Skip to content

Add FSDP support for large models #15

@ikanher

Description

@ikanher

We currently support only distributed data parallel.

When working with bigger models (e.g. with language models), the models are so large that they don't fit anymore in a single GPU memory, but need to be distributed across GPUs. This is where Fully Sharded Data Parallel (FSDP) comes in.

Current Opacus already has some kind of support for this, see: https://github.com/meta-pytorch/opacus/blob/main/examples/fsdp_example.py so likely we should be able to support FSDP also.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions