Year: 2025
Type: preprint
Abstract: We study serving retrieval models, specifically late interaction models like ColBERT, to many concurrent users at once and under a small budget, in which the index may not fit in memory. We present ColBERT-serve, a novel serving system that applies a memory-mapping strategy to the ColBERT index, reducing RAM usage by 90% and permitting its deployment on cheap servers, and incorporates a multi-stage architecture with hybrid scoring, reducing ColBERT's query latency and supporting many concurrent queries in parallel.
Source: arXiv (Cornell University)
Authors: Huang, Kaili, Venkatesh, Thejas, Dingankar, Uma, Mallia, Antonio, Campos, Daniel, Jiao, Jian, Potts, Christopher, Zaharia, Matei, Boahen, Kwabena, Khattab, Omar, Sarup, Saarthak, Santhanam, Keshav
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6979330229
Cites:
Cited by:
Related to: