TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

How to evaluate performance of LLM inference frameworks

18 点作者 matt_d9 个月前

2 条评论

Havoc9 个月前
&gt;Aggressively pruning LLMs via quantization can significantly reduce their accuracy and you might be better off using a smaller model in the first place.<p>Not sure that is correct. Quantization charts suggest its a fairly continous spectrum. i.e. an aggressive quant 13B ends up about same as a no quant 7B:<p><a href="https:&#x2F;&#x2F;www.researchgate.net&#x2F;figure&#x2F;Performance-degradation-of-quantized-models-Chart-available-at_fig1_377817624" rel="nofollow">https:&#x2F;&#x2F;www.researchgate.net&#x2F;figure&#x2F;Performance-degradation-...</a>
brrrrrm9 个月前
If you&#x27;re hitting a memory wall it means you&#x27;re not scaling. This stuff really doesn&#x27;t apply to scaled up inference but rather local small batch execution