Skip to content

[Feature Request] Return LSE from all TRT LLM attention kernels #2169

@divchenko

Description

@divchenko

Currently only some APIs return LSEs, while underlying kernels are capable of returning LSEs in all cases. Extend API to always return LSEs if asked. Example implementation: fw-ai/flashinfer#7

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions