Multimodal chat with Qwen 3.8 27B on ZeroGPU.
Package and upload MLX MoE models for Flash‑MoE inference