-
Notifications
You must be signed in to change notification settings - Fork 455
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
Update ORT training to be compatible with transformers 4.31 #1227
Conversation
The documentation is not available anymore as the PR was closed or merged. |
Related Issue: #1133 |
Gently pinging @pacman100 for the context. Current issue I met with the optimizer while using deepspeed zero stage 2
|
@JingyaHuang any updates? |
@JingyaHuang @pacman100 we are waiting for this issue to help unblock integrations within our team. It would be great if this could be fasttracked. Thanks! |
Thanks @pacman100 for helping out! So @prathikr, @kshama-msft, with the help of @pacman100 from the accelerate team, we are having ORTTrainer compatible with transformers 4.31 and accelerate 0.10. Can you review and also try out my branch to ensure the fix? Thanks! |
Thank you @JingyaHuang this PR resolved my issue. Please merge ASAP. |
Thanks Jingya for the prompt fix! |
What does this PR do?
As per title.
Maybe in the next PRs
These PRs are for the inference (evaluation / prediction) of trainer APIs. Given that the Trainer APIs are mainly used for training, and the ORT inference can always be done with
ORTModel
(s) in the optimum library, there is no priority nor eta for the following PRs. If anyone is interested in contributing, please feel free to open a PR and tag me for review. I am willing to handle them, but I have no vision of when I can have the bandwidth for them...