Comment by amboo7

I also have an M1 Max 64GB: Qwen 3.6 benefits from MTP (after rounds of parameter optimization). MLX was unstable (haven't tried it recently), faster at TG but slower at PP, so inconclusive.