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。

Show HN: Fine Tune LLM's Without Code

1 点作者 Nevin1901超过 1 年前
Hi,<p>During the recent explosion of LLM&#x27;s, I&#x27;ve been playing around with Llama-2 7b on my Mac, and have wanted to fine tune the model.<p>I&#x27;ve created a service which lets you upload a json&#x2F;jsonl file, and it&#x27;ll automatically fine tune Llama&#x2F;Mistral&#x2F;anything from hf on that dataset.<p>You don&#x27;t have to &quot;burn&quot; money trying to get fine tuning scripts to work or use gpt-3.5. You just upload your dataset, let the model fine tune, and then download the model to run locally.<p>I personally used my service to fine tune Code Llama on python docstring comments. I then took the model and converted it to .gguf (for use in llama.cpp) and I&#x27;m now able to generate python docstrings without sending my code to OpenAi (it works surprisingly well).<p>I hope this helps some people. I&#x27;m manually onboarding users and I&#x27;m interested in any feedback you might have.

暂无评论

暂无评论