first tests with a qwen code model (Qwen3-Coder 30B-A3B) on my framework laptop - codex didnt work so well, but from a technically working perspective, opencode appears to be fine. the first test script contained some errors, it is currently tasked in trying to find and correct its errors 👀 something of around 36t/s initially measured vs Qwen3.6 35B-A3B @42.3 t/s on my m2pro mac mini
0 Boosts
0 Likes
interesting insight, i think i never really looked into whats happening on compaction
done in ... 21m 33s · 18.9 tok/s 😅
@milan And here was me just wondering if Milan was dabbling in a little AI these days. 😎