fix(mtp): support contiguous context-parallel THD rolls - #5946
Closed
FDecaYed wants to merge 1 commit into
Closed
Conversation
Add configurable boundary-fill semantics and roll packed THD MTP inputs across contiguous context-parallel shards without crossing sequence boundaries. Reconstructed from the MTP portion of NVIDIA#5795 on top of the padded-cu-seqlens fix in NVIDIA#4495. Signed-off-by: Deyu Fu <Deyu.Foo@gmail.com>
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Contributor
Author
|
/ok to test f57f2df |
Contributor
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Extend MTP packed-sequence rolling so contiguous context-parallel THD shards roll across rank boundaries without crossing packed-sequence boundaries.
cp_partition_modecontract used by the dynamic-CP stack;Truefor padding masks;Stack
This PR is stacked on #4495, which fixes padded cumulative-sequence boundaries for MTP THD rolling.
The future dynamic-CP integration supplies
PackedSeqParams.cp_partition_mode; this focused PR deliberately does not pull that broader data/attention stack into MTP. Its test exercises the same duck-typed contract directly.Provenance
Testing