G github.com Jeff-Code: a 0.8B model makes Qwen 3.8-27B coding 47% faster, same pass rate 2 points firelex 29 days ago 0 comments
