Ben Clavié extracts BM25-ready sparse vocabularies from frozen dense retrievers using Sparse Autoencoders
The technique requires no model fine-tuning or modification.
——0——
The technique requires no model fine-tuning or modification.
Users praised researchers for extracting BM25-ready sparse features from frozen dense retrievers via autoencoders, calling the work interesting and anticipating the full blog post, while one accused them of joining the trend late.
7 comments with sentiment.
