Skip to content

Questions about training #16

@XiaozhenQiao

Description

@XiaozhenQiao

This is very interesting work, but I encountered a problem when reproducing the training on PI3: the results are similar to PI3 during full sequence training, but once I enable chunking, the model's performance drops significantly, and I don't know why.

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