Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We鈥檒l occasionally send you account related emails.

Already on GitHub? Sign in to your account

cudagraph error with set_ on inputs #126187

Open
eellison opened this issue May 14, 2024 · 1 comment
Open

cudagraph error with set_ on inputs #126187

eellison opened this issue May 14, 2024 · 1 comment
Labels
module: cuda graphs Ability to capture and then replay streams of CUDA kernels oncall: pt2 triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Comments

@eellison
Copy link
Contributor

eellison commented May 14, 2024

馃悰 Describe the bug

We need to track storage liveness for correctness and expect the storages to be on the outputs.

A possible fix here would be to add the inputs which are mutated as outputs, but needs further investigation.

Repro (right now internal only) https://www.internalfb.com/phabricator/paste/view/P1360460962.

Error logs

...

Minified repro

...

Versions

master

cc @mcarilli @ezyang @peterbell10 @msaroufim @bdhirsh @anijain2305 @chauhang

@eellison eellison added triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module module: cuda graphs Ability to capture and then replay streams of CUDA kernels oncall: pt2 labels May 14, 2024
@bdhirsh
Copy link
Contributor

bdhirsh commented May 14, 2024

hmm @eellison if we (eventually) care about FSDP2 + cudagraphs, does that mean we're going to have to make sure that the set_() happens outside of the graph? (I think that will only happen if there's a graph break mid-layer, so maybe this isn't too important of an issue..)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
module: cuda graphs Ability to capture and then replay streams of CUDA kernels oncall: pt2 triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
None yet
Development

No branches or pull requests

2 participants