I run qwen 3.8 27b on my m5 mbp, with 48gb of unified ram and I’m getting around 10-15 tok/s.

3.6 35b a3b, I’m getting upwards of 100