deepseek-v4.1IMAGE
DeepSeek V4.1 Flash on one RTX 5090 plus 256GB RAM: about 31 tok/s
Runs the ~340GB Maya-S quant on a Windows box with one RTX 5090 and 256GB RAM at about 31 tok/s (up to 35); disabling --mmap-experts lifted it from 7.7 to 25 tok/s, while AVX512 gave no speedup.
stou
@Stou_1Archived 8 days agoOriginally posted 2026.10.24
stou @Stou_1Originally posted 2026.10.24
DeepSeek-V4.1-Flash（約 340GB、Maya-S）を RTX 5090 1 枚＋RAM 256GB の Windows 機で動かして、生成 約 31 tok/s（速いときは 35）。Project Maya v1.1.25。効いたのは --mmap-experts off（7.7→25 tok/s）と、Q8_BATCH・AVX2 の 2 つの環境変数。 AVVX512を使えば早くなると思っていたけど違った
x.com/stou_1/status/2113776800400883939Archived 2026.10.24 12:06