Repository navigation
test(fsdp): verify optimizer CUDA graphs in v2 - #6498
Merged
Merged
Conversation
|
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. |
wujingyue
changed the title
wujingyue
marked this pull request as ready for review
August 12, 2026 23:28
Phlip79
approved these changes
Aug 12, 2026
skyw
approved these changes
Aug 13, 2026
jaredcasper
approved these changes
Aug 13, 2026
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
wujingyue
force-pushed
the
optimizer-cuda-graph-mfsdp
branch
from
August 13, 2026 16:58
37bc223 to
c98aecd
Compare
wujingyue
enabled auto-merge
August 13, 2026 17:04
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/31737737376 |
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/31745334721 |
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/31746289164 |
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/31752937667 |
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/31755638967 |
wujingyue
added a commit
to wujingyue/Megatron-Bridge
that referenced
this pull request
Sep 1, 2026
…ware optimizer, HSDP, and optimizer CUDA graphs Megatron-LM landed dedicated MFSDP v2 support for these, each with its own commits and tests, after this validation function was written: - clip_grad: MFSDP v2 gradient clipping across dense and expert meshes (NVIDIA/Megatron-LM#6489) - use_precision_aware_optimizer: mixed-precision optimizers in MFSDP v2 (NVIDIA/Megatron-LM#6506) - num_distributed_optimizer_instances / outer_dp_sharding_strategy (HSDP): expose HSDP and HFSDP through the MFSDP v2 adapter (NVIDIA/Megatron-LM#6873); the adapter itself now enforces the correct paired constraint, so no replacement check is needed here - optimizer_cuda_graph: verified for MFSDP v2 (NVIDIA/Megatron-LM#6498) Verified end-to-end against Megatron-LM main + Megatron-Bridge with the DeepSeek-V3 MFSDP v2 + EP proxy config (clip_grad=1.0): 20/20 iterations, finite decreasing loss, no skipped or NaN iterations.
2 of 5 tasks
wujingyue
added a commit
to wujingyue/Megatron-Bridge
that referenced
this pull request
Sep 1, 2026
…ware optimizer, HSDP, and optimizer CUDA graphs Megatron-LM landed dedicated MFSDP v2 support for these, each with its own commits and tests, after this validation function was written: - clip_grad: MFSDP v2 gradient clipping across dense and expert meshes (NVIDIA/Megatron-LM#6489) - use_precision_aware_optimizer: mixed-precision optimizers in MFSDP v2 (NVIDIA/Megatron-LM#6506) - num_distributed_optimizer_instances / outer_dp_sharding_strategy (HSDP): expose HSDP and HFSDP through the MFSDP v2 adapter (NVIDIA/Megatron-LM#6873); the adapter itself now enforces the correct paired constraint, so no replacement check is needed here - optimizer_cuda_graph: verified for MFSDP v2 (NVIDIA/Megatron-LM#6498) Verified end-to-end against Megatron-LM main + Megatron-Bridge with the DeepSeek-V3 MFSDP v2 + EP proxy config (clip_grad=1.0): 20/20 iterations, finite decreasing loss, no skipped or NaN iterations. Signed-off-by: Jingyue Wu <wujingyue@gmail.com>
ksivaman
pushed a commit
to ksivaman/Megatron-LM
that referenced
this pull request
Sep 1, 2026
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
wujingyue
added a commit
to wujingyue/Megatron-Bridge
that referenced
this pull request
Sep 11, 2026
…ware optimizer, HSDP, and optimizer CUDA graphs Megatron-LM landed dedicated MFSDP v2 support for these, each with its own commits and tests, after this validation function was written: - clip_grad: MFSDP v2 gradient clipping across dense and expert meshes (NVIDIA/Megatron-LM#6489) - use_precision_aware_optimizer: mixed-precision optimizers in MFSDP v2 (NVIDIA/Megatron-LM#6506) - num_distributed_optimizer_instances / outer_dp_sharding_strategy (HSDP): expose HSDP and HFSDP through the MFSDP v2 adapter (NVIDIA/Megatron-LM#6873); the adapter itself now enforces the correct paired constraint, so no replacement check is needed here - optimizer_cuda_graph: verified for MFSDP v2 (NVIDIA/Megatron-LM#6498) Verified end-to-end against Megatron-LM main + Megatron-Bridge with the DeepSeek-V3 MFSDP v2 + EP proxy config (clip_grad=1.0): 20/20 iterations, finite decreasing loss, no skipped or NaN iterations. Signed-off-by: Jingyue Wu <wujingyue@gmail.com>
This branch was previously deployed
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.
Summary
Enable
optimizer_cuda_graphfor MFSDP v2 by removing its configuration guard.Extend the MCore adapter optimizer integration test to exercise eager execution plus optimizer-step CUDA-graph warmup, capture, and replay against the existing unsharded reference.
Capturable Transformer Engine FusedAdam requires main gradients to use the configured main-weight dtype. The CUDA-graph test derives that dtype from the MFSDP configuration; the eager variant retains its default bf16 main gradients. See NVIDIA/TransformerEngine#3358.
Validation
python -m torch.distributed.run --nproc-per-node 2 -m pytest -q tests/unit_tests/distributed/mfsdp_v2/test_mcore_adapter.py::TestMcoreAdapter::test_build_train_and_steppython -m black --check tests/unit_tests/distributed/mfsdp_v2/test_mcore_adapter.pypython -m isort --check-only tests/unit_tests/distributed/mfsdp_v2/test_mcore_adapter.py