Skip to content

Code error in reproducing the result of PyramidKV paper. #54

@yankeee-0411

Description

@yankeee-0411

self.kv_seq_len is accumulated when generating samples in the loops. Only the first sample uses the pyramidKV while the remaining ones don't. We need to set the self.kv_seq_len to zero in each new iteration.

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