[test] fix: shrink CP packed SFT functional model#4383
Merged
Conversation
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
yaoyu-33
added
area:training
Contributor
Author
|
/ok to test b17315a |
Contributor
ReviewLGTM. Clean fix that correctly addresses the OOM issue by shrinking the test model shape while preserving the coverage paths (CP + packing + checkpoint load). Good details:
Suggested test casesNo perf tests impacted. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
The CP packed SFT functional test now creates a pretrain checkpoint and then loads it for SFT in the same pytest process. Using the full Llama 3.2 1B recipe shape makes the second DDP buffer allocation run near the H100 memory limit and can OOM before the SFT path starts.
This keeps the test coverage but shrinks only the test model shape for both the pretrain and SFT providers, so the checkpoint remains compatible while still exercising:
It also runs a GC / CUDA cache cleanup barrier between the two phases.
Verification
python3 -m py_compile tests/functional_tests/test_groups/training/test_seqpacking_cp_example.pygit diff --checkuvx ruff check tests/functional_tests/test_groups/training/test_seqpacking_cp_example.pyuvx ruff format --check tests/functional_tests/test_groups/training/test_seqpacking_cp_example.pyNote:
uv run ruff ...is blocked on this local host by the pinnednvidia-resiliency-extwheel platform tag, so I used isolateduvx rufffor the file-scoped lint checks. GPU validation should come from CI.