Skip to content

Distributed Training with Multiple GPUs Slower Than Expected #19

Description

@khanhnguyenvinh2002

Hello,

I've been attempting to train a model using my dataset and wanted to leverage multiple GPUs to speed up the process. However, I've encountered an issue where using multiple GPUs seems to slow down the training rather than speeding it up. I'm seeking guidance on configuring distributed training in this context properly.

Upon inspecting the get_dataloader function in the code, I noticed that DistributedSampler is not being utilized. Additionally, the function employs a while loop combined with a yield statement. I'm unsure if these aspects are contributing to the inefficiency.

Any insights or advice on setting up distributed training correctly would be greatly appreciated. Thank you in advance for your help!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions