yolov3: reduce batch size due to cudagraphs OOM #2008
Closed
Conversation
This file contains 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
xmfan
had a problem deploying
to
docker-s3-upload
GitHub Actions
Error
xmfan
had a problem deploying
to
docker-s3-upload
GitHub Actions
Error
xmfan
force-pushed
the
xmfan/yolov3_reduce_bs
branch
from
fe396c3
to
c5fb66e
Compare
xmfan
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
xmfan
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
xuzhao9
reviewed
Oct 24, 2023
There was a problem hiding this comment.
We got to be careful here because the original code repo uses 16 as the default batch size (https://github.com/ultralytics/yolov3/blob/master/train.py#L449), and we don't know how bs=8 will affect the E2E training accuracy.
I am leaning to keep the default batch size to be consistent with upstream, and only customize it in the dynamo runner or other userbenchmarks that test CUDAGraph.
|
Moved the implementation to pytorch/benchmark/dynamo/torchbench.py: pytorch/pytorch#111959 |
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Outdated suggestions cannot be applied.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.
yolov3 w/ cudagraphs (known to use more memory) is failing perf test due to OOM (https://hud.pytorch.org/benchmark/torchbench/inductor_with_cudagraphs?startTime=Mon,%2016%20Oct%202023%2020:19:47%20GMT&stopTime=Mon,%2023%20Oct%202023%2020:19:47%20GMT&granularity=hour&mode=training&dtype=amp&lBranch=main&lCommit=0b424ee0b7bfe09e0a438a63e8336e95eea85901&rBranch=main&rCommit=29048be41ca3aa8974795d93b9ea9fd6dee415fc)
I'm reducing the batch size from 16 to 8 to keep the same batch size for all yolov3 HUD benchmarks