🔥 Same M5 Max, same 4-bit Qwen3.8-27B: Kapitsa 143 tok/s, mlx-serve 128, Ollam… · @QwenAIFans · TwKit

@QwenAIFans

🔥 Same M5 Max, same 4-bit Qwen3.8-27B: Kapitsa 143 tok/s, mlx-serve 128, Ollama 71, LM Studio 33. Caveat: Kapitsa's own bench, and the others ran about a week earlier. via @kaimi_local 👇 https://github.com/nace-ai/kapitsa/blob/main/docs/benchmarks.md

曝光 482 · 评论 4 · 点赞 6 · 书签 3