I have to retract what I posted as the fastest speed of Qwen 3.8 Flash on a single DGX…
This is a AI post classified by Jev as AI infra & evals (a model release), kept by the AI Radar because it carries real work, not commentary.
I have to retract what I posted as the fastest speed of Qwen 3.8 Flash on a single DGX Spark. Cruz shipped a custom ExLlamaV3 fork and a native engine path. I ran the same model on the same Spark: 102.6 max, 80 tok/s on code, 70 tok/s on prose. My previous benchmark and community recipes was 48.6 tok/s. I ran my full grid on it so you don't have to. This might now be the best way to run this model, without breaking the bank on one DGX Spark. THE QUANT EXL3 at 3.05 bits per weight. Trellis coding packs information more efficiently than plain rounding, so 3 EXL3 bits should land near 4.5
Posted by Yume_X (2.1k followers) 1 h ago · 22 likes · 2.3k views · view the original post on X. Kept by the AI Radar as AI infra & evals.
More AI work like this
- How I AI: spend 9¢ to look at ~1750 PRs and group them into thematic investments. 17,364… — @clairevo
- Last night, in a basement in Florida, one conversation ran across two kinds of silicon… — @volatilemarkts
- 如何自己训练一个 Jev 一样的判断模型？ — @AstroHanRay
- Qwen-Image-2.1 on one Spark: day-0 Comfy, 23/23, H3 kept — @MichaelGannotti
- UPDATE we COOKING ! — @Tech2Wild
- just launched teams on http://classifier.dev! — @michael_chomsky
- we've been exploring how to use Jev to build a better harness — @hwchase17
- Step-5-preview inference speed has improved, time to run more evals. — @ItsmeAjayKV
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 34.9k posts from 5k X accounts over the last 14 days, 1.5k tools, 19 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-20 20:06 UTC. Full method.