Unverified50% confidenceBenchmarkExact time

Qwen 27B的Q4_K_M/GGUF版本在同一台机器上解码速度约为10.7 tok/s，约为MLX版本的1/3.8

1

Sources

50%

Confidence

Long-term

Relevance

10/2/2026

First Seen

Sources

M4 Max实测Qwen 27B：本地模型的tokens/s背后真相

Related Entities

Qwen 27Bmodel Q4_K_Mconcept MLXtechnology

Related Claims

Unverified16G显存部署Qwen3 27B只能选到Q3_K_XL的GGUF量化版本，这是16G显存能容纳的精度上限77% similar UnverifiedQwen3 27B本地推理速度约为每秒10个tokens77% similar VerifiedQwen2.5-Coder-32B在RTX 4090上以Q4量化运行，推理速度可达每秒40-60个token76% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/968804

API

curl https://kongchang.com/api/v1/knowledge/claims/968804

MCP

get_claim(id=968804)