Fast chat with Qwen3.8-Flash-Next on 4xH200
Package and upload MLX MoE models for Flash‑MoE inference