Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

WU KONG checkpoint分享 #2

Open
zhanghang1995 opened this issue Sep 7, 2024 · 4 comments
Open

WU KONG checkpoint分享 #2

zhanghang1995 opened this issue Sep 7, 2024 · 4 comments

Comments

@zhanghang1995
Copy link

开发者你好,请问有已经训练好的权重分享,支持本地推理吗

@XR-stb
Copy link
Owner

XR-stb commented Sep 7, 2024

现在仓库里面就是我在用的参数,见judge.py,还在调整,所以这部分还有点凌乱. 目前打虎先锋感觉还不能收敛(打了40h了差不多),还在逐步调整参数。

@zhanghang1995
Copy link
Author

好的,感谢啦。
1)请问是否方便提供一份训练40h的权重,供测试下,看看效果,方便的话,感谢您同步一份邮箱([email protected]),也看看是否有新的更新路径?
2)请问比如虎先锋场景训练的模型,是否能直接迁移至其他的boss场景,迁移是否需要重新制定奖励规则;
3)针对每一帧的图像信息读取,有考虑引入大模型的能力(或者对接API)?

@XR-stb
Copy link
Owner

XR-stb commented Sep 7, 2024

  1. 已发到对应邮箱
  2. 可以直接迁移,奖励分你可以自己调整,现在的奖励分和模型收敛速度还不是很好。只能打打幽魂这种boss。
    不过restart.py 死亡自动走路到boss的逻辑你要自己调整
  3. 这样感觉会增加每帧处理的速度。如果不能再0.5内处理完一帧(现在每帧再用cudann加速的情况下也平均0.25s了),AI就会变成高ping战士,只能挨打。

@XR-stb
Copy link
Owner

XR-stb commented Sep 7, 2024

对了B站图灵的猫在做这一块的内容了,他视频里面说了做完后会公开这部分的代码。你也可以关注它后面的内容https://www.bilibili.com/video/BV1qE421c7mU/

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants