‹ BackHN Continuity

Thread

Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

741 points · 330 comments · snehesht

  1. Tepix · · focus · HN ↗
    Q2 quantization. Not interested.
    1. tcdent · · focus · HN ↗
      All of these projects targeting low spec systems and "100 tok/s" are the same 2 bit quant without much else. Conveniently none of them include any mention of accuracy in their published numbers. 4 bit is the floor.
      1. latentsea · · focus · HN ↗
        You can run a 4 bit quant with this. Personally, I switched to running IQ3_XXS and am getting better outputs than 27B and at faster speeds.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.