Hacker News Reader

How does this compare to llama.cpp in terms of performance?

MLX is a bit faster (low double digit percentage), but uses a bit more RAM. Worthwhile tradeoff for many.

On my M4 Pro MLX has almost 2x tok/s