[II] Bind lazy B12X DCP pools to graph channels - #337
Closed
voipmonitor wants to merge 1 commit into
Closed
voipmonitor wants to merge 1 commit into
voipmonitor wants to merge 1 commit into
Conversation
Record each active B12X DCP graph channel by process-group identity. A pool initialized after the graph context opens now prepares and enters that channel before its first operation, and collective calls select the active channel instead of the eager scheduler channel. Eager execution is unchanged. The change depends on the graph-owner integration in vLLM #324. Validation: Ruff format/check and git diff --check pass. Thirty-seven relevant DCP tests pass in the CUDA 13.3/PyTorch 2.13 Kimi-K3 image; four focused eager/graph channel tests pass.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. 🗂️ Base branches to auto review (1)
Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Author
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Status
Qualified.
Behavior
B12X DCP pools initialized after a CUDA graph context opens join that graph's semantic channel before their first operation. LSE reduce-scatter and query-head gather select the channel associated with their process group; eager execution continues to use
vllm:eager:dcp.Technical reason
Projection geometry can create a B12X pool after the graph-owner context has already entered. Binding only the pools present at context entry leaves such a pool on the eager channel, which violates B12X graph ownership and prevents safe graph replay.
Compatibility
The change is generic to B12X DCP collectives and does not alter model geometry, supported world sizes, tensor formats, numerical operations, or NCCL fallback behavior. It depends on the graph-owner integration in #324.
Validation
git diff --checkpasses.tests/distributed/test_dcp_a2a.pypass in the CUDA 13.3/PyTorch 2.13 Kimi-K3 image.The remaining full-file test categories are outside this change: packed-LSE dtype expectations, NCCL tests requiring a larger Docker shared-memory allocation, and a test for the removed B12X
_load_extensionsymbol.