Skip to content

The training accuracy of the custom dataset is close to 0 #52

@pengc-bjtu

Description

@pengc-bjtu

Hello author, I trained the custom dataset with the method you mentioned earlier. At first, the training loss was NAN, and then after I used gradient clipping, the loss was normal, but the accuracy was close to 0. I feel like there's something wrong with the configuration file, can you help me analyze it?my wechat account is “pctthappy”

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