Jeff-Code: a 0.8B model makes Qwen 3.8-27B coding 47% faster, same pass rategithub.com2 points·firelex··0 commentsOpen articleSaveView on HN