utsav1033/carmen

An AI writes Apple Metal GPU kernels, a judge it can't fool proves them, and the winners run inside a real model: Qwen 2.5 0.5B decode 3% faster with identical output; 75/75 planted bugs caught.

2 best points · 0 comments · 1 Hacker News post · 2 GitHub stars · View on GitHub

Hacker News discussions

• Show HN: Carmen – harness for reliable AI-written Metal kernels (2 points, 0 comments, by utsav1033, 2026-10-27)