Comment by stoneforger

M4 mini pro 24gb qwen3-8b-mlx and others. Speed is fine, problem is context window. In theory CoreML would be better from an efficiency perspective but I think it's non-trivial to run models with CoreML ( could be wrong )