-
Notifications
You must be signed in to change notification settings - Fork 447
[BUG] Outdated Janus implementation needs refactoring #194
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
Comments
这个问题在本PR中已经被解决,目前本PR能够实现正常的Janus和Janus Pro模型的SFT和DPO训练 |
您那边是可以正常跑的?? |
我拉取的代码如下: |
这个问题本月稍早时已经经过测试被发现,是开发ti2t时兼容性未保证所导致的,目前修复已经写好,将在本PR的下一个commit中提交 |
好的 方便问问下次commit大概是什么时候么 麻烦了! |
请问最近有更新计划吗 |
Required prerequisites
What version of align-anything are you using?
Newest (cutoff 4.27)
System information
System version:
3.11.11 (main, Dec 11 2024, 16:28:39) [GCC 11.2.0] linux
Align Anything version: newest, cutoff date 0427, after commit eea5af6
Problem description
The existing Janus implementation was merged in February 2025. Since then, align-anything has undergone several major refactorings, which have resulted in the following issues with the Janus implementation:
TypeError: DPOTrainer.loss() got an unexpected keyword argument 'batch'
errorReproducible example code
/
Traceback
Expected behavior
No response
Additional context
I'm currently fixing these bugs. To be specific:
After all the bugs are fixed and tested, I will open a PR and merge these modifications. All the updates during this process will be reported in this thread.
The text was updated successfully, but these errors were encountered: