Re:My local model setup on an M4 Pro Mac Mini
by lwsio on Wednesday September 02, 2026 @06:06AM
(#49534161)
My perf sucks compared to yours. Added it to the post - same model averages 325 tok/s in processing prompts, and 34 tok/s in token generation. What am I doing wrong..?
[ Parent
| on HN ]
Re:My local model setup on an M4 Pro Mac Mini
by argee on Wednesday September 02, 2026 @11:14AM
(#49537572)
Wow, that's just about half the perf. I'm not sure what you're doing differently, though our hardware is a bit different: I am on a Macbook Pro M4 Pro, while you're on a Mac Mini.
I would try a different version of the model from HuggingFace while ensuring it's MLX. I'm also using LM Studio, not oMLX, and I've seen some threads like these:
https://www.reddit.com/r/LocalLLaMA/comments/1spuwir/omlx_10...