next-plaid makes late-interaction exact rescoring nearly free with one int8 op, 6x speedup
lateinteraction · x · 2026-09-03
Developer chaochunh nearly eliminated the slow half of PLAID-style late-interaction search — exact rescoring — in next-plaid 1.7.0 using a single int8 operation. It works on existing indexes, runs CPU-only, and the benchmark shows the once-dominant "decompress" segment is gone, delivering a 6x speedup. The library underpins ColGREP, and the work drew praise from search community members.
Related event: next-plaid 1.7.0 Speeds Up Late-Interaction Rescoring 6x with int8(2 posts)→
More from Infra
- Goldman: US to Outspend China $806B to $110B on AI Infra in 2026, China Fights Back on Efficiency — FinanceYF5 · 2026-09-23
- Nokia Signs Four Cloud Data Deals in 90 Days, But Its AI-RAN Bet Rests on One Chipmaker — shashib · 2026-09-23
- China Weighs Curbs on Broadcom Switches Behind Up to 90% of State Data Centers — rohanpaul_ai · 2026-09-23
- tinygrad hits ~200 tok/s MiMo-V2.6-Pro on MI300X, brought up via GLM-5.3 — AIFlow_ML · 2026-09-23
- Rumors: 64GB+ VRAM RTX 5090 in R&D but not coming anytime soon — AIFlow_ML · 2026-09-23
- Swarm-built inference engine runs Qwen Image-2.1: 1K images in under 0.5s — bingxu_ · 2026-09-23