Fix shift-parallel CUDA graph capture and dispatch - #258
Merged
sfc-gh-yewang merged 4 commits intoMar 28, 2026
Conversation
xylian86
requested review from
sfc-gh-goliaro,
sfc-gh-jrasley,
sfc-gh-mhidayetoglu,
sfc-gh-reyazda and
sfc-gh-yewang
as code owners
March 28, 2026 06:14
sfc-gh-yewang
approved these changes
Mar 28, 2026
sfc-gh-mhidayetoglu
pushed a commit
to sfc-gh-mhidayetoglu/ArcticInference
that referenced
this pull request
Aug 12, 2026
Cherry-picked from public repo: snowflakedb#258 Co-authored-by: Xinyu Lian <lian7@illinois.edu> Made-with: Cursor
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR fixes CUDA graph integration for the shift-parallel model.
It wraps the shift model with
CUDAGraphWrapperin full-cudagraph mode, registers shift-model cudagraph keys with the dispatcher, enablesca_commduring shift graph capture, and restricts/pads shift capture to batch sizes >= 8.Why
The shift-parallel path was not fully aligned with cudagraph capture/runtime dispatch, which could lead to incorrect dispatch behavior and prevented using the intended custom all-reduce path during graph capture.