PrismML 将 27B 大模型压缩至手机可运行

@xiaohu

PrismML 将 27B 模型塞进你的 iPhone 里

而且智商几乎没怎么缩水 PrismML公司基于Qwen3.6-27B模型，将约 54GB 的 27B 模型压到约 3.9–5.9GB

使其能在手机上运行

在「高强度思考模式」下用 15 项测试对比原版：

Ternary 版本（5.9GB 电脑版）：保留了原版 95% 的战斗力！

1-bit 版本（3.9GB 手机版）：也保留了 90% 的战斗力！

在电脑（RTX 5090）上最快能达到每秒 163 个字；在苹果 M5 Max 芯片上也能达到每秒 87 个字。

!Image 2: Video thumbnail

01:19

9 Replies

1 Retweets

22 Likes

8,124 Views 

One Sentence Summary

PrismML, based on the Qwen3.6-27B model, uses compression to run it on a phone, retaining 90%-95% performance.

Summary

Xiaohu introduces PrismML's model compression achievements: compressing the ~54GB 27B model to 3.9-5.9GB, enabling mobile deployment while retaining 90%-95% performance, and provides data on inference speeds on RTX 5090 and M5 Max.

AI Screening

82

Influence Score 13

Published Today

Language

Chinese

Tags

PrismML

Model Compression

On-Device AI

Qwen

27B

Make your daily reading actually fit you.A daily brief built from the sources you follow. Get started free