-
Notifications
You must be signed in to change notification settings - Fork 433
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
Deprecate deepspeed #3512
Deprecate deepspeed #3512
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
🫡
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
FSDP is all you need
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Two and a half years later and I still am confused by DeepSpeed checkpointing
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
seems that all the deepspeed related tests are failing
What does this PR do?
This PR deprecates the usage of DeepSpeed in Composer. We have not updated DeepSpeed in a long time and do not actively maintain the integration. Composer is tightly coupled to PyTorch FSDP which provides similar functionality, and we are developing all new functionality on top of FSDP and PyTorch native, not DeepSpeed, for parallelism. Before removal we will provide a migration guide for anyone still using DeepSpeed. In the meantime, users are welcome to open a GitHub issue if they have any trouble.