Best way to run Qwen3.8 27B on M4 Pro 48gb?
retained text
I’m looking for a setup with the best balance of tps, ttft, context length and enough headroom for other tasks on the Mac. I’m getting a lot of conflicting opinions with regards to MTP, MLX, quantization etc. Please don’t recommend options less than 10 tps.
document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xevqqs/best_way_to_run_qwen38_27b_on_m4_pro_48gb