Skip to content
Parameters 70BLlama 3.3 Community LicenseEstimated

Llama 3.3 70B Instruct

需要 48GB 级显存的大家伙:本地能力的天花板之一。

Official source:https://github.com/meta-llama/llama-modelsMetaYou are leaving AIWorksWell for Meta. That site is not under our control; its content, privacy and security are its own responsibility. Please judge for yourself.

What hardware it needs

Estimated
Parameters
70B
Architecture
Dense
Context
128K
License
Llama 3.3 Community License
Min VRAM
47G
RAM (CPU-only)
64G
Download size
—G

Requirements by quantisation

The same model at different quantisations is a different download — the most common misjudgement in local AI.

Suggested quantisationSizeMin VRAMQuality
F16280G322GOriginal precision
Q8_0148.4G171GPractically lossless
Q6_K114.8G132GNear lossless
Q5_K_M99.4G114GVery good
Q4_K_M★85.4G98GBest value for most people
Q3_K_M70G81GNoticeable loss
Q2_K53.2G61GLast resort only

Method:显存为 Q4_K_M 档参考

Your device

A plain-language profile of every model you can run yourself: parameter count, hardware it needs, which quantisation fits. No rankings, no prices — just what it takes to run.

Save your device profile once and this page will tell you whether it runs on your machine.Add your device·No account? Try a one-off check

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:Llama 3.3 70B Instruct
用途:需要 48GB 级显存的大家伙:本地能力的天花板之一。
官方来源:https://github.com/meta-llama/llama-models
许可:Llama 3.3 Community License
Scroll inside the box with your mouse wheel — copying still gives you the full prompt1267 chars

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

About this entry

它是什么

70B 级稠密模型的定位很清楚:能力接近云端大模型,但门槛也在那里——量化后仍需 40GB 以上可用显存,普通消费级显卡单人基本跑不动。来源:https://github.com/meta-llama/llama-modelshttps://github.com/meta-llama/llama-models

怎么用起来

两条现实路径:① 多卡/大显存工作站(48GB+);② 走在线 API。Mac Studio 等大统一内存机型也可以,但速度取决于内存带宽。来源:https://github.com/meta-llama/llama-modelshttps://github.com/meta-llama/llama-models

以上坑来自 https://github.com/meta-llama/llama-modelshttps://github.com/meta-llama/llama-models 与官方文档。

来源与署名

Mentioned from:https://github.com/meta-llama/llama-modelshttps://github.com/meta-llama/llama-models · 原作者 Meta(Llama 3.3 Community License)。