Jeff-Code: a 0.8B model makes Qwen 3.8-27B coding 47% faster, same pass rate(github.com)
2 pointsby firelex3 hours ago1 comment