Easily use and train state of the art late-interaction retrieval methods (ColBERT) in any RAG pipeline. Designed for modularity and ease-of-use, backed by research.
Apache License 2.0
3.07k
stars
210
forks
source link
Use Matryoshka + Binary Quantisation Embeddings #260
Ragatouille supports using models for encoding trained on Matryoshka embeddings but I wasn't able to find a way to use them for indexing. Secondly although ragatouille supports storing compressed vectors to disk but it would be great to combine them with binary/int 8 quantised embeddings. Would love to contribute if it's feasible.
Ragatouille supports using models for encoding trained on Matryoshka embeddings but I wasn't able to find a way to use them for indexing. Secondly although ragatouille supports storing compressed vectors to disk but it would be great to combine them with binary/int 8 quantised embeddings. Would love to contribute if it's feasible.