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

Ring Allgather + GEMM Overlap HostIR Implementation #3626

Open
wants to merge 13 commits into
base: main
Choose a base branch
from

Conversation

nsarka
Copy link
Member

@nsarka nsarka commented Dec 20, 2024

This PR implements the HostIR version of the Ring AG+GEMM overlap scheme. It also has some minor code-cleanliness updates to the ATen version.

@nsarka nsarka self-assigned this Dec 20, 2024
for ([[maybe_unused]] const auto& _ :
c10::irange(params.number_of_iterations)) {
// I don't know why but this seems necessary...
at::manual_seed(getATenRandomSeed());
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I spent a while before realizing that without this line, the test fails with a data validation error. I pasted the line from one of Sam's tests (comment included)

@nsarka nsarka force-pushed the nsarka/ring-ag-overlap-hostir branch from 3f3c77a to a9ef116 Compare December 20, 2024 16:19
@nsarka
Copy link
Member Author

nsarka commented Dec 20, 2024

!build

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant