Skip to content

Model is not reproducible even with fixed PyTorch seed #67

@linden-everest

Description

@linden-everest

I noticed that the model does not produce reproducible results even when I set the PyTorch random seed using torch.manual_seed(), and also control other randomness-related settings (e.g., setting torch.backends.cudnn.deterministic = True and torch.backends.cudnn.benchmark = False). Each run still gives slightly different results.

Has anyone managed to make this model deterministic? Any suggestions are appreciated.

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