8000 Trying to use forward AD with _scaled_dot_product_flash_attention that does not support it because it has not been implemented yet. · Issue #128971 · pytorch/pytorch · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content
Trying to use forward AD with _scaled_dot_product_flash_attention that does not support it because it has not been implemented yet. #128971
< 8000 svg focusable="false" aria-label="Issue" class="octicon octicon-issue-opened Octicon-sc-9kayk9-0 hTWZgt" role="img" viewBox="0 0 16 16" width="16" height="16" fill="currentColor" display="inline-block" overflow="visible" style="vertical-align:text-bottom">Open
@lciti

Description

@lciti

🚀 The feature, motivation and pitch

As suggested in the error message, I am reporting this error so its implementation can be prioritized.

Trying to use forward AD with _scaled_dot_product_flash_attention that does not support it because it has not been implemented yet.

Trying to use forward AD with _scaled_dot_product_flash_attention_for_cpu that does not support it because it has not been implemented yet.

Alternatives

No response

Additional context

torch.__version__: '2.3.1+cu121'

cc @ezyang @albanD @gqchen @pearu @nikitaved @soulitzer @Varal7

Metadata

Metadata

Assignees

No one assigned

    Labels

    actionablemodule: autogradRelated to torch.autograd, and the autograd engine in generaltriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

      0