AGC Paper Outperforms hpool with Attention-Learned Pooling

antoine_chaffin · x · 2026-07-07

Following their pooling paper, several methods have emerged to compress ColBERT models into fewer tokens. A recent Attention-Guided Clustering (AGC) paper demonstrates that learning pooling via attention yields better information retention than hpool.

Related event: hpool Enables Flexible and Adaptive Compression for Retrieval Models(5 posts)→

Original post →

More from Research

Research channel →