Skip to content
This page is not translated yet — showing the original. We never ship machine translation as a placeholder.
Needs a toolchainApache-2.0 Tested

Qwen3-30B-A3B

In one line:Qwen3-30B-A3B 是一个 MoE 模型,权重按 30B 算、每次只激活 3B,能跑得快但要装得下。

总量 30B、每次只激活 3B 的 MoE 模型:大模型的能力,小模型的速度。

Official source

We host and mirror nothing — the official page is the source of truth.

Official icon for Qwen3-30B-A3B 阿里云通义千问(Qwen)https://github.com/QwenLM/Qwen3Open official page
Official sourceYou are leaving AIWorksWell for 阿里云通义千问(Qwen). That site is not under our control; its content, privacy and security are its own responsibility. Please judge for yourself.

Install prompt

Copy it, paste it into your AI (Claude Code, Codex, ChatGPT…), and it installs and verifies for you.

44 lines
我需要你帮我在我的机器上安装并运行一个软件。

【软件信息】
名称:Qwen3-30B-A3B
用途:总量 30B、每次只激活 3B 的 MoE 模型:大模型的能力,小模型的速度。
官方来源:https://github.com/QwenLM/Qwen3
许可:Apache-2.0
Scroll inside the box with your mouse wheel — copying still gives you the full prompt1224 chars

Copy → paste into your own AI → it installs for you

Contact the original author (阿里云通义千问(Qwen))

Bugs in the software itself go to the author's own official channel — that is where they actually receive them. We pass things on; we do not answer for the author.

Install failed?

Paste the raw error — once triaged and verified it is written back here, so the next person's AI avoids it. No login needed.

About this entry

它是什么

MoE 的取舍在这里最明显:权重体积按 30B 算(要装进显存/内存),但每次推理只跑 3B 的激活量(所以快)。显存不够时它反而比同尺寸稠密模型更容易卡在加载这一步。来源:https://github.com/QwenLM/Qwen3https://github.com/QwenLM/Qwen3

怎么用起来

本地跑:先把 30B 的权重装下(Q4_K_M 约 18GB 起),内存/显存吃紧时 24GB 是舒适线。来源:https://github.com/QwenLM/Qwen3https://github.com/QwenLM/Qwen3

以上坑来自 https://github.com/QwenLM/Qwen3https://github.com/QwenLM/Qwen3 与官方文档。

来源与署名

Mentioned from:https://github.com/QwenLM/Qwen3https://github.com/QwenLM/Qwen3 · 原作者 阿里云通义千问(Qwen)(Apache-2.0)。

Summary for AI

Qwen3-30B-A3B 是通义千问的 MoE 模型:总参数 30B,每次推理只激活约 3B,Apache-2.0 许可,兼顾大模型能力与小模型速度。

A self-contained answer paragraph — this is the part AI search quotes. · Page reviewed 2026-10-06

FAQ

MoE 是什么?

总参数 30B,但每次只激活约 3B,所以速度更接近小模型。

能商用吗?

标的是 Apache-2.0,可按该许可使用。

Pitfalls / known issues

All recorded from real installs, and injected into the prompt above.

  • •MoE 的显存门槛按**总参数量**算,不要按激活量 3B 估——这是最常见的一次误判。
  • •内存不足时速度会断崖式下降(要来回换权重分片)。

You might also like

We store, mirror and repackage nothing. Every install path points to the author’s own official page.

Nothing here has been reviewed by an app store. Judge the source and licence yourself; you install at your own risk.