transformers 5.2.0

Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

Stars: 156589, Watchers: 156589, Forks: 32121, Open Issues: 2282

The huggingface/transformers repo was created 7 years ago and the last code push was 5 hours ago.
The project is extremely popular with a mindblowing 156589 github stars!

How to Install transformers

You can install transformers using pip

pip install transformers

or add it to a project with poetry

poetry add transformers

Package Details

Author

The Hugging Face team (past and future) with the help of all our contributors (https://github.com/huggingface/transformers/graphs/contributors)

License

Apache 2.0 License

Homepage

https://github.com/huggingface/transformers

PyPi:

https://pypi.org/project/transformers/

GitHub Repo:

https://github.com/huggingface/transformers

Classifiers

• Scientific/Engineering/Artificial Intelligence

Related Packages

Errors

A list of common transformers errors.

Code Examples

Here are some transformers code examples and snippets.

GitHub Issues

The transformers package has 2282 open issues on GitHub

• Add pytest-random-order for reproducible test randomization
• Fix Llama4 vision attention scaling for TP compatibility
• fix(config): Fix incorrect default vision/audio config initialization in Phi4MultimodalConfig
• Phi4MultimodalConfig incorrectly initializes default vision/audio configs when passed as None
• refactor: use BatchLinear in Qwen2Moe to enable PEFT/LoRA support
• Remove stale TODO comments in UDOP tied weights
• Fix: AutoVideoProcessor returns incorrect shape for batched tensor inputs (#43450)
• [SAM 3 Video] Sam3VisionEncoderOutput object has no attribute 'fpn_position_embeddings'
• [SAM3 Video] get_text_features returns CLIPTextModelOutput but code expects tensor
• Fix Apertus model loading (NotImplementedError: Cannot copy out of meta tensor; no data!)
• Introduce standardized BatchLinear module for MoE architectures to facilitate PEFT and Quantization
• cli: Improve serve missing dependency error
• Fix MusicgenForConditionalGeneration config_class inheritance
• argparser: Allow optional bool flags without values
• RagTokenizer: add encode and patch_token(_id) forwarding

See more issues on GitHub