模型 / ALIBABA QWEN

Qwen3.6 27B

The memory figure is a conservative Q4 fit budget, not the official full-precision checkpoint size. Exact speed depends on the quantized build and runtime.

一手來源

總參數27B

估算記憶體~18.5 GBQ4 / NVFP4 estimate

上下文262K nativeText + Image

授權條款Apache-2.0Multilingual

01 / 適合

coding agents
reasoning
multimodal work

02 / 估算適配

Mac Studio M5 Ultra · 96 GB餘量充足 Mac mini M6 · 24 GB勉強適配 MacBook Pro M5 Max · 64 GB餘量充足 Framework Desktop · Ryzen AI Max+ 395 · 128 GB餘量充足 Radeon AI PRO R9700 · 32 GB餘量充足

03 / 部署方式

以下是執行方式範例，不是已驗證的安裝流程。容量估算對應量化檔案，官方倉庫可能是全精度權重；MLX、GGUF 與 TensorRT 需要相符的轉換檔或引擎。

vLLM進階vllm serve Qwen/Qwen3.6-27B --dtype auto

llama.cpp中等llama-server -m /path/to/verified-model.gguf --port 9931

LM Studio簡單Search “Qwen/Qwen3.6-27B” → Load a Q4/INT4 build → Start Server

MLX LM中等mlx_lm.server --model Qwen/Qwen3.6-27B

注意

The memory figure is a conservative Q4 fit budget, not the official full-precision checkpoint size. Exact speed depends on the quantized build and runtime.