这是indexloc提供的服务,不要输入任何密码
Skip to content

[QST] how to use cutlass in pytorch backward #2755

@xkwnzxxjw

Description

@xkwnzxxjw

I have a question: if I develop a custom MM operator using Cutlass and use it in PyTorch. If I use this operator for network training, PyTorch cannot perform backpropagation on this custom MM operator. Do I need to develop a backward MM operator? If so, how can I integrate it into PyTorch?

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