Skip to content

Only disable flash attention in the decoder when dtw flag is enabled #3662

@predict-woo

Description

@predict-woo

Currently, the only way to use dtw is by disabling flash attention for the encoder and decoder.

However, we only really need to disable it in the decoder for dtw to work.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions