fix: correct 3D mRoPE position_ids sharding in context parallelism#1482
Open
fix: correct 3D mRoPE position_ids sharding in context parallelism#1482
Conversation
mRoPE position_ids have shape [3, B, S] rather than standard [B, S]. Detect ndim==3 and shard along dim=2 instead of dim=1 to correctly split the sequence dimension. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: HuiyingLi <willwin.lee@gmail.com>
HuiyingLi
requested review from
ZhiyuLi-Nvidia,
adil-a,
akoumpa and
hemildesai
as code owners
Contributor
Author
|
/ok to test cf44135 |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Cherry-pick from internal
zhiqi-devbranch (commit033eb9c1).Author: zhiqil zhiqil@nvidia.com
mRoPE
position_idshave shape[3, B, S]rather than the standard[B, S]. The CP sharding code inmake_cp_batch_and_ctxwas usingcp_seq_dims=1for all buffers, which shardsposition_idson the wrong dimension (the 3-component mRoPE dim instead of the sequence dim). This causes anAssertionErrorin PyTorch'scontext_parallelwhencp_size > 1with VLM models like Qwen2.5-VL.Fix: Detect
position_ids.ndim == 3and setcp_seq_dims=2for that buffer so the sequence dimension is correctly sharded.Files changed
nemo_automodel/components/distributed/cp_utils.py— detect 3D mRoPE and shard on dim=2tests/unit_tests/distributed/test_cp_utils.py— 3 new unit tests covering 3D mRoPE, 2D standard, and 3D + loss_mask cases