Operator level microbenchmarking #3154
Open
Conversation
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
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D77676673 |
SSYernar
added a commit
to SSYernar/torchrec
that referenced
this pull request
Jul 9, 2025
Summary: This change introduces microbenchmarking for PyTorch operators. Since we need to capture and measure each operator call (which is happening under the hood of PyTorch), we need to use torch.profiler.profile. Example operators are `aten:mm`, `aten::sigmoid`, `cudaLaunchKernel`, etc… Use `--benchmark_operators` to enable the operator-level benchmarking. Use `--limit_operator_results` argument to specify the number of top runtime operators to benchmark. Use `--target_operators` argument to list PyTorch operators to benchmark. Example output: ``` TrainPipelineSparseDist | Malloc retries (P50/P90/P100): 0.0 / 0.0 / 0.0 | Runtime (P90): 442.08 ms | Peak Memory alloc (P90): 24.23 GB | Peak Memory reserved (P90): 26.21 GB operator_aten::copy_ | Malloc retries (P50/P90/P100): -1.0 / -1.0 / -1.0 | Runtime (P90): 39.21 ms | Peak Memory alloc (P90): 0.00 GB | Peak Memory reserved (P90): -0.00 GB ... ``` Differential Revision: D77676673
SSYernar
force-pushed
the
export-D77676673
branch
from
da06c9d to
f67ee85
Compare
Summary: Pull Request resolved: pytorch#3154 This change introduces microbenchmarking for PyTorch operators. Since we need to capture and measure each operator call (which is happening under the hood of PyTorch), we need to use torch.profiler.profile. Example operators are `aten:mm`, `aten::sigmoid`, `cudaLaunchKernel`, etc… Use `--benchmark_operators` to enable the operator-level benchmarking. Use `--limit_operator_results` argument to specify the number of top runtime operators to benchmark. Use `--target_operators` argument to list PyTorch operators to benchmark. Example output: ``` TrainPipelineSparseDist | Malloc retries (P50/P90/P100): 0.0 / 0.0 / 0.0 | Runtime (P90): 442.08 ms | Peak Memory alloc (P90): 24.23 GB | Peak Memory reserved (P90): 26.21 GB operator_aten::copy_ | Malloc retries (P50/P90/P100): -1.0 / -1.0 / -1.0 | Runtime (P90): 39.21 ms | Peak Memory alloc (P90): 0.00 GB | Peak Memory reserved (P90): -0.00 GB ... ``` Differential Revision: D77676673
|
This pull request was exported from Phabricator. Differential Revision: D77676673 |
SSYernar
force-pushed
the
export-D77676673
branch
from
f67ee85 to
484444e
Compare
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:
This change introduces microbenchmarking for PyTorch operators.
Since we need to capture and measure each operator call (which is happening under the hood of PyTorch), we need to use torch.profiler.profile. Example operators are
aten:mm,aten::sigmoid,cudaLaunchKernel, etc…Use
--benchmark_operatorsto enable the operator-level benchmarking.Use
--limit_operator_resultsargument to specify the number of top runtime operators to benchmark.Use
--target_operatorsargument to list PyTorch operators to benchmark.Example output:
Differential Revision: D77676673