Skip to content

Question about next vision token prediction #3

@lntzm

Description

@lntzm

Thank you for sharing your code. As the paper describes the vision loss as AR, the code of this line seems to directly reconstruct the current token rather than the next token. I wonder where the label shift operation for AR is, or if I might have misunderstood this part? Thx.

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