极道 ·

一张AMD MI300X卡跑DeepSeek V4 flash：总吞吐能干到830 tok/s

💡 原文中文，约4400字，阅读约需11分钟。

📝

内容提要

一张显卡跑大模型？这事没那么简单。 GitHub上一个名叫deepseek-v4-flash-mi300x的仓库，公开了用一张AMD MI300X显卡跑起3040亿参数大模型DeepSeek V4 Flash的完整方案。实测单流解码中位数达到168.6 tok/s，预填冲到7.9K...

🏷️

DeepSeek低价风暴打服硅谷！海外平台争相倒贴V4 Flash
那还说啥了梁圣，我订阅费全给你就是了呗
DeepSeek-V4-Flash登顶全球调用量榜首，国产大模型包揽前五
【TechWeb】8月4日消息，7月31日，DeepSeek正式开放V4-Flash正式版API公测。短短一周之内，这款国产大模型便在全球AI开发者市场掀...
24GB显存Windows电脑上竟然跑起了DeepSeek-V4-Flash-0731？
Redditor网友惊呼：简直不敢相信我的家用电脑上居然运行着 DeepSeek-V4-Flash-0731，这可是早期版本啊！太不可思议了！二十个月前...
AMD’s data center business is booming while gaming takes a backseat
Driven by demand for AI capacity, AMD's data center revenue more than dou...
DeepSeek V4 Flash is 90% off through Novita on AI Gateway
DeepSeek v4 Flash is 90% off on AI Gateway when you route to Novita. The disc...
What Modern QA Engineers Actually Do: It's More Than Finding Bugs
Ask someone what a QA engineer does, and you'll probably hear a familiar ...