HuggingFace Transformers

The industry standard Python library for downloading and fine-tuning 100,000+ pretrained models across NLP, computer vision, and audio.

Transformers provides a unified API for 100,000+ pretrained models including BERT, GPT-2, and Whisper. It enables seamless interoperability between PyTorch, TensorFlow, and JAX: allowing teams to switch frameworks without refactoring core logic. The library integrates with the Hugging Face Hub to manage 10,000+ datasets and 50,000+ demos (Spaces). Using the Pipeline abstraction, developers execute complex tasks like translation or summarization in under five lines of code. It remains the primary engine for modern AI research and production at scale.

https://github.com/huggingface/transformers