Skip to content

8-bit optimizers dont work with FSDP #89

Description

@prajdabre

When I use an 8-bit ADAM with FSDP, I get an error as follows:

RuntimeError: output tensor must have the same type as input tensor

If my understanding is correct, there seems to be a casting issue. Is there any workaround this?

TIA.

Metadata

Metadata

Labels

Contributions WelcomeWe welcome contributions to fix this issue!DuplicateThis issue or pull request already existsFSDPOptimizersIssues or feature requests relating to optimizersTo Discuss Internally

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions