Skip to content
New issue

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

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

Already on GitHub? # to your account

<<论文复现>>如何对齐paddle中随机数种子与pytorch中随机数种子 #26637

Closed
xjhcassy opened this issue Aug 25, 2020 · 8 comments
Closed

Comments

@xjhcassy
Copy link

No description provided.

@LielinJiang
Copy link
Contributor

当前版本还无法做到。具体是用在什么层中呢?

@xjhcassy
Copy link
Author

全局的,像torch.manual_seed(opt.manual_seed)这样

@zhiqiu
Copy link
Contributor

zhiqiu commented Aug 25, 2020

全局的,像torch.manual_seed(opt.manual_seed)这样

hi,我们正在支持torch类似的manual_seed和generator功能。

@GuoQuanhao
Copy link

这个问题解决了吗,paddle有对应的API了吗 torch.manual_seed

@GuoQuanhao
Copy link

paddle.framework.seed(seed) 好像在2.0版本找到了

@ToscanaGoGithub
Copy link

似乎是不能完全对齐的,就算是设置一样的随机数种子,根据这个种子初始化一些参数的时候仍然会与一些区别。只能依靠加载预训练模型来进行对齐

@paddle-bot-old
Copy link

Since you haven't replied for more than a year, we have closed this issue/pr.
If the problem is not solved or there is a follow-up one, please reopen it at any time and we will continue to follow up.
由于您超过一年未回复,我们将关闭这个issue/pr。
若问题未解决或有后续问题,请随时重新打开,我们会继续跟进。

@yyinsomnia
Copy link

paddle.seed(),用paddle.randint测试过自身是能能稳定随机。
但是用同样的seed,paddle和torch randint出来是不一样的,这个要对齐确实有点勉为其难了。

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

No branches or pull requests

6 participants