Fix multi-copy benchmark OSError and update libc10.so path #310
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
added 2 commits
November 19, 2025 08:58
Running benchmarks with a large number of copies can trigger:
OSError: [Errno 24] Too many open files.
Example command:
(fbgemm_gpu_env)$ ulimit -n 1048576
(fbgemm_gpu_env)$ python ./bench/tbe/tbe_inference_benchmark.py nbit-cpu
\
--num-embeddings=40000000 --bag-size=2 --embedding-dim=96 \
--batch-size=162 --num-tables=8 --weights-precision=int4 \
--output-dtype=fp32 --copies=96 --iters=30000
Cherry-pick the latest upstream commit from the FBGEMM main branch to
fix this issue.
This patch enables switching to the file_system sharing strategy via:
export PYTORCH_SHARE_STRATEGY='file_system'
A recent change to the installation location of libc10.so causes the following warning and breaks the functionality introduced in facebookresearch#290: [WARN] Linux + aarch64 detected and target binary is tbe_inference_benchmark, but libc10.so not found at: DCPerf/benchmarks/build/miniconda/envs/fbgemm_build_oss_env/lib/python3.13/site-packages/torch/lib/libc10.so [WARN] Running without LD_PRELOAD This patch updates the expected path to the correct libc10.so location.
meta-cla
bot
added
the
CLA Signed
|
@YifanYuan3 has imported this pull request. If you are a Meta employee, you can view this in D87577418. |
meta-codesync bot
pushed a commit
that referenced
this pull request
Nov 20, 2025
Summary: This PR includes two fixes: 1. Fix OSError 24 (“Too many open files”) in the multi-copy benchmark by cherry-picking the latest upstream FBGEMM patch. Enables switching to the file_system sharing strategy via: export PYTORCH_SHARE_STRATEGY='file_system'. 2. Fix missing libc10.so caused by a recent path change, which breaks functionality from PR #290. Updates the expected path to the correct library location. Pull Request resolved: #310 Reviewed By: charles-typ Differential Revision: D87577418 Pulled By: YifanYuan3 fbshipit-source-id: 768acca7dcfeba552aa103b125d9ac8cbd1fd68e
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
This PR includes two fixes:
Fix OSError 24 (“Too many open files”) in the multi-copy benchmark by cherry-picking the latest upstream FBGEMM patch.
Enables switching to the file_system sharing strategy via:
export PYTORCH_SHARE_STRATEGY='file_system'.
Fix missing libc10.so caused by a recent path change, which breaks functionality from PR Fix incorrect SVE2 detection in FBGEMM benchmark #290.
Updates the expected path to the correct library location.