Thread

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

741 points · 330 comments · snehesht

pilooch · 1791126702 · focus · HN ↗
My goto private setup, runs ~50t/sex on a dgx spark with sglang, nvfp4. Excellent model.
1. apitman · 1791128296 · focus · HN ↗
  This is a meaningless metric without knowing how long the sex takes
  1. swiftcoder · 1791128566 · focus · HN ↗
    30 seconds at most

Unofficial Hacker News client; not affiliated with Y Combinator.