penta2himajin/qwisp

Fast, optionally-lossless local inference for Qwen3.6-35B-A3B (MoE) on Apple Silicon — OpenAI-compatible server + CLI over a raw-Metal decode engine (Seedless) on MLX.

语言：Swift · GitHub 星标：17 · Fork：1

话题：apple-silicon · llm-inference · local-llm · metal · mlx · moe · openai-compatible · qwen

在 GitHub 上查看原仓库