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’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Transpose for reduce window. #21197

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Transpose for reduce window. #21197

wants to merge 1 commit into from

Conversation

copybara-service[bot]
Copy link

Transpose for reduce window.

Implements the transpose rule for general reduce window support. We do this by padding out the cotangents and running a reduce_window on the cotangents, using mostly the same windowing properties, with the specific difference in how we approach dilation and padding during the reduction (as these are accounted for during the padding of the cotangents).

Note - that this CL highlights a small bug/issue in the tracing mechanism itself. We cannot reuse the primals side of the JVP, and are forced to recompute it, due to the fact that attempting to use primals from the initial JVP computation flags the primals side arrays as unknown.

Copy link

google-cla bot commented May 12, 2024

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

Implements the transpose rule for general reduce window support. We do this by padding out the cotangents and running a reduce_window on the cotangents, using mostly the same windowing properties, with the specific difference in how we approach dilation and padding during the reduction (as these are accounted for during the padding of the cotangents).

Note - that this CL highlights a small bug/issue in the tracing mechanism itself. We cannot reuse the primals side of the JVP, and are forced to recompute it, due to the fact that attempting to use primals from the initial JVP computation flags the primals side arrays as unknown.

PiperOrigin-RevId: 632679634
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

0 participants