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。

Ask HN: Local LLM Experience

1 点作者 enginaar12 个月前
i gave Llama 3 Instruct 7B Q8 a try using LM Studio on a 16 GB Macbook Pro w&#x2F; M2. not sure what do at first I asked it to be a dungeon master to get some creative conversation going and i&#x27;m satisfied with both the performance and creativity if not impressed (link: https:&#x2F;&#x2F;pastebin.com&#x2F;raw&#x2F;iM4U8skk if you&#x27;re interested)<p>configuration:<p>- laptop: macbookpro m2 w&#x2F; 16 gb ram<p>- context length: 8192 (max)<p>- gpu layers: 33 (max)cpu threads: 8<p>response:<p>- time to first token: ~2s by the end of the conversation (4892 total token count)<p>- speed: ~7-8 tok&#x2F;s<p>- memory usage: 13GB (system total)<p>- memory pressure: slightly over 50% (&gt;90% when coding with containers)<p>now, these results are on par with ChatGPT 4. i compared a few general knowledge questions and coding problems, including some niche libraries, and it seem to do very well against ChatGTP 4 as well.<p>i want to compare your experience and wonder your opinions. is it possible to run copilot alike with a local server?

暂无评论

暂无评论