$5, 10-minute SFT on Qwen3.6-35B-A3B lifts GPQA +8% and MMLU-Pro +12%
josh_wills · x · 2026-09-21
Developer ekzhang1 ran an evening SFT on Qwen3.6-35B-A3B via Tinker to better handle "Jev-y" prompts: cost $5, 10 minutes of training, +8% on GPQA diamond and +12% on MMLU-Pro, with Jared Palmer's Kev evaled for comparison.
He also open-sourced openjev-sglang: a TypeSafe/Jev-compatible HTTP API served by Qwen3.6-35B-A3B on SGLang, one B200 per container leveraging SGLang 0.5.19's Rust frontend, radix caching and breakable prefill CUDA graphs, with a FastAPI/uvloop async API layer and one-command deploy on Modal.
More from Infra
- KubeCon keynote: computing is shifting from human speed to agent speed — HowDevelop · 2026-11-12
- Leaked memo: Meta infrastructure head Santosh Janardhan to leave in early 2027 — dinabass · 2026-11-12
- Omarchy 4.6 lets you share local AI models across your tailnet, says DHH — alexcovo_eth · 2026-11-12
- Eric Schmidt's foundation funds French NGO fighting against data centers — StephanSturges · 2026-11-12
- Epoch: Open-Weight GLM 5.3 Hacks Six Firefox Versions for ~$1,000 Each; Claude Fable 5.2 Sets ECI Record — Epoch AI · 2026-11-12
- Polymarket prices 28% odds of a state data center moratorium; Torus raises $500M — Polymarket · 2026-11-12