MetaGPT icon indicating copy to clipboard operation
MetaGPT copied to clipboard

humaneval

Open codeAlwaysPass opened this issue 1 year ago • 2 comments

请问可以提供测试humaneval的相关方法吗,现在想要进行测试但是完全不知道该如何下手,generate_repo返回的是一个Projectrepo类型不能被转换成jsonl,希望作者能加上去,谢谢。

codeAlwaysPass avatar Jul 27 '24 09:07 codeAlwaysPass

ProjectRepo对应的是一个文件夹路径。如果要转JSON,本质上是将一个文件夹转JSON。 你希望转成什么样的JSON? @codeAlwaysPass

iorisa avatar Jul 29 '24 07:07 iorisa

我希望转成一个可以让humaneval评估的jsonl文件,目前我是自己写python脚本把文件夹里面的代码写入jsonl,但可能是由于我的代码问题这样得到的jsonl获得的准确率只有0.27。

codeAlwaysPass avatar Jul 29 '24 08:07 codeAlwaysPass

由于版本迭代新增功能且迭代较快,engineer 中增加了不少 repo 级别的处理和设计。针对 HumanEval,我建议你先熟悉并使用 v0.4-release 或者之前的版本。在这个版本中,你可以获取 engineer 的执行结果(msg = Message(content=code, role=self.profile, cause_by=WriteCode))。你可以选择在 msg 生成之前直接将code保存到JSONL文件,或者在 engineer 执行结束后,通过 memory 的 get_by_actions 函数来获取代码,再将其写入 JSONL 文件。:)

stellaHSR avatar Oct 11 '24 06:10 stellaHSR

There will be some scripts related to humaneval in version 1.0

geekan avatar Oct 20 '24 07:10 geekan

请问现在有可以测试humaneval的script了吗?

wkwk-ai avatar Dec 17 '24 06:12 wkwk-ai

This issue has no activity in the past 30 days. Please comment on the issue if you have anything to add.

github-actions[bot] avatar Jan 18 '25 00:01 github-actions[bot]

This issue was closed due to 45 days of inactivity. If you feel this issue is still relevant, please reopen the issue to continue the discussion.

github-actions[bot] avatar Feb 01 '25 00:02 github-actions[bot]

请问现在有可以测试humaneval的script了吗?

YuanDaoze avatar Apr 10 '25 17:04 YuanDaoze

Follow up on this.

q8888620002 avatar Jun 11 '25 16:06 q8888620002