformer)는 지역 어텐션을 사용합니다. 종종 특정 토큰에 대해 지역 컨텍스트(예: 왼쪽과 오른쪽에 있는 두 개의 토큰은 무엇인가요?)만으로도 작업을 수행하는데 충분합니다. 
+또한 작은 창(window)을 가진 어텐션 레이어를 쌓음으로써 마지막 레이어는 창 내의 토큰뿐만 아니라 더 많은 수의 토큰에 대한 수용 영역(receptive field)을 갖게 되어 전체 문장의 표현을 구축할 수 있습니다.
+
+사전에 선택된 일부 입력 토큰들은 전역 어텐션을 받습니다. 이 몇 개의 토큰에 대해서는 어텐션 행렬이 모든 토큰에 접근할 수 있으며, 이 과정은 대칭적으로 이루어집니다. 
+다른 모든 토큰들은 로컬 창 내의 토큰들에 더해 해당 특정 토큰들에도 접근할 수 있습니다. 이는 논문의 Figure 2d에서 나타나며, 아래에 샘플 어텐션 마스크가 제시되어 있습니다:
+
+
+<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>
+
+
+적은 파라미터의 어텐션 행렬을 사용하면 모델이 더 큰 시퀀스 입력 길이를 가질 수 있습니다.
+
+## 다른 방법들[[other_tricks]]
+
+### 축별 위치 인코딩[[axial_positional_encodings]]
+
+[Reformer](#reformer)는 축별 위치 인코딩(axial positional encodings)을 사용합니다. 기존의 트랜스포머 모델에서는 위치 인코딩 행렬 E는 크기가 \\(l \times d\\)인 행렬이며, 
+여기서 \\(l\\)은 시퀀스 길이(sequence length)이고 \\(d\\)는 숨겨진 상태(hidden state)의 차원입니다. 매우 긴 텍스트의 경우, 이 행렬은 매우 크며 GPU 상에서 공간을 많이 차지할 수 있습니다. 
+이를 완화하기 위해, 축별 위치 인코딩은 큰 행렬 E를 두 개의 작은 행렬 E1과 E2로 분해합니다. 이때 E1의 크기는 \\(l_{1} \times d_{1}\\)이고, E2의 크기는 \\(l_{2} \times d_{2}\\)입니다. 
+이때 \\(l_{1} \times l_{2} = l\\)이고 \\(d_{1} + d_{2} = d\\)(길이에 대한 곱셈 연산을 사용하면 훨씬 작아집니다). E의 시간 단계 j에 대한 임베딩은 E1에서 시간 단계 \\(j \% l1\\)의 임베딩과 E2에서 시간 단계  \\(j // l1\\)의 임베딩을 연결하여 얻습니다.

docs/source/ko/task_summary.mdx

@@ -0,0 +1,337 @@
+<!--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.
+-->
+
+# 🤗 Transformers로 할 수 있는 것[[what__transformers_can_do]]
+
+🤗 Transformers는 자연어처리(NLP), 컴퓨터 비전, 오디오 및 음성 처리 작업에 대한 사전훈련된 최첨단 모델 라이브러리입니다. 
+이 라이브러리는 트랜스포머 모델뿐만 아니라 컴퓨터 비전 작업을 위한 현대적인 합성곱 신경망과 같은 트랜스포머가 아닌 모델도 포함하고 있습니다. 
+
+스마트폰, 앱, 텔레비전과 같은 오늘날 가장 인기 있는 소비자 제품을 살펴보면, 딥러닝 기술이 그 뒤에 사용되고 있을 확률이 높습니다. 
+스마트폰으로 촬영한 사진에서 배경 객체를 제거하고 싶다면 어떻게 할까요? 이는 파놉틱 세그멘테이션 작업의 예입니다(아직 이게 무엇인지 모른다면, 다음 섹션에서 설명하겠습니다!).
+
+이 페이지는 다양한 음성 및 오디오, 컴퓨터 비전, NLP 작업을 🤗 Transformers 라이브러리를 활용하여 다루는 간단한 예제를 3줄의 코드로 제공합니다. 
+
+## 오디오[[audio]]
+
+
+음성 및 오디오 처리 작업은 다른 모달리티와 약간 다릅니다. 이는 주로 오디오가 연속적인 신호로 입력되기 때문입니다. 
+텍스트와 달리 원본 오디오 파형(waveform)은 문장이 단어로 나눠지는 것처럼 깔끔하게 이산적인 묶음으로 나눌 수 없습니다. 
+이를 극복하기 위해 원본 오디오 신호는 일정한 간격으로 샘플링됩니다. 해당 간격 내에서 더 많은 샘플을 취할 경우 샘플링률이 높아지며, 오디오는 원본 오디오 소스에 더 가까워집니다.
+
+과거의 접근 방식은 오디오에서 유용한 특징을 추출하기 위해 오디오를 전처리하는 것이었습니다. 
+하지만 현재는 원본 오디오 파형을 특성 인코더에 직접 넣어서 오디오 표현(representation)을 추출하는 것이 더 일반적입니다. 
+이렇게 하면 전처리 단계가 단순해지고 모델이 가장 중요한 특징을 학습할 수 있습니다.
+
+### 오디오 분류[[audio_classification]]
+
+
+오디오 분류는 오디오 데이터에 미리 정의된 클래스 집합의 레이블을 지정하는 작업입니다. 이는 많은 구체적인 응용 프로그램을 포함한 넓은 범주입니다.
+
+일부 예시는 다음과 같습니다:
+
+* 음향 장면 분류: 오디오에 장면 레이블("사무실", "해변", "경기장")을 지정합니다.
+* 음향 이벤트 감지: 오디오에 소리 이벤트 레이블("차 경적", "고래 울음소리", "유리 파손")을 지정합니다.
+* 태깅: 여러 가지 소리(새 지저귐, 회의에서의 화자 식별)가 포함된 오디오에 레이블을 지정합니다.
+* 음악 분류: 음악에 장르 레이블("메탈", "힙합", "컨트리")을 지정합니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> classifier = pipeline(task="audio-classification", model="superb/hubert-base-superb-er")
+>>> preds = classifier("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac")
+>>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
+>>> preds
+[{'score': 0.4532, 'label': 'hap'},
+ {'score': 0.3622, 'label': 'sad'},
+ {'score': 0.0943, 'label': 'neu'},
+ {'score': 0.0903, 'label': 'ang'}]
+```
+
+### 자동 음성 인식[[automatic_speech_recognition]]
+
+
+자동 음성 인식(ASR)은 음성을 텍스트로 변환하는 작업입니다. 
+음성은 인간의 자연스러운 의사소통 형태이기 때문에 ASR은 가장 일반적인 오디오 작업 중 하나입니다. 
+오늘날 ASR 시스템은 스피커, 전화 및 자동차와 같은 "스마트" 기술 제품에 내장되어 있습니다. 
+우리는 가상 비서에게 음악 재생, 알림 설정 및 날씨 정보를 요청할 수 있습니다.
+
+하지만 트랜스포머 아키텍처가 해결하는 데 도움을 준 핵심 도전 과제 중 하나는 양이 데이터 양이 적은 언어(low-resource language)에 대한 것입니다. 대량의 음성 데이터로 사전 훈련한 후 데이터 양이 적은 언어에서 레이블이 지정된 음성 데이터 1시간만으로 모델을 미세 조정하면 이전의 100배 많은 레이블이 지정된 데이터로 훈련된 ASR 시스템보다 훨씬 더 높은 품질의 결과를 얻을 수 있습니다. 
+```py
+>>> from transformers import pipeline
+
+>>> transcriber = pipeline(task="automatic-speech-recognition", model="openai/whisper-small")
+>>> 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.'}
+```
+
+## 컴퓨터 비전[[computer_vision]]
+
+컴퓨터 비전 작업 중 가장 초기의 성공적인 작업 중 하나는 [합성곱 신경망(CNN)](glossary#convolution)을 사용하여 우편번호 숫자 이미지를 인식하는 것이었습니다. 이미지는 픽셀로 구성되어 있으며 각 픽셀은 숫자 값으로 표현됩니다. 이로써 이미지를 픽셀 값의 행렬로 나타내는 것이 쉬워집니다. 특정한 픽셀 값의 조합은 이미지의 색상을 의미합니다.
+
+컴퓨터 비전 작업은 일반적으로 다음 두 가지 방법으로 접근 가능합니다:
+
+1. 합성곱을 사용하여 이미지의 낮은 수준 특징에서 높은 수준의 추상적인 요소까지 계층적으로 학습합니다.
+
+2. 이미지를 패치로 나누고 트랜스포머를 사용하여 점진적으로 각 이미지 패치가 서로 어떠한 방식으로 연관되어 이미지를 형성하는지 학습합니다. `CNN`에서 선호하는 상향식 접근법과는 달리, 이 방식은 흐릿한 이미지로 초안을 그리고 점진적으로 선명한 이미지로 만들어가는 것과 유사합니다.
+
+### 이미지 분류[[image_classification]]
+
+
+이미지 분류는 한 개의 전체 이미지에 미리 정의된 클래스 집합의 레이블을 지정하는 작업입니다. 
+
+대부분의 분류 작업과 마찬가지로, 이미지 분류에는 다양한 실용적인 용도가 있으며, 일부 예시는 다음과 같습니다:
+
+
+* 의료: 질병을 감지하거나 환자 건강을 모니터링하기 위해 의료 이미지에 레이블을 지정합니다.
+* 환경: 위성 이미지를 분류하여 산림 벌채를 감시하고 야생 지역 관리를 위한 정보를 제공하거나 산불을 감지합니다. 
+* 농업: 작물 이미지를 분류하여 식물 건강을 확인하거나 위성 이미지를 분류하여 토지 이용 관찰에 사용합니다.
+* 생태학: 동물이나 식물 종 이미지를 분류하여 야생 동물 개체군을 조사하거나 멸종 위기에 처한 종을 추적합니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> classifier = pipeline(task="image-classification")
+>>> preds = classifier(
+...     "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]
+>>> print(*preds, sep="\n")
+{'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'}
+```
+
+### 객체 탐지[[object_detection]]
+
+
+이미지 분류와 달리 객체 탐지는 이미지 내에서 여러 객체를 식별하고 바운딩 박스로 정의된 객체의 위치를 파악합니다. 
+
+객체 탐지의 몇 가지 응용 예시는 다음과 같습니다:
+
+* 자율 주행 차량: 다른 차량, 보행자 및 신호등과 같은 일상적인 교통 객체를 감지합니다.
+* 원격 감지: 재난 모니터링, 도시 계획 및 기상 예측 등을 수행합니다.
+* 결함 탐지: 건물의 균열이나 구조적 손상, 제조 결함 등을 탐지합니다.
+
+
+```py
+>>> from transformers import pipeline
+
+>>> detector = pipeline(task="object-detection")
+>>> preds = detector(
+...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"
+... )
+>>> preds = [{"score": round(pred["score"], 4), "label": pred["label"], "box": pred["box"]} for pred in preds]
+>>> preds
+[{'score': 0.9865,
+  'label': 'cat',
+  'box': {'xmin': 178, 'ymin': 154, 'xmax': 882, 'ymax': 598}}]
+```
+
+### 이미지 분할[[image_segmentation]]
+
+
+이미지 분할은 픽셀 차원의 작업으로, 이미지 내의 모든 픽셀을 클래스에 할당합니다. 이는 객체 탐지와 다릅니다. 객체 탐지는 바운딩 박스를 사용하여 이미지 내의 객체를 레이블링하고 예측하는 반면, 분할은 더 세분화된 작업입니다. 분할은 픽셀 수준에서 객체를 감지할 수 있습니다. 
+
+이미지 분할에는 여러 유형이 있습니다:
+
+* 인스턴스 분할: 개체의 클래스를 레이블링하는 것 외에도, 개체의 각 구분된 인스턴스에도 레이블을 지정합니다 ("개-1", "개-2" 등).
+* 파놉틱 분할: 의미적 분할과 인스턴스 분할의 조합입니다. 각 픽셀을 의미적 클래스로 레이블링하는 **동시에** 개체의 각각 구분된 인스턴스로도 레이블을 지정합니다.
+
+분할 작업은 자율 주행 차량에서 유용하며, 주변 환경의 픽셀 수준 지도를 생성하여 보행자와 다른 차량 주변에서 안전하게 탐색할 수 있습니다. 또한 의료 영상에서도 유용합니다. 분할 작업이 픽셀 수준에서 객체를 감지할 수 있기 때문에 비정상적인 세포나 장기의 특징을 식별하는 데 도움이 될 수 있습니다. 이미지 분할은 의류 가상 시착이나 카메라를 통해 실제 세계에 가상 개체를 덧씌워 증강 현실 경험을 만드는 등 전자 상거래 분야에서도 사용될 수 있습니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> segmenter = pipeline(task="image-segmentation")
+>>> preds = segmenter(
+...     "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]
+>>> print(*preds, sep="\n")
+{'score': 0.9879, 'label': 'LABEL_184'}
+{'score': 0.9973, 'label': 'snow'}
+{'score': 0.9972, 'label': 'cat'}
+```
+
+### 깊이 추정[[depth_estimation]]
+
+깊이 추정은 카메라로부터 이미지 내부의 각 픽셀의 거리를 예측합니다. 이 컴퓨터 비전 작업은 특히 장면 이해와 재구성에 중요합니다. 예를 들어, 자율 주행 차량은 보행자, 교통 표지판 및 다른 차량과 같은 객체와의 거리를 이해하여 장애물과 충돌을 피해야 합니다. 깊이 정보는 또한 2D 이미지에서 3D 표현을 구성하는 데 도움이 되며 생물학적 구조나 건물의 고품질 3D 표현을 생성하는 데 사용될 수 있습니다.
+
+깊이 추정에는 두 가지 접근 방식이 있습니다:
+
+* 스테레오: 약간 다른 각도에서 촬영된 동일한 이미지 두 장을 비교하여 깊이를 추정합니다.
+* 단안: 단일 이미지에서 깊이를 추정합니다.
+
+
+```py
+>>> from transformers import pipeline
+
+>>> depth_estimator = pipeline(task="depth-estimation")
+>>> preds = depth_estimator(
+...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"
+... )
+```
+
+## 자연어처리[[natural_language_processing]]
+
+텍스트는 인간이 의사 소통하는 자연스러운 방식 중 하나이기 때문에 자연어처리 역시 가장 일반적인 작업 유형 중 하나입니다. 모델이 인식하는 형식으로 텍스트를 변환하려면 토큰화해야 합니다. 이는 텍스트 시퀀스를 개별 단어 또는 하위 단어(토큰)로 분할한 다음 이러한 토큰을 숫자로 변환하는 것을 의미합니다. 결과적으로 텍스트 시퀀스를 숫자 시퀀스로 표현할 수 있으며, 숫자 시퀀스를 다양한 자연어처리 작업을 해결하기 위한 모델에 입력할 수 있습니다!
+
+### 텍스트 분류[[text_classification]]
+
+다른 모달리티에서의 분류 작업과 마찬가지로 텍스트 분류는 미리 정의된 클래스 집합에서 텍스트 시퀀스(문장 수준, 단락 또는 문서 등)에 레이블을 지정합니다. 텍스트 분류에는 다양한 실용적인 응용 사례가 있으며, 일부 예시는 다음과 같습니다:
+
+* 감성 분석: 텍스트를 `긍정` 또는 `부정`과 같은 어떤 극성에 따라 레이블링하여 정치, 금융, 마케팅과 같은 분야에서 의사 결정에 정보를 제공하고 지원할 수 있습니다.
+* 콘텐츠 분류: 텍스트를 주제에 따라 레이블링(날씨, 스포츠, 금융 등)하여 뉴스 및 소셜 미디어 피드에서 정보를 구성하고 필터링하는 데 도움이 될 수 있습니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> classifier = pipeline(task="sentiment-analysis")
+>>> preds = classifier("Hugging Face is the best thing since sliced bread!")
+>>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
+>>> preds
+[{'score': 0.9991, 'label': 'POSITIVE'}]
+```
+
+### 토큰 분류[[token_classification]]
+
+모든 자연어처리 작업에서는 텍스트가 개별 단어나 하위 단어로 분리되어 전처리됩니다. 분리된 단어를 [토큰](/glossary#token)이라고 합니다. 토큰 분류는 각 토큰에 미리 정의된 클래스 집합의 레이블을 할당합니다.
+
+토큰 분류의 두 가지 일반적인 유형은 다음과 같습니다:
+
+* 개체명 인식 (NER): 토큰을 조직, 인물, 위치 또는 날짜와 같은 개체 범주에 따라 레이블링합니다. NER은 특히 유전체학적인 환경에서 유전자, 단백질 및 약물 이름에 레이블을 지정하는 데 널리 사용됩니다.
+* 품사 태깅 (POS): 명사, 동사, 형용사와 같은 품사에 따라 토큰에 레이블을 할당합니다. POS는 번역 시스템이 동일한 단어가 문법적으로 어떻게 다른지 이해하는 데 도움이 됩니다 (명사로 사용되는 "bank(은행)"과 동사로 사용되는 "bank(예금을 예치하다)"과 같은 경우).
+
+
+```py
+>>> from transformers import pipeline
+
+>>> classifier = pipeline(task="ner")
+>>> preds = classifier("Hugging Face is a French company based in New York City.")
+>>> preds = [
+...     {
+...         "entity": pred["entity"],
+...         "score": round(pred["score"], 4),
+...         "index": pred["index"],
+...         "word": pred["word"],
+...         "start": pred["start"],
+...         "end": pred["end"],
+...     }
+...     for pred in preds
+... ]
+>>> print(*preds, sep="\n")
+{'entity': 'I-ORG', 'score': 0.9968, 'index': 1, 'word': 'Hu', 'start': 0, 'end': 2}
+{'entity': 'I-ORG', 'score': 0.9293, 'index': 2, 'word': '##gging', 'start': 2, 'end': 7}
+{'entity': 'I-ORG', 'score': 0.9763, 'index': 3, 'word': 'Face', 'start': 8, 'end': 12}
+{'entity': 'I-MISC', 'score': 0.9983, 'index': 6, 'word': 'French', 'start': 18, 'end': 24}
+{'entity': 'I-LOC', 'score': 0.999, 'index': 10, 'word': 'New', 'start': 42, 'end': 45}
+{'entity': 'I-LOC', 'score': 0.9987, 'index': 11, 'word': 'York', 'start': 46, 'end': 50}
+{'entity': 'I-LOC', 'score': 0.9992, 'index': 12, 'word': 'City', 'start': 51, 'end': 55}
+```
+
+### 질의응답[[question_answering]]
+
+질의응답은 또 하나의 토큰 차원의 작업으로, 문맥이 있을 때(개방형 도메인)와 문맥이 없을 때(폐쇄형 도메인) 질문에 대한 답변을 반환합니다. 이 작업은 가상 비서에게 식당이 영업 중인지와 같은 질문을 할 때마다 발생할 수 있습니다. 고객 지원 또는 기술 지원을 제공하거나 검색 엔진이 요청한 정보를 검색하는 데 도움을 줄 수 있습니다.
+
+질문 답변에는 일반적으로 두 가지 유형이 있습니다:
+
+* 추출형: 질문과 문맥이 주어졌을 때, 모델이 주어진 문맥의 일부에서 가져온 텍스트의 범위를 답변으로 합니다.
+* 생성형: 질문과 문맥이 주어졌을 때, 주어진 문맥을 통해 답변을 생성합니다. 이 접근 방식은 [`QuestionAnsweringPipeline`] 대신 [`Text2TextGenerationPipeline`]을 통해 처리됩니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> question_answerer = pipeline(task="question-answering")
+>>> preds = question_answerer(
+...     question="What is the name of the repository?",
+...     context="The name of the repository is huggingface/transformers",
+... )
+>>> print(
+...     f"score: {round(preds['score'], 4)}, start: {preds['start']}, end: {preds['end']}, answer: {preds['answer']}"
+... )
+score: 0.9327, start: 30, end: 54, answer: huggingface/transformers
+```
+
+### 요약[[summarization]]
+
+요약은 원본 문서의 의미를 최대한 보존하면서 긴 문서를 짧은 문서로 만드는 작업입니다. 요약은 `sequence-to-sequence` 작업입니다. 입력보다 짧은 텍스트 시퀀스를 출력합니다. 요약 작업은 독자가 장문 문서들의 주요 포인트를 빠르게 이해하는 데 도움을 줄 수 있습니다. 입법안, 법률 및 금융 문서, 특허 및 과학 논문은 요약 작업이 독자의 시간을 절약하고 독서 보조 도구로 사용될 수 있는 몇 가지 예시입니다.
+
+질문 답변과 마찬가지로 요약에는 두 가지 유형이 있습니다:
+
+* 추출형: 원본 텍스트에서 가장 중요한 문장을 식별하고 추출합니다.
+* 생성형: 원본 텍스트에서 목표 요약을 생성합니다. 입력 문서에 없는 새로운 단어를 포함할 수도 있습니다. [`SummarizationPipeline`]은 생성형 접근 방식을 사용합니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> summarizer = pipeline(task="summarization")
+>>> summarizer(
+...     "In this work, we presented the Transformer, the first sequence transduction model based entirely on attention, replacing the recurrent layers most commonly used in encoder-decoder architectures with multi-headed self-attention. For translation tasks, the Transformer can be trained significantly faster than architectures based on recurrent or convolutional layers. On both WMT 2014 English-to-German and WMT 2014 English-to-French translation tasks, we achieve a new state of the art. In the former task our best model outperforms even all previously reported ensembles."
+... )
+[{'summary_text': ' The Transformer is the first sequence transduction model based entirely on attention . It replaces the recurrent layers most commonly used in encoder-decoder architectures with multi-headed self-attention . For translation tasks, the Transformer can be trained significantly faster than architectures based on recurrent or convolutional layers .'}]
+```
+
+### 번역[[translation]]
+
+번역은 한 언어로 된 텍스트 시퀀스를 다른 언어로 변환하는 작업입니다. 이는 서로 다른 배경을 가진 사람들이 서로 소통하는 데 도움을 주는 중요한 역할을 합니다. 더 넓은 대중에게 콘텐츠를 번역하여 전달하거나, 새로운 언어를 배우는 데 도움이 되는 학습 도구가 될 수도 있습니다. 요약과 마찬가지로, 번역은 `sequence-to-sequence` 작업입니다. 즉, 모델은 입력 시퀀스를 받아서 출력이 되는 목표 시퀀스를 반환합니다.
+
+초기의 번역 모델은 대부분 단일 언어로 이루어져 있었지만, 최근에는 많은 언어 쌍 간에 번역을 수행할 수 있는 다중 언어 모델에 대한 관심이 높아지고 있습니다.
+
+```py
+>>> from transformers import pipeline
+
+>>> text = "translate English to French: Hugging Face is a community-based open-source platform for machine learning."
+>>> translator = pipeline(task="translation", model="t5-small")
+>>> translator(text)
+[{'translation_text': "Hugging Face est une tribune communautaire de l'apprentissage des machines."}]
+```
+
+### 언어 모델링[[language_modeling]]
+
+언어 모델링은 텍스트 시퀀스에서 단어를 예측하는 작업입니다. 사전 훈련된 언어 모델은 많은 다른 하위 작업에 따라 미세 조정될 수 있기 때문에 매우 인기 있는 자연어처리 작업이 되었습니다. 최근에는 제로 샷(zero-shot) 또는 퓨 샷(few-shot) 학습이 가능한 대규모 언어 모델(Large Language Models, LLM)에 대한 많은 관심이 발생하고 있습니다. 이는 모델이 명시적으로 훈련되지 않은 작업도 해결할 수 있다는 것을 의미합니다! 언어 모델은 유창하고 설득력 있는 텍스트를 생성하는 데 사용될 수 있지만, 텍스트가 항상 정확하지는 않을 수 있으므로 주의가 필요합니다.
+
+언어 모델링에는 두 가지 유형이 있습니다:
+
+* 인과적 언어 모델링: 이 모델의 목적은 시퀀스에서 다음 토큰을 예측하는 것이며, 미래 토큰이 마스킹 됩니다.
+    ```py
+    >>> from transformers import pipeline
+
+    >>> prompt = "Hugging Face is a community-based open-source platform for machine learning."
+    >>> generator = pipeline(task="text-generation")
+    >>> generator(prompt)  # doctest: +SKIP
+    ```
+
+* 마스킹된 언어 모델링: 이 모델의 목적은 시퀀스 내의 마스킹된 토큰을 예측하는 것이며, 시퀀스 내의 모든 토큰에 대한 접근이 제공됩니다.
+    
+    ```py
+    >>> text = "Hugging Face is a community-based open-source <mask> for machine learning."
+    >>> fill_mask = pipeline(task="fill-mask")
+    >>> preds = fill_mask(text, top_k=1)
+    >>> preds = [
+    ...     {
+    ...         "score": round(pred["score"], 4),
+    ...         "token": pred["token"],
+    ...         "token_str": pred["token_str"],
+    ...         "sequence": pred["sequence"],
+    ...     }
+    ...     for pred in preds
+    ... ]
+    >>> preds
+    [{'score': 0.2236,
+      'token': 1761,
+      'token_str': ' platform',
+      'sequence': 'Hugging Face is a community-based open-source platform for machine learning.'}]
+    ```
+
+이 페이지를 통해 각 모달리티의 다양한 작업 유형과 각 작업의 실용적 중요성에 대해 추가적인 배경 정보를 얻으셨기를 바랍니다. 다음 [섹션](tasks_explained)에서는 🤗 Transformer가 이러한 작업을 해결하는 **방법**에 대해 알아보실 수 있습니다.

examples/flax/question-answering/run_qa.py

@@ -61,7 +61,7 @@ from transformers.utils import check_min_version, get_full_repo_name, send_examp
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 Array = Any
 Dataset = datasets.arrow_dataset.Dataset

examples/flax/text-classification/run_flax_glue.py

@@ -54,7 +54,7 @@ from transformers.utils import check_min_version, get_full_repo_name, send_examp
 
 logger = logging.getLogger(__name__)
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 Array = Any
 Dataset = datasets.arrow_dataset.Dataset

examples/flax/token-classification/run_flax_ner.py

@@ -55,7 +55,7 @@ from transformers.utils.versions import require_version
 
 logger = logging.getLogger(__name__)
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/token-classification/requirements.txt")
 

examples/pytorch/audio-classification/run_audio_classification.py

@@ -45,7 +45,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.14.0", "To fix: pip install -r examples/pytorch/audio-classification/requirements.txt")
 

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

@@ -54,7 +54,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/contrastive-image-text/requirements.txt")
 

examples/pytorch/image-classification/run_image_classification.py

@@ -55,7 +55,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-classification/requirements.txt")
 

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -453,10 +453,11 @@ def main():
             resume_step = None
             completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_step
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/image-pretraining/run_mae.py

@@ -43,7 +43,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-pretraining/requirements.txt")
 

examples/pytorch/image-pretraining/run_mim.py

@@ -48,7 +48,7 @@ Any model supported by the AutoModelForMaskedImageModeling API can be used.
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-pretraining/requirements.txt")
 

examples/pytorch/image-pretraining/run_mim_no_trainer.py

@@ -53,7 +53,7 @@ Any model supported by the AutoModelForMaskedImageModeling API can be used.
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-pretraining/requirements.txt")
 
@@ -666,7 +666,7 @@ def main():
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_steps
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/language-modeling/run_clm.py

@@ -55,7 +55,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -572,7 +572,7 @@ def main():
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_steps
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/language-modeling/run_mlm.py

@@ -53,7 +53,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
@@ -616,7 +616,7 @@ def main():
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_steps
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/language-modeling/run_plm.py

@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
 

examples/pytorch/multiple-choice/run_swag.py

@@ -47,7 +47,7 @@ from transformers.utils import PaddingStrategy, check_min_version, send_example_
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -56,7 +56,7 @@ from transformers.utils import PaddingStrategy, check_min_version, get_full_repo
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 # You should update this to your particular problem to have better documentation of `model_type`
@@ -559,10 +559,11 @@ def main():
             resume_step = None
             completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/question-answering/run_qa.py

@@ -49,7 +49,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/question-answering/run_qa_beam_search.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -56,7 +56,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 
@@ -811,10 +811,11 @@ def main():
             resume_step = None
             completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 
@@ -830,7 +830,7 @@ def main():
             resume_step = int(training_difference.replace("step_", ""))
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/question-answering/run_seq2seq_qa.py

@@ -46,7 +46,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")
 

examples/pytorch/semantic-segmentation/run_semantic_segmentation.py

@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=2.0.0", "To fix: pip install -r examples/pytorch/semantic-segmentation/requirements.txt")
 

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -50,7 +50,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -556,10 +556,11 @@ def main():
             resume_step = None
             completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/speech-recognition/README.md

@@ -26,6 +26,10 @@ limitations under the License.
 		- [Librispeech](#librispeech-ctc)
 		- [Common Voice](#common-voice-ctc)
 		- [Multilingual Librispeech](#multilingual-librispeech-ctc)
+- [Automatic Speech Recognition with CTC and Adapter Layers](#connectionist-temporal-classification-with-adapters)
+	- [Massive Multilingual Speech (MMS)](#mms-model)
+	- [Examples](#examples-ctc-adapter)
+		- [Common Voice](#common-voice-ctc-adapter)
 - [Automatic Speech Recognition with Sequence-to-Sequence](#sequence-to-sequence)
 	- [Whisper Model](#whisper-model)
 	- [Speech-Encoder-Decoder Model](#warm-started-speech-encoder-decoder-model)
@@ -243,6 +247,111 @@ they can serve as a baseline to improve upon.
 | [Multilingual Librispeech](https://huggingface.co/datasets/multilingual_librispeech)| `"german"`  | [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53)  | 0.13  | -     | 1 GPU Titan 24 GB RAM  |  15h04                 | [here](https://huggingface.co/patrickvonplaten/wav2vec2-xlsr-53-300m-mls-german-ft)      |  [run.sh](https://huggingface.co/patrickvonplaten/wav2vec2-xlsr-53-300m-mls-german-ft/blob/main/run.sh) |
 | [Multilingual Librispeech](https://huggingface.co/datasets/multilingual_librispeech)| `"german"`  | [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m)  | 0.15 | -     | 1 GPU Titan 24 GB RAM  |  15h04                 | [here](https://huggingface.co/patrickvonplaten/wav2vec2-300m-mls-german-ft)      |  [run.sh](https://huggingface.co/patrickvonplaten/wav2vec2-300m-mls-german-ft/blob/main/run.sh) |
 
+## Connectionist Temporal Classification With Adapters
+
+The script [`run_speech_recognition_ctc_adapter.py`](https://github.com/huggingface/transformers/blob/main/examples/pytorch/speech-recognition/run_speech_recognition_ctc_adapter.py) can be used to fine-tune adapter layers for [Wav2Vec2-like models like MMS](https://huggingface.co/docs/transformers/main/en/model_doc/mms) for automatic speech recognition.
+
+### MMS Model
+
+The [Massive Multilingual Speech (MMS) model](https://huggingface.co/facebook/mms-1b-all) has been pre-trained and fine-tuned
+on 1000+ languages. The model makes use of adapter attention layers to fine-tune only a small part 
+of the model on a specific language. The model already comes with fine-tuned adapter layers for 1000+ languages and 
+can be used for inference for 1000+ languages out of the box.
+
+However, for improved performance or more specific use cases one can re-initialize the adapter weights, freeze all 
+other weights and fine-tune them on a specific dataset as shown in the [example below](#examples-ctc-adapter).
+
+Note that the adapter weights include low dimensional linear layers for every attention block as well as the final language
+model head layers.
+
+### Examples CTC Adapter
+
+In the following we will look at how one can fine-tune adapter weights for any of the 
+[MMS CTC checkpoints](https://huggingface.co/models?pipeline_tag=automatic-speech-recognition&other=mms&sort=downloads) in less than 1 hour.
+
+#### Common Voice CTC Adapter
+
+As in the examples [above](#examples-ctc), we fine-tune on Common Voice's 6 dataset in Turkish as an example.
+Contrary to [`run_speech_recognition_ctc.py`](https://github.com/huggingface/transformers/blob/main/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py) before there is a `--target_language` which has to be defined to state for which 
+language or concept the adapter layers shall be trained. The adapter weights will then 
+accordingly be called `adapter.{<target_language}.safetensors`.
+
+Let's run an example script. Make sure to be logged in so that your model can be directly uploaded to the Hub.
+```
+huggingface-cli login
+```
+
+Now, let's run an example and upload it to the Hub under `wav2vec2-common_voice-tr-mms-demo`.
+
+```sh
+python run_speech_recognition_ctc.py \
+	--dataset_name="common_voice" \
+	--model_name_or_path="facebook/mms-1b-all" \
+	--dataset_config_name="tr" \
+	--output_dir="./wav2vec2-common_voice-tr-mms-demo" \
+	--num_train_epochs="4" \
+	--per_device_train_batch_size="32" \
+	--learning_rate="1e-3" \
+	--warmup_steps="100" \
+	--evaluation_strategy="steps" \
+	--text_column_name="sentence" \
+	--length_column_name="input_length" \
+	--save_steps="200" \
+	--eval_steps="100" \
+	--save_total_limit="3" \
+  --target_language="tur" \
+	--gradient_checkpointing \
+	--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” <20> \
+	--fp16 \
+	--group_by_length \
+	--do_train --do_eval \
+  --push_to_hub
+```
+
+This should take less than 10 minutes on most GPUs and you should very quickly get word error rates 
+below 27%.
+
+For an example run, you can have a look at [`patrickvonplaten/wav2vec2-common_voice-tr-mms-demo`](https://huggingface.co/patrickvonplaten/wav2vec2-common_voice-tr-mms-demo).
+
+
+If you'd like to train another adapter model with the same base model, you can simply re-use the same `--output_dir`,
+but make sure to pass the `--output_dir` folder also to `--tokenizer_name_or_path` so that the vocabulary is not 
+overwritten but **extended**. Assuming you would like to train adapter weights on Swedish in addition to Turkish and save 
+the adapter weights in the same model repo, you can run:
+
+```sh
+python run_speech_recognition_ctc.py \
+	--dataset_name="common_voice" \
+	--model_name_or_path="facebook/mms-1b-all" \
+	--dataset_config_name="sw" \
+	--output_dir="./wav2vec2-common_voice-tr-mms-demo" \
+	--tokenizer_name_or_path="./wav2vec2-common_voice-tr-mms-demo" \
+	--num_train_epochs="4" \
+	--per_device_train_batch_size="32" \
+	--learning_rate="1e-3" \
+	--warmup_steps="100" \
+	--evaluation_strategy="steps" \
+	--text_column_name="sentence" \
+	--length_column_name="input_length" \
+	--save_steps="200" \
+	--eval_steps="100" \
+	--save_total_limit="3" \
+  --target_language="swe" \
+	--gradient_checkpointing \
+	--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” <20> \
+	--fp16 \
+	--group_by_length \
+	--do_train --do_eval \
+  --push_to_hub
+```
+
+Now you should have both `adapter.tur.safetensors` and `adapter.swe.safetensors` in the model repo
+and you can load the respective language with:
+```py
+model.load_adapter("tur")  # or "swe"
+```
+respectively.
+
 ## Sequence to Sequence
 
 The script [`run_speech_recognition_seq2seq.py`](https://github.com/huggingface/transformers/blob/main/examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py) can be used to fine-tune any [Speech Sequence-to-Sequence Model](https://huggingface.co/docs/transformers/main/en/model_doc/auto#transformers.AutoModelForSpeechSeq2Seq) for automatic speech 

examples/pytorch/speech-recognition/run_speech_recognition_ctc.py

@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
 

examples/pytorch/speech-recognition/run_speech_recognition_ctc_adapter.py

@@ -0,0 +1,799 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+""" Fine-tuning a 🤗 Transformers CTC adapter model for automatic speech recognition"""
+
+import functools
+import json
+import logging
+import os
+import re
+import sys
+import warnings
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional, Union
+
+import datasets
+import evaluate
+import numpy as np
+import torch
+from datasets import DatasetDict, load_dataset
+from safetensors.torch import save_file as safe_save_file
+
+import transformers
+from transformers import (
+    AutoConfig,
+    AutoFeatureExtractor,
+    AutoModelForCTC,
+    AutoProcessor,
+    AutoTokenizer,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    Wav2Vec2Processor,
+    set_seed,
+)
+from transformers.models.wav2vec2.modeling_wav2vec2 import WAV2VEC2_ADAPTER_SAFE_FILE
+from transformers.trainer_utils import get_last_checkpoint, is_main_process
+from transformers.utils import check_min_version, send_example_telemetry
+from transformers.utils.versions import require_version
+
+
+# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
+check_min_version("4.31.0.dev0")
+
+require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
+
+
+logger = logging.getLogger(__name__)
+
+
+def list_field(default=None, metadata=None):
+    return field(default_factory=lambda: default, metadata=metadata)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
+    )
+    tokenizer_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={"help": "Path to pretrained tokenizer or tokenizer identifier from huggingface.co/models"},
+    )
+    cache_dir: Optional[str] = field(
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
+    )
+    final_dropout: float = field(
+        default=0.0,
+        metadata={"help": "The dropout probability for the final projection layer."},
+    )
+    mask_time_prob: float = field(
+        default=0.05,
+        metadata={
+            "help": (
+                "Probability of each feature vector along the time axis to be chosen as the start of the vector"
+                "span to be masked. Approximately ``mask_time_prob * sequence_length // mask_time_length`` feature"
+                "vectors will be masked along the time axis."
+            )
+        },
+    )
+    mask_time_length: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the time axis."},
+    )
+    mask_feature_prob: float = field(
+        default=0.0,
+        metadata={
+            "help": (
+                "Probability of each feature vector along the feature axis to be chosen as the start of the vectorspan"
+                " to be masked. Approximately ``mask_feature_prob * sequence_length // mask_feature_length`` feature"
+                " bins will be masked along the time axis."
+            )
+        },
+    )
+    mask_feature_length: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the feature axis."},
+    )
+    layerdrop: float = field(default=0.0, metadata={"help": "The LayerDrop probability."})
+    ctc_loss_reduction: Optional[str] = field(
+        default="mean", metadata={"help": "The way the ctc loss should be reduced. Should be one of 'mean' or 'sum'."}
+    )
+    adapter_attn_dim: int = field(
+        default=16,
+        metadata={
+            "help": "The hidden dimension of the adapter layers that will be randomly initialized and trained. The higher the dimension, the more capacity is given to the adapter weights. Note that only the adapter weights are fine-tuned."
+        },
+    )
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+
+    Using `HfArgumentParser` we can turn this class
+    into argparse arguments to be able to specify them on
+    the command line.
+    """
+
+    dataset_name: str = field(
+        metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    target_language: Optional[str] = field(
+        metadata={
+            "help": (
+                "The target language on which the adapter attention layers"
+                " should be trained on in ISO 693-3 code, e.g. `tur` for Turkish"
+                " Wav2Vec2's MMS ISO codes can be looked up here: https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mms.html"
+                " If you are not training the adapter layers on a language, simply choose"
+                " another accronym that fits your data."
+            )
+        },
+    )
+    dataset_config_name: str = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    train_split_name: str = field(
+        default="train+validation",
+        metadata={
+            "help": (
+                "The name of the training data set split to use (via the datasets library). Defaults to "
+                "'train+validation'"
+            )
+        },
+    )
+    eval_split_name: str = field(
+        default="test",
+        metadata={
+            "help": "The name of the evaluation data set split to use (via the datasets library). Defaults to 'test'"
+        },
+    )
+    audio_column_name: str = field(
+        default="audio",
+        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
+    )
+    text_column_name: str = field(
+        default="text",
+        metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of training examples to this "
+                "value if set."
+            )
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of validation examples to this "
+                "value if set."
+            )
+        },
+    )
+    chars_to_ignore: Optional[List[str]] = list_field(
+        default=None,
+        metadata={"help": "A list of characters to remove from the transcripts."},
+    )
+    eval_metrics: List[str] = list_field(
+        default=["wer"],
+        metadata={"help": "A list of metrics the model should be evaluated on. E.g. `'wer cer'`"},
+    )
+    max_duration_in_seconds: float = field(
+        default=20.0,
+        metadata={
+            "help": (
+                "Filter audio files that are longer than `max_duration_in_seconds` seconds to"
+                " 'max_duration_in_seconds`"
+            )
+        },
+    )
+    min_duration_in_seconds: float = field(
+        default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
+    )
+    preprocessing_only: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "Whether to only do data preprocessing and skip training. This is especially useful when data"
+                " preprocessing errors out in distributed training due to timeout. In this case, one should run the"
+                " preprocessing in a non-distributed setup with `preprocessing_only=True` so that the cached datasets"
+                " can consequently be loaded in distributed training"
+            )
+        },
+    )
+    use_auth_token: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "If :obj:`True`, will use the token generated when running"
+                ":obj:`huggingface-cli login` as HTTP bearer authorization for remote files."
+            )
+        },
+    )
+    unk_token: str = field(
+        default="[UNK]",
+        metadata={"help": "The unk token for the tokenizer"},
+    )
+    pad_token: str = field(
+        default="[PAD]",
+        metadata={"help": "The padding token for the tokenizer"},
+    )
+    word_delimiter_token: str = field(
+        default="|",
+        metadata={"help": "The word delimiter token for the tokenizer"},
+    )
+    overwrite_lang_vocab: bool = field(
+        default=False,
+        metadata={"help": ("If :obj:`True`, will overwrite existing `target_language` vocabulary of tokenizer.")},
+    )
+
+
+@dataclass
+class DataCollatorCTCWithPadding:
+    """
+    Data collator that will dynamically pad the inputs received.
+    Args:
+        processor (:class:`~transformers.AutoProcessor`)
+            The processor used for proccessing the data.
+        padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
+            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
+            among:
+            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+              sequence if provided).
+            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
+              maximum acceptable input length for the model if that argument is not provided.
+            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
+              different lengths).
+        max_length (:obj:`int`, `optional`):
+            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
+        max_length_labels (:obj:`int`, `optional`):
+            Maximum length of the ``labels`` returned list and optionally padding length (see above).
+        pad_to_multiple_of (:obj:`int`, `optional`):
+            If set will pad the sequence to a multiple of the provided value.
+            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
+            7.5 (Volta).
+    """
+
+    processor: AutoProcessor
+    padding: Union[bool, str] = "longest"
+    pad_to_multiple_of: Optional[int] = None
+    pad_to_multiple_of_labels: Optional[int] = None
+
+    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
+        # split inputs and labels since they have to be of different lenghts and need
+        # different padding methods
+        input_features = [{"input_values": feature["input_values"]} for feature in features]
+        label_features = [{"input_ids": feature["labels"]} for feature in features]
+
+        batch = self.processor.pad(
+            input_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of,
+            return_tensors="pt",
+        )
+
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
+
+        # replace padding with -100 to ignore loss correctly
+        labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
+
+        batch["labels"] = labels
+        if "attention_mask" in batch:
+            batch["attention_mask"] = batch["attention_mask"].to(torch.long)
+
+        return batch
+
+
+def create_vocabulary_from_data(
+    datasets: DatasetDict,
+    word_delimiter_token: Optional[str] = None,
+    unk_token: Optional[str] = None,
+    pad_token: Optional[str] = None,
+):
+    # Given training and test labels create vocabulary
+    def extract_all_chars(batch):
+        all_text = " ".join(batch["target_text"])
+        vocab = list(set(all_text))
+        return {"vocab": [vocab], "all_text": [all_text]}
+
+    vocabs = datasets.map(
+        extract_all_chars,
+        batched=True,
+        batch_size=-1,
+        keep_in_memory=True,
+        remove_columns=datasets["train"].column_names,
+    )
+
+    # take union of all unique characters in each dataset
+    vocab_set = functools.reduce(
+        lambda vocab_1, vocab_2: set(vocab_1["vocab"][0]) | set(vocab_2["vocab"][0]), vocabs.values()
+    )
+
+    vocab_dict = {v: k for k, v in enumerate(sorted(vocab_set))}
+
+    # replace white space with delimiter token
+    if word_delimiter_token is not None:
+        vocab_dict[word_delimiter_token] = vocab_dict[" "]
+        del vocab_dict[" "]
+
+    # add unk and pad token
+    if unk_token is not None:
+        vocab_dict[unk_token] = len(vocab_dict)
+
+    if pad_token is not None:
+        vocab_dict[pad_token] = len(vocab_dict)
+
+    return vocab_dict
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_speech_recognition_ctc_adapter", model_args, data_args)
+
+    # Detecting last checkpoint.
+    last_checkpoint = None
+    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
+        last_checkpoint = get_last_checkpoint(training_args.output_dir)
+        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
+            raise ValueError(
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Use --overwrite_output_dir to overcome."
+            )
+        elif last_checkpoint is not None:
+            logger.info(
+                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
+                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
+            )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+    logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
+    logger.info("Training/evaluation parameters %s", training_args)
+
+    # Set seed before initializing model.
+    set_seed(training_args.seed)
+
+    # 1. First, let's load the dataset
+    raw_datasets = DatasetDict()
+
+    if training_args.do_train:
+        raw_datasets["train"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.train_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+
+        if data_args.audio_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--audio_column_name '{data_args.audio_column_name}' not found in dataset '{data_args.dataset_name}'."
+                " Make sure to set `--audio_column_name` to the correct audio column - one of"
+                f" {', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.text_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--text_column_name {data_args.text_column_name} not found in dataset '{data_args.dataset_name}'. "
+                "Make sure to set `--text_column_name` to the correct text column - one of "
+                f"{', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.max_train_samples is not None:
+            raw_datasets["train"] = raw_datasets["train"].select(range(data_args.max_train_samples))
+
+    if training_args.do_eval:
+        raw_datasets["eval"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.eval_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+
+        if data_args.max_eval_samples is not None:
+            raw_datasets["eval"] = raw_datasets["eval"].select(range(data_args.max_eval_samples))
+
+    # 2. We remove some special characters from the datasets
+    # that make training complicated and do not help in transcribing the speech
+    # E.g. characters, such as `,` and `.` do not really have an acoustic characteristic
+    # that could be easily picked up by the model
+    chars_to_ignore_regex = (
+        f'[{"".join(data_args.chars_to_ignore)}]' if data_args.chars_to_ignore is not None else None
+    )
+    text_column_name = data_args.text_column_name
+
+    def remove_special_characters(batch):
+        if chars_to_ignore_regex is not None:
+            batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
+        else:
+            batch["target_text"] = batch[text_column_name].lower() + " "
+        return batch
+
+    with training_args.main_process_first(desc="dataset map special characters removal"):
+        raw_datasets = raw_datasets.map(
+            remove_special_characters,
+            remove_columns=[text_column_name],
+            desc="remove special characters from datasets",
+        )
+
+    # save special tokens for tokenizer
+    word_delimiter_token = data_args.word_delimiter_token
+    unk_token = data_args.unk_token
+    pad_token = data_args.pad_token
+
+    # 3. Next, let's load the config as we might need it to create
+    # the tokenizer
+    # load config
+    config = AutoConfig.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # 4. Next, if no tokenizer file is defined,
+    # we create the vocabulary of the model by extracting all unique characters from
+    # the training and evaluation datasets
+    # We need to make sure that only first rank saves vocabulary
+    # make sure all processes wait until vocab is created
+    tokenizer_name_or_path = model_args.tokenizer_name_or_path
+    tokenizer_kwargs = {}
+
+    vocab_dict = {}
+    if tokenizer_name_or_path is not None:
+        # load vocabulary of other adapter languages so that new language can be appended
+        tokenizer = AutoTokenizer.from_pretrained(tokenizer_name_or_path, use_auth_token=data_args.use_auth_token)
+        vocab_dict = tokenizer.vocab.copy()
+        if tokenizer.target_lang is None:
+            raise ValueError("Make sure to load a multi-lingual tokenizer with a set target language.")
+
+        if data_args.target_language in tokenizer.vocab and not data_args.overwrite_lang_vocab:
+            logger.info(
+                "Adapter language already exists."
+                " Skipping vocabulary creating. If you want to create a new vocabulary"
+                f" for {data_args.target_language} make sure to add '--overwrite_lang_vocab'"
+            )
+        else:
+            tokenizer_name_or_path = None
+
+    if tokenizer_name_or_path is None:
+        # save vocab in training output dir
+        tokenizer_name_or_path = training_args.output_dir
+
+        vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
+
+        with training_args.main_process_first():
+            if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
+                try:
+                    os.remove(vocab_file)
+                except OSError:
+                    # in shared file-systems it might be the case that
+                    # two processes try to delete the vocab file at the some time
+                    pass
+
+        with training_args.main_process_first(desc="dataset map vocabulary creation"):
+            if not os.path.isfile(vocab_file):
+                os.makedirs(tokenizer_name_or_path, exist_ok=True)
+                lang_dict = create_vocabulary_from_data(
+                    raw_datasets,
+                    word_delimiter_token=word_delimiter_token,
+                    unk_token=unk_token,
+                    pad_token=pad_token,
+                )
+
+                # if we doing adapter language training, save
+                # vocab with adpter language
+                if data_args.target_language is not None:
+                    vocab_dict[data_args.target_language] = lang_dict
+
+                # save vocab dict to be loaded into tokenizer
+                with open(vocab_file, "w") as file:
+                    json.dump(vocab_dict, file)
+
+        # if tokenizer has just been created
+        # it is defined by `tokenizer_class` if present in config else by `model_type`
+        tokenizer_kwargs = {
+            "config": config if config.tokenizer_class is not None else None,
+            "tokenizer_type": config.model_type if config.tokenizer_class is None else None,
+            "unk_token": unk_token,
+            "pad_token": pad_token,
+            "word_delimiter_token": word_delimiter_token,
+            "target_lang": data_args.target_language,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        use_auth_token=data_args.use_auth_token,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # adapt config
+    config.update(
+        {
+            "final_dropout": model_args.final_dropout,
+            "mask_time_prob": model_args.mask_time_prob,
+            "mask_time_length": model_args.mask_time_length,
+            "mask_feature_prob": model_args.mask_feature_prob,
+            "mask_feature_length": model_args.mask_feature_length,
+            "gradient_checkpointing": training_args.gradient_checkpointing,
+            "layerdrop": model_args.layerdrop,
+            "ctc_loss_reduction": model_args.ctc_loss_reduction,
+            "pad_token_id": tokenizer.pad_token_id,
+            "vocab_size": len(tokenizer),
+            "adapter_attn_dim": model_args.adapter_attn_dim,
+        }
+    )
+
+    # create model
+    model = AutoModelForCTC.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        config=config,
+        use_auth_token=data_args.use_auth_token,
+        ignore_mismatched_sizes=True,
+    )
+
+    # if attn adapter is defined, freeze all non-adapter weights
+    if model.config.adapter_attn_dim is not None:
+        model.init_adapter_layers()
+        # first we freeze the whole base model
+        model.freeze_base_model()
+
+        # next we unfreeze all adapter layers
+        adapter_weights = model._get_adapters()
+        for param in adapter_weights.values():
+            param.requires_grad = True
+
+    # 6. Now we preprocess the datasets including loading the audio, resampling and normalization
+    # Thankfully, `datasets` takes care of automatically loading and resampling the audio,
+    # so that we just need to set the correct target sampling rate and normalize the input
+    # via the `feature_extractor`
+
+    # make sure that dataset decodes audio with correct sampling rate
+    dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
+    if dataset_sampling_rate != feature_extractor.sampling_rate:
+        raw_datasets = raw_datasets.cast_column(
+            data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        )
+
+    # derive max & min input length for sample rate & max duration
+    max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
+    min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
+    audio_column_name = data_args.audio_column_name
+    num_workers = data_args.preprocessing_num_workers
+
+    # Preprocessing the datasets.
+    # We need to read the audio files as arrays and tokenize the targets.
+    def prepare_dataset(batch):
+        # load audio
+        sample = batch[audio_column_name]
+
+        inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
+        batch["input_values"] = inputs.input_values[0]
+        batch["input_length"] = len(batch["input_values"])
+
+        # encode targets
+        batch["labels"] = tokenizer(batch["target_text"]).input_ids
+        return batch
+
+    with training_args.main_process_first(desc="dataset map preprocessing"):
+        vectorized_datasets = raw_datasets.map(
+            prepare_dataset,
+            remove_columns=next(iter(raw_datasets.values())).column_names,
+            num_proc=num_workers,
+            desc="preprocess datasets",
+        )
+
+        def is_audio_in_length_range(length):
+            return length > min_input_length and length < max_input_length
+
+        # filter data that is shorter than min_input_length
+        vectorized_datasets = vectorized_datasets.filter(
+            is_audio_in_length_range,
+            num_proc=num_workers,
+            input_columns=["input_length"],
+        )
+
+    # 7. Next, we can prepare the training.
+    # Let's use word error rate (WER) as our evaluation metric,
+    # instantiate a data collator and the trainer
+
+    # Define evaluation metrics during training, *i.e.* word error rate, character error rate
+    eval_metrics = {metric: evaluate.load(metric) for metric in data_args.eval_metrics}
+
+    # for large datasets it is advised to run the preprocessing on a
+    # single machine first with ``args.preprocessing_only`` since there will mostly likely
+    # be a timeout when running the script in distributed mode.
+    # In a second step ``args.preprocessing_only`` can then be set to `False` to load the
+    # cached dataset
+    if data_args.preprocessing_only:
+        logger.info(f"Data preprocessing finished. Files cached at {vectorized_datasets.cache_files}")
+        return
+
+    def compute_metrics(pred):
+        pred_logits = pred.predictions
+        pred_ids = np.argmax(pred_logits, axis=-1)
+
+        pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id
+
+        pred_str = tokenizer.batch_decode(pred_ids)
+        # we do not want to group tokens when computing the metrics
+        label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)
+
+        metrics = {k: v.compute(predictions=pred_str, references=label_str) for k, v in eval_metrics.items()}
+
+        return metrics
+
+    # Now save everything to be able to create a single processor later
+    # make sure all processes wait until data is saved
+    with training_args.main_process_first():
+        # only the main process saves them
+        if is_main_process(training_args.local_rank):
+            # save feature extractor, tokenizer and config
+            feature_extractor.save_pretrained(training_args.output_dir)
+            tokenizer.save_pretrained(training_args.output_dir)
+            config.save_pretrained(training_args.output_dir)
+
+    try:
+        processor = AutoProcessor.from_pretrained(training_args.output_dir)
+    except (OSError, KeyError):
+        warnings.warn(
+            "Loading a processor from a feature extractor config that does not"
+            " include a `processor_class` attribute is deprecated and will be removed in v5. Please add the following "
+            " attribute to your `preprocessor_config.json` file to suppress this warning: "
+            " `'processor_class': 'Wav2Vec2Processor'`",
+            FutureWarning,
+        )
+        processor = Wav2Vec2Processor.from_pretrained(training_args.output_dir)
+
+    # Instantiate custom data collator
+    data_collator = DataCollatorCTCWithPadding(processor=processor)
+
+    # Initialize Trainer
+    trainer = Trainer(
+        model=model,
+        data_collator=data_collator,
+        args=training_args,
+        compute_metrics=compute_metrics,
+        train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=vectorized_datasets["eval"] if training_args.do_eval else None,
+        tokenizer=processor,
+    )
+
+    # 8. Finally, we can start training
+
+    # Training
+    if training_args.do_train:
+        # use last checkpoint if exist
+        if last_checkpoint is not None:
+            checkpoint = last_checkpoint
+        elif os.path.isdir(model_args.model_name_or_path):
+            checkpoint = model_args.model_name_or_path
+        else:
+            checkpoint = None
+
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
+        trainer.save_model()
+
+        metrics = train_result.metrics
+        max_train_samples = (
+            data_args.max_train_samples
+            if data_args.max_train_samples is not None
+            else len(vectorized_datasets["train"])
+        )
+        metrics["train_samples"] = min(max_train_samples, len(vectorized_datasets["train"]))
+
+        trainer.log_metrics("train", metrics)
+        trainer.save_metrics("train", metrics)
+        trainer.save_state()
+
+    # Evaluation
+    results = {}
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+        metrics = trainer.evaluate()
+        max_eval_samples = (
+            data_args.max_eval_samples if data_args.max_eval_samples is not None else len(vectorized_datasets["eval"])
+        )
+        metrics["eval_samples"] = min(max_eval_samples, len(vectorized_datasets["eval"]))
+
+        trainer.log_metrics("eval", metrics)
+        trainer.save_metrics("eval", metrics)
+
+    # Write model card and (optionally) push to hub
+    config_name = data_args.dataset_config_name if data_args.dataset_config_name is not None else "na"
+    kwargs = {
+        "finetuned_from": model_args.model_name_or_path,
+        "tasks": "automatic-speech-recognition",
+        "tags": ["automatic-speech-recognition", data_args.dataset_name, "mms"],
+        "dataset_args": (
+            f"Config: {config_name}, Training split: {data_args.train_split_name}, Eval split:"
+            f" {data_args.eval_split_name}"
+        ),
+        "dataset": f"{data_args.dataset_name.upper()} - {config_name.upper()}",
+    }
+    if "common_voice" in data_args.dataset_name:
+        kwargs["language"] = config_name
+
+    # make sure that adapter weights are saved seperately
+    adapter_file = WAV2VEC2_ADAPTER_SAFE_FILE.format(data_args.target_language)
+    adapter_file = os.path.join(training_args.output_dir, adapter_file)
+    logger.info(f"Saving adapter weights under {adapter_file}...")
+    safe_save_file(model._get_adapters(), adapter_file, metadata={"format": "pt"})
+
+    if training_args.push_to_hub:
+        trainer.push_to_hub(**kwargs)
+    else:
+        trainer.create_model_card(**kwargs)
+
+    return results
+
+
+if __name__ == "__main__":
+    main()

examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
 

examples/pytorch/summarization/run_summarization.py

@@ -52,7 +52,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")
 

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -56,7 +56,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")
@@ -628,10 +628,11 @@ def main():
             resume_step = None
             completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

examples/pytorch/test_pytorch_examples.py

@@ -63,6 +63,7 @@ if SRC_DIRS is not None:
     import run_semantic_segmentation
     import run_seq2seq_qa as run_squad_seq2seq
     import run_speech_recognition_ctc
+    import run_speech_recognition_ctc_adapter
     import run_speech_recognition_seq2seq
     import run_summarization
     import run_swag
@@ -446,6 +447,38 @@ class ExamplesTests(TestCasePlus):
             result = get_results(tmp_dir)
             self.assertLess(result["eval_loss"], result["train_loss"])
 
+    def test_run_speech_recognition_ctc_adapter(self):
+        tmp_dir = self.get_auto_remove_tmp_dir()
+        testargs = f"""
+            run_speech_recognition_ctc_adapter.py
+            --output_dir {tmp_dir}
+            --model_name_or_path hf-internal-testing/tiny-random-wav2vec2
+            --dataset_name hf-internal-testing/librispeech_asr_dummy
+            --dataset_config_name clean
+            --train_split_name validation
+            --eval_split_name validation
+            --do_train
+            --do_eval
+            --learning_rate 1e-4
+            --per_device_train_batch_size 2
+            --per_device_eval_batch_size 1
+            --remove_unused_columns False
+            --overwrite_output_dir True
+            --preprocessing_num_workers 16
+            --max_steps 10
+            --target_language tur
+            --seed 42
+        """.split()
+
+        if is_cuda_and_apex_available():
+            testargs.append("--fp16")
+
+        with patch.object(sys, "argv", testargs):
+            run_speech_recognition_ctc_adapter.main()
+            result = get_results(tmp_dir)
+            self.assertTrue(os.path.isfile(os.path.join(tmp_dir, "./adapter.tur.safetensors")))
+            self.assertLess(result["eval_loss"], result["train_loss"])
+
     def test_run_speech_recognition_seq2seq(self):
         tmp_dir = self.get_auto_remove_tmp_dir()
         testargs = f"""

examples/pytorch/text-classification/run_glue.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
 

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -501,10 +501,16 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_step
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()

examples/pytorch/text-classification/run_xnli.py

@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
 

examples/pytorch/token-classification/run_ner.py

@@ -49,7 +49,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/token-classification/requirements.txt")
 

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -55,7 +55,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/token-classification/requirements.txt")
@@ -659,10 +659,16 @@ def main():
         if "epoch" in training_difference:
             starting_epoch = int(training_difference.replace("epoch_", "")) + 1
             resume_step = None
+            completed_steps = starting_epoch * num_update_steps_per_epoch
         else:
-            resume_step = int(training_difference.replace("step_", ""))
+            # need to multiply `gradient_accumulation_steps` to reflect real steps
+            resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
+            completed_steps = resume_step // args.gradient_accumulation_stepp
+
+    # update the progress_bar if load from checkpoint
+    progress_bar.update(completed_steps)
 
     for epoch in range(starting_epoch, args.num_train_epochs):
         model.train()

examples/pytorch/translation/run_translation.py

@@ -52,7 +52,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/translation/requirements.txt")
 

examples/pytorch/translation/run_translation_no_trainer.py

@@ -57,7 +57,7 @@ from transformers.utils.versions import require_version
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = get_logger(__name__)
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/translation/requirements.txt")
@@ -613,7 +613,7 @@ def main():
             resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
             starting_epoch = resume_step // len(train_dataloader)
             resume_step -= starting_epoch * len(train_dataloader)
-            completed_steps = resume_step
+            completed_steps = resume_step // args.gradient_accumulation_stepp
 
     # update the progress_bar if load from checkpoint
     progress_bar.update(completed_steps)

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

@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version(
     "datasets>=1.8.0", "To fix: pip install -r examples/tensorflow/contrastive-image-text/requirements.txt"

examples/tensorflow/image-classification/run_image_classification.py

@@ -54,7 +54,7 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/image-classification/requirements.txt")
 

examples/tensorflow/multiple-choice/run_swag.py

@@ -50,7 +50,7 @@ from transformers.utils import PaddingStrategy, check_min_version, send_example_
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/tensorflow/question-answering/run_qa.py

@@ -48,7 +48,7 @@ from transformers.utils import CONFIG_NAME, TF2_WEIGHTS_NAME, check_min_version,
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/tensorflow/summarization/run_summarization.py

@@ -53,7 +53,7 @@ from transformers.utils.versions import require_version
 
 # region Checking dependencies
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")
 

examples/tensorflow/text-classification/run_glue.py

@@ -47,7 +47,7 @@ from transformers.utils import check_min_version, send_example_telemetry
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 task_to_keys = {
     "cola": ("sentence", None),

examples/tensorflow/translation/run_translation.py

@@ -56,7 +56,7 @@ from transformers.utils.versions import require_version
 
 # region Dependencies and constants
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.30.0.dev0")
+check_min_version("4.31.0.dev0")
 
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")
 

setup.py

@@ -76,6 +76,7 @@ from pathlib import Path
 
 from setuptools import Command, find_packages, setup
 
+a = 3
 
 # Remove stale transformers.egg-info directory to avoid https://github.com/pypa/pip/issues/5466
 stale_egg_info = Path(__file__).parent / "transformers.egg-info"
@@ -98,7 +99,7 @@ if stale_egg_info.exists():
 # 2. once modified, run: `make deps_table_update` to update src/transformers/dependency_versions_table.py
 _deps = [
     "Pillow",
-    "accelerate>=0.20.1",
+    "accelerate>=0.20.3",
     "av==9.2.0",  # Latest version of PyAV (10.0.0) has issues with audio stream.
     "beautifulsoup4",
     "black~=23.1",
@@ -428,7 +429,7 @@ install_requires = [
 
 setup(
     name="transformers",
-    version="4.30.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="4.31.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     author="The Hugging Face team (past and future) with the help of all our contributors (https://github.com/huggingface/transformers/graphs/contributors)",
     author_email="transformers@huggingface.co",
     description="State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow",

src/transformers/__init__.py

@@ -18,7 +18,7 @@
 # to defer the actual importing for when the objects are requested. This way `import transformers` provides the names
 # in the namespace without actually importing anything (and especially none of the backends).
 
-__version__ = "4.30.0.dev0"
+__version__ = "4.31.0.dev0"
 
 from typing import TYPE_CHECKING
 
@@ -281,6 +281,10 @@ _import_structure = {
     "models.efficientformer": ["EFFICIENTFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "EfficientFormerConfig"],
     "models.efficientnet": ["EFFICIENTNET_PRETRAINED_CONFIG_ARCHIVE_MAP", "EfficientNetConfig"],
     "models.electra": ["ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP", "ElectraConfig", "ElectraTokenizer"],
+    "models.encodec": [
+        "ENCODEC_PRETRAINED_CONFIG_ARCHIVE_MAP",
+        "EncodecConfig",
+    ],
     "models.encoder_decoder": ["EncoderDecoderConfig"],
     "models.ernie": [
         "ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP",
@@ -825,6 +829,7 @@ except OptionalDependencyNotAvailable:
     ]
 else:
     _import_structure["models.audio_spectrogram_transformer"].append("ASTFeatureExtractor")
+    _import_structure["models.encodec"].append("EncodecFeatureExtractor")
     _import_structure["models.mctct"].append("MCTCTFeatureExtractor")
     _import_structure["models.speech_to_text"].append("Speech2TextFeatureExtractor")
     _import_structure["models.speecht5"].append("SpeechT5FeatureExtractor")
@@ -1568,6 +1573,13 @@ else:
             "load_tf_weights_in_electra",
         ]
     )
+    _import_structure["models.encodec"].extend(
+        [
+            "ENCODEC_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "EncodecModel",
+            "EncodecPreTrainedModel",
+        ]
+    )
     _import_structure["models.encoder_decoder"].append("EncoderDecoderModel")
     _import_structure["models.ernie"].extend(
         [
@@ -4100,6 +4112,10 @@ if TYPE_CHECKING:
     from .models.efficientformer import EFFICIENTFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, EfficientFormerConfig
     from .models.efficientnet import EFFICIENTNET_PRETRAINED_CONFIG_ARCHIVE_MAP, EfficientNetConfig
     from .models.electra import ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP, ElectraConfig, ElectraTokenizer
+    from .models.encodec import (
+        ENCODEC_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        EncodecConfig,
+    )
     from .models.encoder_decoder import EncoderDecoderConfig
     from .models.ernie import ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP, ErnieConfig
     from .models.ernie_m import ERNIE_M_PRETRAINED_CONFIG_ARCHIVE_MAP, ErnieMConfig
@@ -4598,6 +4614,7 @@ if TYPE_CHECKING:
         from .utils.dummy_speech_objects import *
     else:
         from .models.audio_spectrogram_transformer import ASTFeatureExtractor
+        from .models.encodec import EncodecFeatureExtractor
         from .models.mctct import MCTCTFeatureExtractor
         from .models.speech_to_text import Speech2TextFeatureExtractor
         from .models.speecht5 import SpeechT5FeatureExtractor
@@ -5210,6 +5227,11 @@ if TYPE_CHECKING:
             ElectraPreTrainedModel,
             load_tf_weights_in_electra,
         )
+        from .models.encodec import (
+            ENCODEC_PRETRAINED_MODEL_ARCHIVE_LIST,
+            EncodecModel,
+            EncodecPreTrainedModel,
+        )
         from .models.encoder_decoder import EncoderDecoderModel
         from .models.ernie import (
             ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST,

src/transformers/commands/download.py

@@ -18,7 +18,7 @@ from . import BaseTransformersCLICommand
 
 
 def download_command_factory(args):
-    return DownloadCommand(args.model, args.cache_dir, args.force)
+    return DownloadCommand(args.model, args.cache_dir, args.force, args.trust_remote_code)
 
 
 class DownloadCommand(BaseTransformersCLICommand):
@@ -31,16 +31,26 @@ class DownloadCommand(BaseTransformersCLICommand):
         download_parser.add_argument(
             "--force", action="store_true", help="Force the model to be download even if already in cache-dir"
         )
+        download_parser.add_argument(
+            "--trust-remote-code",
+            action="store_true",
+            help="Whether or not to allow for custom models defined on the Hub in their own modeling files. Use only if you've reviewed the code as it will execute on your local machine",
+        )
         download_parser.add_argument("model", type=str, help="Name of the model to download")
         download_parser.set_defaults(func=download_command_factory)
 
-    def __init__(self, model: str, cache: str, force: bool):
+    def __init__(self, model: str, cache: str, force: bool, trust_remote_code: bool):
         self._model = model
         self._cache = cache
         self._force = force
+        self._trust_remote_code = trust_remote_code
 
     def run(self):
         from ..models.auto import AutoModel, AutoTokenizer
 
-        AutoModel.from_pretrained(self._model, cache_dir=self._cache, force_download=self._force)
-        AutoTokenizer.from_pretrained(self._model, cache_dir=self._cache, force_download=self._force)
+        AutoModel.from_pretrained(
+            self._model, cache_dir=self._cache, force_download=self._force, trust_remote_code=self._trust_remote_code
+        )
+        AutoTokenizer.from_pretrained(
+            self._model, cache_dir=self._cache, force_download=self._force, trust_remote_code=self._trust_remote_code
+        )

src/transformers/configuration_utils.py

@@ -784,6 +784,13 @@ class PretrainedConfig(PushToHubMixin):
             ):
                 serializable_config_dict[key] = value
 
+        if hasattr(self, "quantization_config"):
+            serializable_config_dict["quantization_config"] = (
+                self.quantization_config.to_dict()
+                if not isinstance(self.quantization_config, dict)
+                else self.quantization_config
+            )
+
         self.dict_torch_dtype_to_str(serializable_config_dict)
 
         return serializable_config_dict

src/transformers/convert_slow_tokenizer.py

@@ -54,12 +54,15 @@ class SentencePieceExtractor:
 
         # Merges
         merges = []
-        for piece_l in vocab.keys():
-            for piece_r in vocab.keys():
-                merge = f"{piece_l}{piece_r}"
-                piece_score = vocab_scores.get(merge, None)
-                if piece_score:
-                    merges += [(piece_l, piece_r, piece_score)]
+        for merge, piece_score in vocab_scores.items():
+            local = []
+            for index in range(1, len(merge)):
+                piece_l, piece_r = merge[:index], merge[index:]
+                if piece_l in vocab and piece_r in vocab:
+                    local.append((piece_l, piece_r, piece_score))
+            local = sorted(local, key=lambda x: (vocab[x[0]], vocab[x[1]]))
+            merges.extend(local)
+
         merges = sorted(merges, key=lambda val: val[2], reverse=reverse)
         merges = [(val[0], val[1]) for val in merges]
         return vocab, merges
@@ -1134,9 +1137,9 @@ class LlamaConverter(SpmConverter):
             )
             tokenizer.add_special_tokens(
                 [
-                    AddedToken("<unk>", normalized=True),
-                    AddedToken("<s>", normalized=True),
-                    AddedToken("</s>", normalized=True),
+                    AddedToken("<unk>", normalized=False),
+                    AddedToken("<s>", normalized=False),
+                    AddedToken("</s>", normalized=False),
                 ]
             )
         else:

src/transformers/dependency_versions_check.py

@@ -23,9 +23,21 @@ from .utils.versions import require_version, require_version_core
 # order specific notes:
 # - tqdm must be checked before tokenizers
 
-pkgs_to_check_at_runtime = "python tqdm regex requests packaging filelock numpy tokenizers".split()
-if sys.version_info < (3, 7):
-    pkgs_to_check_at_runtime.append("dataclasses")
+pkgs_to_check_at_runtime = [
+    "python",
+    "tqdm",
+    "regex",
+    "requests",
+    "packaging",
+    "filelock",
+    "numpy",
+    "tokenizers",
+    "huggingface-hub",
+    "safetensors",
+    "accelerate",
+    "pyyaml",
+]
+
 if sys.version_info < (3, 8):
     pkgs_to_check_at_runtime.append("importlib_metadata")
 
@@ -37,6 +49,14 @@ for pkg in pkgs_to_check_at_runtime:
 
             if not is_tokenizers_available():
                 continue  # not required, check version only if installed
+        elif pkg == "accelerate":
+            # must be loaded here, or else tqdm check may fail
+            from .utils import is_accelerate_available
+
+            # Maybe switch to is_torch_available in the future here so that Accelerate is hard dep of
+            # Transformers with PyTorch
+            if not is_accelerate_available():
+                continue  # not required, check version only if installed
 
         require_version_core(deps[pkg])
     else:

src/transformers/dependency_versions_table.py

@@ -3,7 +3,7 @@
 # 2. run `make deps_table_update``
 deps = {
     "Pillow": "Pillow",
-    "accelerate": "accelerate>=0.20.1",
+    "accelerate": "accelerate>=0.20.3",
     "av": "av==9.2.0",
     "beautifulsoup4": "beautifulsoup4",
     "black": "black~=23.1",

src/transformers/generation/beam_search.py

@@ -15,7 +15,7 @@
 
 from abc import ABC, abstractmethod
 from collections import UserDict
-from typing import List, Optional, Tuple, Union
+from typing import Dict, List, Optional, Tuple, Union
 
 import numpy as np
 import torch
@@ -211,7 +211,7 @@ class BeamSearchScorer(BeamScorer):
         pad_token_id: Optional[int] = None,
         eos_token_id: Optional[Union[int, List[int]]] = None,
         beam_indices: Optional[torch.LongTensor] = None,
-    ) -> Tuple[torch.Tensor]:
+    ) -> Dict[str, torch.Tensor]:
         cur_len = input_ids.shape[-1] + 1  # add up to the length which the next_scores is calculated on
         batch_size = len(self._beam_hyps)
         if not (batch_size == (input_ids.shape[0] // self.group_size)):

src/transformers/generation/configuration_utils.py

@@ -288,7 +288,8 @@ class GenerationConfig(PushToHubMixin):
 
         # Additional attributes without default values
         if not self._from_model_config:
-            # we don't want to copy values from the model config if we're initializing a `GenerationConfig` from a model's default configuration file
+            # we don't want to copy values from the model config if we're initializing a `GenerationConfig` from a
+            # model's default configuration file
             for key, value in kwargs.items():
                 try:
                     setattr(self, key, value)
@@ -569,9 +570,9 @@ class GenerationConfig(PushToHubMixin):
         if "_commit_hash" in kwargs and "_commit_hash" in config_dict:
             kwargs["_commit_hash"] = config_dict["_commit_hash"]
 
-        # remove all the arguments that are in the config_dict
-
-        config = cls(**config_dict, **kwargs)
+        # The line below allows model-specific config to be loaded as well through kwargs, with safety checks.
+        # See https://github.com/huggingface/transformers/pull/21269
+        config = cls(**{**config_dict, **kwargs})
         unused_kwargs = config.update(**kwargs)
 
         logger.info(f"Generate config {config}")

src/transformers/generation/flax_logits_process.py

@@ -129,6 +129,8 @@ class FlaxTopPLogitsWarper(FlaxLogitsWarper):
     def __init__(self, top_p: float, filter_value: float = -float("Inf"), min_tokens_to_keep: int = 1):
         if not isinstance(top_p, float) or (top_p < 0 or top_p > 1.0):
             raise ValueError(f"`top_p` has to be a float > 0 and < 1, but is {top_p}")
+        if not isinstance(min_tokens_to_keep, int) or (min_tokens_to_keep < 0):
+            raise ValueError(f"`min_tokens_to_keep` has to be a non-negative integer, but is {min_tokens_to_keep}")
 
         self.top_p = top_p
         self.filter_value = filter_value

src/transformers/generation/flax_utils.py

@@ -319,7 +319,7 @@ class FlaxGenerationMixin:
                         "You have modified the pretrained model configuration to control generation. This is a"
                         " deprecated strategy to control generation and will be removed soon, in a future version."
                         " Please use a generation configuration file (see"
-                        " https://huggingface.co/docs/transformers/main_classes/text_generation)"
+                        " https://huggingface.co/docs/transformers/main_classes/text_generation )"
                     )
                     self.generation_config = new_generation_config
             generation_config = self.generation_config

src/transformers/generation/logits_process.py

@@ -255,6 +255,8 @@ class TopPLogitsWarper(LogitsWarper):
         top_p = float(top_p)
         if top_p < 0 or top_p > 1.0:
             raise ValueError(f"`top_p` has to be a float > 0 and < 1, but is {top_p}")
+        if not isinstance(min_tokens_to_keep, int) or (min_tokens_to_keep < 0):
+            raise ValueError(f"`min_tokens_to_keep` has to be a non-negative integer, but is {min_tokens_to_keep}")
 
         self.top_p = top_p
         self.filter_value = filter_value
@@ -266,9 +268,8 @@ class TopPLogitsWarper(LogitsWarper):
 
         # Remove tokens with cumulative top_p above the threshold (token with 0 are kept)
         sorted_indices_to_remove = cumulative_probs <= (1 - self.top_p)
-        if self.min_tokens_to_keep > 1:
-            # Keep at least min_tokens_to_keep
-            sorted_indices_to_remove[..., -self.min_tokens_to_keep :] = 0
+        # Keep at least min_tokens_to_keep
+        sorted_indices_to_remove[..., -self.min_tokens_to_keep :] = 0
 
         # scatter sorted tensors to original indexing
         indices_to_remove = sorted_indices_to_remove.scatter(1, sorted_indices, sorted_indices_to_remove)

src/transformers/generation/streamers.py

@@ -81,7 +81,7 @@ class TextStreamer(BaseStreamer):
 
     def put(self, value):
         """
-        Recives tokens, decodes them, and prints them to stdout as soon as they form entire words.
+        Receives tokens, decodes them, and prints them to stdout as soon as they form entire words.
         """
         if len(value.shape) > 1 and value.shape[0] > 1:
             raise ValueError("TextStreamer only supports batch size 1")

src/transformers/generation/tf_logits_process.py

@@ -160,6 +160,8 @@ class TFTopPLogitsWarper(TFLogitsWarper):
     def __init__(self, top_p: float, filter_value: float = -float("Inf"), min_tokens_to_keep: int = 1):
         if not isinstance(top_p, float) or (top_p < 0 or top_p > 1.0):
             raise ValueError(f"`top_p` has to be a float > 0 and < 1, but is {top_p}")
+        if not isinstance(min_tokens_to_keep, int) or (min_tokens_to_keep < 0):
+            raise ValueError(f"`min_tokens_to_keep` has to be a non-negative integer, but is {min_tokens_to_keep}")
 
         self.top_p = top_p
         self.filter_value = filter_value

src/transformers/generation/tf_utils.py

@@ -746,7 +746,7 @@ class TFGenerationMixin:
                         "You have modified the pretrained model configuration to control generation. This is a"
                         " deprecated strategy to control generation and will be removed soon, in a future version."
                         " Please use a generation configuration file (see"
-                        " https://huggingface.co/docs/transformers/main_classes/text_generation)"
+                        " https://huggingface.co/docs/transformers/main_classes/text_generation )"
                     )
                     self.generation_config = new_generation_config
             generation_config = self.generation_config

src/transformers/generation/utils.py

@@ -1260,7 +1260,7 @@ class GenerationMixin:
                         "You have modified the pretrained model configuration to control generation. This is a"
                         " deprecated strategy to control generation and will be removed soon, in a future version."
                         " Please use a generation configuration file (see"
-                        " https://huggingface.co/docs/transformers/main_classes/text_generation)"
+                        " https://huggingface.co/docs/transformers/main_classes/text_generation )"
                     )
                     self.generation_config = new_generation_config
             generation_config = self.generation_config
@@ -4232,6 +4232,15 @@ class GenerationMixin:
 
         # other auxiliary variables
         max_len = stopping_criteria[0].max_length
+        assistant_kv_indexing = (
+            1
+            if "bloom" in assistant_model.__class__.__name__.lower()
+            or (
+                assistant_model.config.architectures is not None
+                and "bloom" in assistant_model.config.architectures[0].lower()
+            )
+            else 0
+        )
 
         this_peer_finished = False  # used by synced_gpus only
         while True:
@@ -4247,7 +4256,6 @@ class GenerationMixin:
 
             # Assistant: main logic start
             cur_len = input_ids.shape[-1]
-            assistant_kv_indexing = 0 if "bloom" not in assistant_model.__class__.__name__.lower() else 1
 
             #  1. Forecast next N tokens using the assistant model. This `for` block can be replaced with a
             # `.generate()` call if we decide to add `past_key_values` as a possible output of generate, as we
@@ -4322,6 +4330,7 @@ class GenerationMixin:
                         encoder_outputs=model_kwargs["encoder_outputs"],
                         output_attentions=output_attentions,
                         output_hidden_states=output_hidden_states,
+                        use_cache=True,
                     )
                 else:
                     outputs = self(
@@ -4330,6 +4339,7 @@ class GenerationMixin:
                         past_key_values=model_kwargs["past_key_values"],
                         output_attentions=output_attentions,
                         output_hidden_states=output_hidden_states,
+                        use_cache=True,
                     )
             else:
                 if self.config.is_encoder_decoder:
@@ -4338,12 +4348,14 @@ class GenerationMixin:
                         encoder_outputs=model_kwargs["encoder_outputs"],
                         output_attentions=output_attentions,
                         output_hidden_states=output_hidden_states,
+                        use_cache=True,
                     )
                 else:
                     outputs = self(
                         candidate_input_ids,
                         output_attentions=output_attentions,
                         output_hidden_states=output_hidden_states,
+                        use_cache=True,
                     )
 
             # 2.2. Process the new logits
@@ -4508,7 +4520,10 @@ def _crop_past_key_values(model, past_key_values, maximum_length):
                 )
             )
         past_key_values = tuple(new_past)
-    elif "bloom" in model.__class__.__name__.lower():  # bloom is special
+    # bloom is special
+    elif "bloom" in model.__class__.__name__.lower() or (
+        model.config.architectures is not None and "bloom" in model.config.architectures[0].lower()
+    ):
         for idx in range(len(past_key_values)):
             new_past.append(
                 (
@@ -4517,7 +4532,10 @@ def _crop_past_key_values(model, past_key_values, maximum_length):
                 )
             )
         past_key_values = tuple(new_past)
-    elif "gptbigcode" in model.__class__.__name__.lower():  # gptbigcode is too
+    # gptbigcode is too
+    elif "gptbigcode" in model.__class__.__name__.lower() or (
+        model.config.architectures is not None and "gptbigcode" in model.config.architectures[0].lower()
+    ):
         if model.config.multi_query:
             for idx in range(len(past_key_values)):
                 past_key_values[idx] = past_key_values[idx][:, :maximum_length, :]

src/transformers/modeling_flax_utils.py

@@ -468,13 +468,14 @@ class FlaxPreTrainedModel(PushToHubMixin, FlaxGenerationMixin):
         # the state dict is unflattened to the match the format of model.params
         return unflatten_dict(state_sharded_dict, sep="/")
 
-    def can_generate(self) -> bool:
+    @classmethod
+    def can_generate(cls) -> bool:
         """
         Returns whether this model can generate sequences with `.generate()`. Returns:
             `bool`: Whether this model can generate sequences with `.generate()`.
         """
         # Detects whether `prepare_inputs_for_generation` has been overwritten, which is a requirement for generation
-        if "GenerationMixin" in str(self.prepare_inputs_for_generation.__func__):
+        if "GenerationMixin" in str(cls.prepare_inputs_for_generation):
             return False
         return True
 

src/transformers/modeling_tf_utils.py

@@ -81,7 +81,7 @@ elif parse(tf.__version__).minor >= 11:
     from keras.engine.keras_tensor import KerasTensor
 else:
     from tensorflow.python.keras import backend as K
-    from tensorflow.python.keras.engine import call_context
+    from tensorflow.python.keras.engine.base_layer_utils import call_context
     from tensorflow.python.keras.engine.keras_tensor import KerasTensor
 
 
@@ -1156,8 +1156,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         if self.built or call_context().in_call:
             self.built = True
         else:
-            self(self.dummy_inputs, training=False)
             self.built = True
+            self(self.dummy_inputs, training=False)
 
     def __init__(self, config, *inputs, **kwargs):
         super().__init__(*inputs, **kwargs)
@@ -1171,12 +1171,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         self.config = config
         self.name_or_path = config.name_or_path
         self.generation_config = GenerationConfig.from_model_config(config) if self.can_generate() else None
-        if not hasattr(self, "serving"):  # Don't overwrite existing serving signatures
-            self.serving = tf.function(
-                self.eager_serving, input_signature=[self._prune_signature(self.input_signature)]
-            )
         # Set the serving spec quickly to ensure that Keras doesn't use the specific dummy input shapes as the spec
-        self._set_save_spec(self.serving.input_signature[0])
+        self._set_save_spec(self._prune_signature(self.input_signature))
 
     def get_config(self):
         return self.config.to_dict()
@@ -1226,15 +1222,31 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         head_mask = tf.cast(head_mask, tf.float32)  # switch to float if need + fp16 compatibility
         return head_mask
 
+    @tf.function
+    def serving(self, inputs):
+        """
+        Args:
+        Method used for serving the model. Does not have a specific signature, but will be specialized as concrete
+        functions when saving with `save_pretrained`.
+            inputs (`Dict[str, tf.Tensor]`):
+                The input of the saved model as a dictionary of tensors.
+        """
+        output = self.call(inputs)
+
+        return self.serving_output(output)
+
     def eager_serving(self, inputs):
         """
-        Method used for serving the model. Intended not to be compiled with a tf.function decorator so that we can use
-        it to generate multiple signatures later.
+        Method used for serving the model. This method is deprecated, and will be removed.
 
         Args:
             inputs (`Dict[str, tf.Tensor]`):
                 The input of the saved model as a dictionary of tensors.
         """
+        warnings.warn(
+            "The function `eager_serving` is deprecated and will be removed in version 4.32.0 of Transformers",
+            FutureWarning,
+        )
         output = self.call(inputs)
 
         return self.serving_output(output)
@@ -1316,7 +1328,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
                     pass  # Layers may not have the same dimensions
         return output
 
-    def can_generate(self) -> bool:
+    @classmethod
+    def can_generate(cls) -> bool:
         """
         Returns whether this model can generate sequences with `.generate()`.
 
@@ -1324,7 +1337,7 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
             `bool`: Whether this model can generate sequences with `.generate()`.
         """
         # Detects whether `prepare_inputs_for_generation` has been overwritten, which is a requirement for generation
-        if "GenerationMixin" in str(self.prepare_inputs_for_generation.__func__):
+        if "GenerationMixin" in str(cls.prepare_inputs_for_generation):
             return False
         return True
 
@@ -2409,17 +2422,19 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
             if getattr(self.config, "torch_dtype", None) is not None and not isinstance(self.config.torch_dtype, str):
                 self.config.torch_dtype = str(self.config.torch_dtype).split(".")[1]
             if signatures is None:
-                if any(spec.dtype == tf.int32 for spec in self.serving.input_signature[0].values()):
+                sig = self._prune_signature(self.input_signature)
+                serving_default = self.serving.get_concrete_function(sig)
+                if any(spec.dtype == tf.int32 for spec in sig.values()):
                     int64_spec = {
                         key: tf.TensorSpec(
                             shape=spec.shape, dtype=tf.int64 if spec.dtype == tf.int32 else spec.dtype, name=spec.name
                         )
-                        for key, spec in self.serving.input_signature[0].items()
+                        for key, spec in sig.items()
                     }
-                    int64_serving = tf.function(self.eager_serving, input_signature=[int64_spec])
-                    signatures = {"serving_default": self.serving, "int64_serving": int64_serving}
+                    int64_serving = self.serving.get_concrete_function(int64_spec)
+                    signatures = {"serving_default": serving_default, "int64_serving": int64_serving}
                 else:
-                    signatures = self.serving
+                    signatures = serving_default
             saved_model_dir = os.path.join(save_directory, "saved_model", str(version))
             self.save(saved_model_dir, include_optimizer=False, signatures=signatures)
             logger.info(f"Saved model created in {saved_model_dir}")

src/transformers/modeling_utils.py

@@ -82,27 +82,17 @@ XLA_USE_BF16 = os.environ.get("XLA_USE_BF16", "0").upper()
 XLA_DOWNCAST_BF16 = os.environ.get("XLA_DOWNCAST_BF16", "0").upper()
 
 if is_accelerate_available():
-    from accelerate import __version__ as accelerate_version
     from accelerate import dispatch_model, infer_auto_device_map, init_empty_weights
     from accelerate.utils import (
+        check_tied_parameters_on_same_device,
         find_tied_parameters,
+        get_balanced_memory,
         load_offloaded_weights,
         offload_weight,
         save_offload_index,
         set_module_tensor_to_device,
     )
 
-    if version.parse(accelerate_version) > version.parse("0.11.0"):
-        from accelerate.utils import get_balanced_memory
-    else:
-        get_balanced_memory = None
-    if version.parse(accelerate_version) > version.parse("0.19.0"):
-        from accelerate.utils import check_tied_parameters_on_same_device
-    else:
-        check_tied_parameters_on_same_device = None
-else:
-    find_tied_parameters = None
-
 if is_safetensors_available():
     from safetensors import safe_open
     from safetensors.torch import load_file as safe_load_file
@@ -1069,6 +1059,8 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
     # a list of `state_dict` keys to ignore when saving the model (useful for keys that aren't
     # trained, but which are either deterministic or tied variables)
     _keys_to_ignore_on_save = None
+    # a list of `state_dict` keys that are potentially tied to another key in the state_dict.
+    _tied_weights_keys = None
 
     is_parallelizable = False
     supports_gradient_checkpointing = False
@@ -1182,7 +1174,8 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         """
         return getattr(self, self.base_model_prefix, self)
 
-    def can_generate(self) -> bool:
+    @classmethod
+    def can_generate(cls) -> bool:
         """
         Returns whether this model can generate sequences with `.generate()`.
 
@@ -1190,7 +1183,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
             `bool`: Whether this model can generate sequences with `.generate()`.
         """
         # Detects whether `prepare_inputs_for_generation` has been overwritten, which is a requirement for generation
-        if "GenerationMixin" in str(self.prepare_inputs_for_generation.__func__):
+        if "GenerationMixin" in str(cls.prepare_inputs_for_generation):
             return False
         return True
 
@@ -1778,8 +1771,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
             # We're going to remove aliases before saving
             ptrs = collections.defaultdict(list)
             for name, tensor in state_dict.items():
-                ident = (tensor.data_ptr(), tensor.device, tensor.shape, tensor.stride())
-                ptrs[ident].append(name)
+                ptrs[id_tensor_storage(tensor)].append(name)
 
             # These are all the pointers of shared tensors.
             shared_ptrs = {ptr: names for ptr, names in ptrs.items() if len(names) > 1}
@@ -2445,6 +2437,11 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                         f" {pretrained_model_name_or_path} but there is a file for Flax weights. Use `from_flax=True`"
                         " to load this model from those weights."
                     )
+                elif use_safetensors:
+                    raise EnvironmentError(
+                        f"Error no file named {_add_variant(SAFE_WEIGHTS_NAME, variant)} found in directory"
+                        f" {pretrained_model_name_or_path}."
+                    )
                 else:
                     raise EnvironmentError(
                         f"Error no file named {_add_variant(WEIGHTS_NAME, variant)}, {TF2_WEIGHTS_NAME},"
@@ -2786,8 +2783,6 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                     "If passing a string for `device_map`, please choose 'auto', 'balanced', 'balanced_low_0' or "
                     "'sequential'."
                 )
-            elif device_map in ["balanced", "balanced_low_0"] and get_balanced_memory is None:
-                raise ValueError(f"`device_map={device_map}` requires a source install of Accelerate.")
 
             kwargs = {"no_split_module_classes": no_split_modules}
             if "special_dtypes" in inspect.signature(infer_auto_device_map).parameters:
@@ -2797,7 +2792,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                     "This model has some weights that should be kept in higher precision, you need to upgrade "
                     "`accelerate` to properly deal with them (`pip install --upgrade accelerate`)."
                 )
-            if device_map != "sequential" and get_balanced_memory is not None:
+            if device_map != "sequential":
                 max_memory = get_balanced_memory(
                     model,
                     dtype=target_dtype,
@@ -2832,8 +2827,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
             model.tie_weights()
             tied_params = find_tied_parameters(model)
             # check if we don't have tied param in different devices
-            if check_tied_parameters_on_same_device is not None:
-                check_tied_parameters_on_same_device(tied_params, device_map)
+            check_tied_parameters_on_same_device(tied_params, device_map)
 
         if from_tf:
             if resolved_archive_file.endswith(".index"):
@@ -3025,8 +3019,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         missing_keys = list(set(expected_keys) - set(loaded_keys))
         unexpected_keys = list(set(loaded_keys) - set(expected_keys))
 
-        if find_tied_parameters is not None:
-            model.tie_weights()
+        if is_accelerate_available():
             tied_params = find_tied_parameters(model)
         else:
             tied_params = []

src/transformers/models/__init__.py

@@ -72,6 +72,7 @@ from . import (
     efficientformer,
     efficientnet,
     electra,
+    encodec,
     encoder_decoder,
     ernie,
     ernie_m,

src/transformers/models/albert/modeling_albert.py

@@ -759,6 +759,7 @@ class AlbertModel(AlbertPreTrainedModel):
     ALBERT_START_DOCSTRING,
 )
 class AlbertForPreTraining(AlbertPreTrainedModel):
+    _tied_weights_keys = ["predictions.decoder.bias", "predictions.decoder.weight"]
     _keys_to_ignore_on_load_missing = [
         "predictions.decoder.weight",
         "predictions.decoder.bias",
@@ -912,6 +913,7 @@ class AlbertSOPHead(nn.Module):
 )
 class AlbertForMaskedLM(AlbertPreTrainedModel):
     _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _tied_weights_keys = ["predictions.decoder.bias", "predictions.decoder.weight"]
     _keys_to_ignore_on_load_missing = [
         "predictions.decoder.weight",
         "predictions.decoder.bias",

src/transformers/models/align/configuration_align.py

@@ -206,8 +206,6 @@ class AlignVisionConfig(PretrainedConfig):
             The epsilon used by the batch normalization layers.
         batch_norm_momentum (`float`, *optional*, defaults to 0.99):
             The momentum used by the batch normalization layers.
-        dropout_rate (`float`, *optional*, defaults to 0.5):
-            The dropout rate to be applied before final classifier layer.
         drop_connect_rate (`float`, *optional*, defaults to 0.2):
             The drop rate for skip connections.
 
@@ -249,7 +247,6 @@ class AlignVisionConfig(PretrainedConfig):
         initializer_range: float = 0.02,
         batch_norm_eps: float = 0.001,
         batch_norm_momentum: float = 0.99,
-        dropout_rate: float = 0.5,
         drop_connect_rate: float = 0.2,
         **kwargs,
     ):
@@ -274,7 +271,6 @@ class AlignVisionConfig(PretrainedConfig):
         self.initializer_range = initializer_range
         self.batch_norm_eps = batch_norm_eps
         self.batch_norm_momentum = batch_norm_momentum
-        self.dropout_rate = dropout_rate
         self.drop_connect_rate = drop_connect_rate
         self.num_hidden_layers = sum(num_block_repeats) * 4
 

src/transformers/models/auto/configuration_auto.py

@@ -80,6 +80,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("efficientformer", "EfficientFormerConfig"),
         ("efficientnet", "EfficientNetConfig"),
         ("electra", "ElectraConfig"),
+        ("encodec", "EncodecConfig"),
         ("encoder-decoder", "EncoderDecoderConfig"),
         ("ernie", "ErnieConfig"),
         ("ernie_m", "ErnieMConfig"),
@@ -273,6 +274,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("efficientformer", "EFFICIENTFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("efficientnet", "EFFICIENTNET_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("electra", "ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("encodec", "ENCODEC_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("ernie", "ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("ernie_m", "ERNIE_M_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("esm", "ESM_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -461,6 +463,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("efficientformer", "EfficientFormer"),
         ("efficientnet", "EfficientNet"),
         ("electra", "ELECTRA"),
+        ("encodec", "EnCodec"),
         ("encoder-decoder", "Encoder decoder"),
         ("ernie", "ERNIE"),
         ("ernie_m", "ErnieM"),

src/transformers/models/auto/feature_extraction_auto.py

@@ -54,6 +54,7 @@ FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
         ("dinat", "ViTFeatureExtractor"),
         ("donut-swin", "DonutFeatureExtractor"),
         ("dpt", "DPTFeatureExtractor"),
+        ("encodec", "EncodecFeatureExtractor"),
         ("flava", "FlavaFeatureExtractor"),
         ("glpn", "GLPNFeatureExtractor"),
         ("groupvit", "CLIPFeatureExtractor"),

src/transformers/models/auto/modeling_auto.py

@@ -79,6 +79,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("efficientformer", "EfficientFormerModel"),
         ("efficientnet", "EfficientNetModel"),
         ("electra", "ElectraModel"),
+        ("encodec", "EncodecModel"),
         ("ernie", "ErnieModel"),
         ("ernie_m", "ErnieMModel"),
         ("esm", "EsmModel"),

src/transformers/models/autoformer/modeling_autoformer.py

@@ -1495,7 +1495,7 @@ class AutoformerModel(AutoformerPreTrainedModel):
     def __init__(self, config: AutoformerConfig):
         super().__init__(config)
 
-        if config.scaling == "mean" or config.scaling:
+        if config.scaling == "mean" or config.scaling is True:
             self.scaler = AutoformerMeanScaler(dim=1, keepdim=True)
         elif config.scaling == "std":
             self.scaler = AutoformerStdScaler(dim=1, keepdim=True)

src/transformers/models/bart/modeling_bart.py

@@ -1166,6 +1166,7 @@ class BartDecoder(BartPretrainedModel):
 )
 class BartModel(BartPretrainedModel):
     _keys_to_ignore_on_load_missing = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
 
     def __init__(self, config: BartConfig):
         super().__init__(config)
@@ -1293,9 +1294,10 @@ class BartModel(BartPretrainedModel):
 )
 class BartForConditionalGeneration(BartPretrainedModel):
     base_model_prefix = "model"
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight", "lm_head.weight"]
     _keys_to_ignore_on_load_missing = [
-        r"final_logits_bias",
-        r"lm_head.weight",
+        "final_logits_bias",
+        "lm_head.weight",
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",
     ]
@@ -1472,6 +1474,7 @@ class BartForConditionalGeneration(BartPretrainedModel):
 )
 class BartForSequenceClassification(BartPretrainedModel):
     _keys_to_ignore_on_load_missing = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
 
     def __init__(self, config: BartConfig, **kwargs):
         super().__init__(config, **kwargs)
@@ -1602,6 +1605,7 @@ class BartForSequenceClassification(BartPretrainedModel):
 )
 class BartForQuestionAnswering(BartPretrainedModel):
     _keys_to_ignore_on_load_missing = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -1740,6 +1744,7 @@ class BartDecoderWrapper(BartPretrainedModel):
 )
 class BartForCausalLM(BartPretrainedModel):
     _keys_to_ignore_on_load_missing = ["lm_head.weight"]
+    _tied_weights_keys = ["lm_head.weight"]
 
     def __init__(self, config):
         config = copy.deepcopy(config)

src/transformers/models/bert/modeling_bert.py

@@ -1054,6 +1054,7 @@ class BertModel(BertPreTrainedModel):
 )
 class BertForPreTraining(BertPreTrainedModel):
     _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias", r"cls.predictions.decoder.weight"]
+    _tied_weights_keys = ["predictions.decoder.bias", "cls.predictions.decoder.weight"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -1161,6 +1162,7 @@ class BertForPreTraining(BertPreTrainedModel):
 class BertLMHeadModel(BertPreTrainedModel):
     _keys_to_ignore_on_load_unexpected = [r"pooler"]
     _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias", r"cls.predictions.decoder.weight"]
+    _tied_weights_keys = ["predictions.decoder.bias", "cls.predictions.decoder.weight"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -1301,6 +1303,7 @@ class BertLMHeadModel(BertPreTrainedModel):
 class BertForMaskedLM(BertPreTrainedModel):
     _keys_to_ignore_on_load_unexpected = [r"pooler"]
     _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias", r"cls.predictions.decoder.weight"]
+    _tied_weights_keys = ["predictions.decoder.bias", "cls.predictions.decoder.weight"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/bert_generation/modeling_bert_generation.py

@@ -861,6 +861,7 @@ class BertGenerationOnlyLMHead(nn.Module):
 )
 class BertGenerationDecoder(BertGenerationPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["lm_head.decoder.weight", "lm_head.decoder.bias", "embeddings.position_ids"]
+    _tied_weights_keys = ["lm_head.decoder.weight", "lm_head.decoder.bias"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/big_bird/modeling_big_bird.py

@@ -2262,6 +2262,7 @@ class BigBirdModel(BigBirdPreTrainedModel):
 
 class BigBirdForPreTraining(BigBirdPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["cls.predictions.decoder.weight", "cls.predictions.decoder.bias"]
+    _tied_weights_keys = ["cls.predictions.decoder.weight", "cls.predictions.decoder.bias"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -2368,6 +2369,7 @@ class BigBirdForPreTraining(BigBirdPreTrainedModel):
 @add_start_docstrings("""BigBird Model with a `language modeling` head on top.""", BIG_BIRD_START_DOCSTRING)
 class BigBirdForMaskedLM(BigBirdPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["cls.predictions.decoder.weight", "cls.predictions.decoder.bias"]
+    _tied_weights_keys = ["cls.predictions.decoder.weight", "cls.predictions.decoder.bias"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -2517,6 +2519,7 @@ class BigBirdForCausalLM(BigBirdPreTrainedModel):
         "cls.predictions.decoder.weight",
         "cls.predictions.decoder.bias",
     ]
+    _tied_weights_keys = ["cls.predictions.decoder.weight", "cls.predictions.decoder.bias"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -2354,6 +2354,7 @@ class BigBirdPegasusDecoder(BigBirdPegasusPreTrainedModel):
 )
 class BigBirdPegasusModel(BigBirdPegasusPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
 
     def __init__(self, config: BigBirdPegasusConfig):
         super().__init__(config)
@@ -2484,9 +2485,10 @@ class BigBirdPegasusModel(BigBirdPegasusPreTrainedModel):
 # Copied from transformers.models.bart.modeling_bart.BartForConditionalGeneration with Bart->BigBirdPegasus, BART->BIGBIRD_PEGASUS
 class BigBirdPegasusForConditionalGeneration(BigBirdPegasusPreTrainedModel):
     base_model_prefix = "model"
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight", "lm_head.weight"]
     _keys_to_ignore_on_load_missing = [
-        r"final_logits_bias",
-        r"lm_head.weight",
+        "final_logits_bias",
+        "lm_head.weight",
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",
     ]
@@ -2663,6 +2665,7 @@ class BigBirdPegasusForConditionalGeneration(BigBirdPegasusPreTrainedModel):
 )
 class BigBirdPegasusForSequenceClassification(BigBirdPegasusPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
 
     def __init__(self, config: BigBirdPegasusConfig, **kwargs):
         super().__init__(config, **kwargs)
@@ -2792,6 +2795,7 @@ class BigBirdPegasusForSequenceClassification(BigBirdPegasusPreTrainedModel):
 )
 class BigBirdPegasusForQuestionAnswering(BigBirdPegasusPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -2924,6 +2928,7 @@ class BigBirdPegasusDecoderWrapper(BigBirdPegasusPreTrainedModel):
 
 class BigBirdPegasusForCausalLM(BigBirdPegasusPreTrainedModel):
     _keys_to_ignore_on_load_missing = ["lm_head.weight"]
+    _tied_weights_keys = ["lm_head.weight"]
 
     def __init__(self, config):
         config = copy.deepcopy(config)