pylate

PyLate is a Python library for optimizing training, inference, and retrieval with state-of-the-art ColBERT (Contextualized Late Interaction) models.

This is your go-to library for leveraging ColBERT models: it's built on the `sentence-transformers` framework for familiarity and efficiency. PyLate streamlines multi-vector model operations, offering features like multi-GPU/FP16/BF16 training and knowledge distillation for peak performance. It integrates efficient indexes (FastPLAID, HNSW) to handle large-scale, multi-vector neighbor search, accelerating both research and real-world retrieval applications for late interaction models.

https://github.com/lightonai/pylate