Add mHC functional CI coverage (stacks on #4483)#4530
Closed
Connor-XY wants to merge 10 commits intoNVIDIA:dsv4from
Closed
Add mHC functional CI coverage (stacks on #4483)#4530Connor-XY wants to merge 10 commits intoNVIDIA:dsv4from
Connor-XY wants to merge 10 commits intoNVIDIA:dsv4from
Conversation
Adds a functional test recipe exercising mHC end-to-end on H100 with TP=2 / PP=2: golden-values comparison against a dev_dgx_h100 baseline, model config, and a wired-up entry in `tests/test_utils/recipes/h100/gpt.yaml`. Stacks on NVIDIA#4483 (transformer mHC reference impl). The recipe sets `enable_hyper_connections=True` and exercises the n-stream P2P shape path added in NVIDIA#4528 alongside the transformer / hybrid integration. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Adds an end-to-end mHC functional test recipe on H100 with TP=2 / PP=2.
Carved out of PR #4469 so the CI / recipe addition can be reviewed by
@NVIDIA/ciseparately from the source changes.This PR's content (the only files this PR is asking to add):
tests/functional_tests/test_cases/gpt/gpt3_mcore_te_tp2_pp2_mhc/model_config.yaml(new): mHC test config (enable_hyper_connections=True,num_residual_streams=4).tests/functional_tests/test_cases/gpt/gpt3_mcore_te_tp2_pp2_mhc/golden_values_dev_dgx_h100.json(new): golden values for parity comparison ondev_dgx_h100.tests/test_utils/recipes/h100/gpt.yaml: wires the new test case into the H100 GPT recipe.Stacking note
This branch is built on top of #4483 so the recipe's test infra resolves
enable_hyper_connectionsfromTransformerConfig. Realistically the recipe also exercises code from #4527 (fused kernels), #4528 (PP), and #4529 (HybridModel) once those merge —model_config.yamldoesn't import them but the integration test will only pass against the full mHC stack. Cannot merge until #4483 (and ideally the sibling PRs) lands."Files changed" shows #4483's content as ancestry — please review only the three files listed above.
Origin
Carved out of PR #4469 at commit
e3d0102ad. The recipe and golden values were stable across the strict-review loop on the parent PR.🤖 Generated by Claude Opus 4.7 (1M context).