配合Claw最优化的高性价比Local LLM配置 Mac Mini M4 24GB Qwen3-30B-A3B MoE Q4 mlx_lm ≥ 0.25.2 pip install mlx-lm --break-system-packages mlx_lm.server --model Qwen/Qwen3-30B-A3B-MLX-4bit --port 8080 vi ~/.openclaw/openclaw.json { "models": { "providers": { "local-mlx": { "baseUrl": "http://127.0.0.1:8080/v1", "apiKey": "none", "api": "openai-responses", "models": [ { "id": "qwen3-30b-a3b", "name": "Qwen3-30B-A3B-4bit", "reasoning": true, "input": ["text"], "cost": { "input": 0, "output": 0 }, "contextWindow": 8192, "maxTokens": 4096 } ] } } }, "agents": { "defaults": { "model": { "primary": "local-mlx/qwen3-30b-a3b" } } } } contextWindow 从 8192->16384->32768 /no_think 模式大概 30-40 token/s 的速度。