Transformers

Transformers is the Python library in which Hugging Face keeps the model definitions for state-of-the-art text, vision, audio, video and multimodal models, covering both inference and training. Because the definition lives in one place, a model supported here is usable across most training frameworks such as Axolotl, Unsloth, DeepSpeed, FSDP and PyTorch Lightning, inference engines like vLLM, SGLang and TGI, and neighbouring libraries including llama.cpp and mlx, which all read from the same definition. The Hugging Face Hub lists more than a million checkpoints built on it, installation is a pip install of transformers[torch] on Python 3.10 or newer with PyTorch 2.5 or newer, and the license is Apache 2.0.