AVO agents evolve attention kernels beating FlashAttention-4 by 10.5% on B200
bingxu_ · x · 2026-09-25
AVO, accepted at NeurIPS 2026, replaces fixed mutation/crossover heuristics in evolutionary search with autonomous coding agents that propose, repair, critique and verify kernel edits. After 7 days of autonomous evolution on Blackwell B200, AVO discovers attention kernels up to 3.5% faster than cuDNN and 10.5% faster than FlashAttention-4; transferring to grouped-query attention takes just 30 more minutes for up to 9.3% gains.
More from Infra
- Altman says OpenAI aims to replace Nvidia in the medium to long term — zephyr_z9 · 2026-11-11
- Chinese University Bought Nvidia H200 Chips Despite US Blacklist — pstAsiatech · 2026-11-11
- Baidu's ERNIE Evolve tops public bests on NanoGPT Speedrun, brings RSI to real inference optimization — xiaohu · 2026-11-11
- Baidu says RSI auto-optimized ERNIE inference, 6.24x decode speedup on Kunxin P800 — xiaohu · 2026-11-11
- Nvidia's supply commitments jump from $119B to $279B in one quarter, locking up 2027-28 memory — tengyanAI · 2026-11-11
- Train 27B models on a 24GB GPU: ternary optimizer cuts memory to 1 byte/weight — oguzeldereli · 2026-11-11