4 ms·

Have you looked at using oMLX? https://omlx.ai/ https://omlx.ai/

• mc8098 2mo ago
  Would second this, I switched to oMLX I get ~75 tok/s on Qwen3.6-35B-A3B-4bit on a 48GB M5 Pro