Repository navigation
Reuse profiler helpers in mFSDP v2 symmetric memory tests - #5873
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. |
cbf1c5a to
1383435
Compare
9153657 to
fb5bf82
Compare
1383435 to
a16d05a
Compare
fb5bf82 to
be3cbc6
Compare
a16d05a to
67a5adf
Compare
be3cbc6 to
949a9df
Compare
67a5adf to
af677a8
Compare
adab7f6 to
46b3e02
Compare
8a049c8 to
85bc135
Compare
85bc135 to
4a06492
Compare
4a06492 to
1b1a649
Compare
|
/ok to test |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/29920736154 |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/29935237896 |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/29960716719 |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/29968252116 |
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com> Signed-off-by: Dmytro Pykhtar <dpykhtar@nvidia.com>
What changed
test_symmetric_memory.pyto profile CPU+CUDA events and reusecollect_linked_kernelsfor all-gather and reduce-scatter kernel attribution.ncclSymkdirectly on the linked collective kernel names instead of keeping local profiler-name/count helpers.Why
This keeps the mFSDP v2 profiler parsing logic in one helper path after #5859 added linked CPU-op attribution for profiler kernels.
Impact
Test-only refactor. There is no production-code behavior change.
Stacking
Stacked on #5859 with base branch
pull-request/5859.Validation
uv run --no-sync isort tests/unit_tests/distributed/mfsdp_v2/profiler_utils.py tests/unit_tests/distributed/mfsdp_v2/test_symmetric_memory.pyuv run --no-sync python -m compileall tests/unit_tests/distributed/mfsdp_v2/profiler_utils.py tests/unit_tests/distributed/mfsdp_v2/test_symmetric_memory.pygit diff --checkuv run --no-sync python -m pytest --collect-only -q tests/unit_tests/distributed/mfsdp_v2/test_symmetric_memory.pyFull distributed CUDA execution is left to CI because this refactor depends on linked profiler correlations from the CI PyTorch profiler build.