# Qwen 3.6 27B 在 MacBook Pro M5 Max 64GB 上实现每秒34个token的本地推理

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-05-14 13:34
- AIHOT 分数：77
- AIHOT 链接：https://aihot.virxact.com/items/cmp5226q00anisljxamkxkoxp
- 原文链接：https://x.com/rohanpaul_ai/status/2054797570946511018

## AI 摘要

近期，Qwen 3.6 27B大型语言模型通过TurboQuant技术被量化为GGUF格式，并整合Multi-Token Prediction技术。在配备M5 Max芯片和64GB内存的MacBook Pro上，该模型实现了每秒34个token的本地推理速度。高达90%的接受率表明，性能提升并非以牺牲输出质量为代价，而是通过避免重复的全成本解码工作来达成。同时，利用llama.cpp进行高效调用，进一步优化了运行效率。这一技术组合显著扩展了“笔记本电脑AI”的应用边界，使得在本地设备上流畅运行大型模型成为可能，提升了用户体验。

## 正文

Qwen 3.6 27B on a MacBook Pro M5 Max 64GB hitting 34tokens per sec， locally with atomic【.】chat

90% acceptance rate， i.e. most draft tokens matched what the main model would have produced， so the speed gain is not from skipping quality checks， but from avoiding repeated full-cost decoding work.

TurboQuant and GGUF handle the storage and runtime side： the model is compressed enough to run locally， while llama.cpp can feed Apple Silicon efficiently instead of waiting on huge weight movement.

Pretty serious local-inference result， changes what "laptop AI" can feel like.

### 引用推文

> atomic.chat：Multi-Token Prediction (MTP) for Qwen on LLaMA.cpp! +40% performance! 90% acceptance rate. Running locally on a MacBook Pro M5 Max 64GB We patched LLaMA.cpp, qu...