3 ms·MLX LM 0.20.1 has the comparable speed as llama.cpp with flash attention1 points by tosh 2y agodeleted 2y ago[deleted]