Skip to content

question about fp32 and mixed precision #4

Description

@yyyouy

Thanks for your work, when I finetuned the huge model in the setting of 1 percent of labels, I met a question about precision. when I used mixed precision, I always got the loss nan question.
but when I use fp32 precision, the accuracy of the classifier is 79.4.
So how can I solve this question? Thank you very much.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    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