Preprint
Aug 2026
Chimera: Efficient Multi-Vector Retrieval via GPU-CPU Co-Processing
Chimera is proposed, a GPU-CPU co-processing system for multi-vector retrieval that eliminates this transfer bottleneck, and significantly outperforms existing approaches, achieving up to 59.5x higher QPS at the same recall level.
Yanqi Chen, Jue-Lin Liu, Alexandra Meliou et al.
· 0 citations