8000 turbomind attention算子接口求助 · Issue #3553 · InternLM/lmdeploy · GitHub
[go: up one dir, main page]

Skip to content
turbomind attention算子接口求助 #3553
@Lubenwei-nb123

Description

@Lubenwei-nb123

Checklist

  1. I have searched related issues but cannot get the expected help.

我在知乎上看到lmdeploy的turbomind attention性能是flash attention的2倍,正好手上V100比较多,想看一下V100上turbomind attention相比xformers的性能会不会更好,但是一直找不到可以直接调用turbomind attention的接口。

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

      0