7x speed improvement for LLaMA in less than 10 lines of codegithub.com2 points·hack_ml··1 commentOpen articleSaveView on HN