Reduce variance of classification references evaluation #4609
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
NicolasHug
commented
Oct 13, 2021
references/classification/train.py
Outdated
| # See FIXME above | ||
| num_processed_samples = utils.reduce_across_processes(num_processed_samples) | ||
| if hasattr(data_loader.dataset, "__len__") and len(data_loader.dataset) != num_processed_samples: | ||
| warnings.warn( |
There was a problem hiding this comment.
We might want to only warn is rank == 0 to avoid having world_size warnings? It's not really a problem when running with submitit, but it can be a bit too much when running with torchrun directly
There was a problem hiding this comment.
I agree with warning only when rank == 0.
| @@ -277,6 +297,10 @@ def main(args): | |||
| model_ema.load_state_dict(checkpoint["model_ema"]) | |||
|
|
|||
| if args.test_only: | |||
| # We disable the cudnn benchmarking because it can noticeably affect the accuracy | |||
| torch.backends.cudnn.benchmark = False | |||
| torch.backends.cudnn.deterministic = True | |||
There was a problem hiding this comment.
I deliberately choose not to set torch.use_deterministic_algorithms(True) here because:
- just removing the cudnn bencharmking is enough to get constant results, at least for resnet18 (maybe not for others)
- using
torch.use_deterministic_algorithms(True)forces the user to set some env variables likeCUBLAS_WORKSPACE_CONFIG=:4096:8, otherwise the script would crash. - Users can always set the new --use-deterministic-algorithms flag if they really want to
datumbox
approved these changes
Oct 13, 2021
references/classification/train.py
Outdated
| # See FIXME above | ||
| num_processed_samples = utils.reduce_across_processes(num_processed_samples) | ||
| if hasattr(data_loader.dataset, "__len__") and len(data_loader.dataset) != num_processed_samples: | ||
| warnings.warn( |
There was a problem hiding this comment.
I agree with warning only when rank == 0.
|
Thanks for the review!
Yes, I just wanted to test the waters with this one first :) |
facebook-github-bot
pushed a commit
that referenced
this pull request
Oct 14, 2021
Reviewed By: fmassa Differential Revision: D31649961 fbshipit-source-id: 34d72840adcc84a50d8b96c2cf4776e5450343e8
mszhanyi
pushed a commit
to mszhanyi/vision
that referenced
this pull request
Oct 19, 2021
cyyever
pushed a commit
to cyyever/vision
that referenced
this pull request
Nov 16, 2021
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.
Closes #4559
Closes #4600
This is a follow up to #4559 (comment). This PR does a few things:
--use-deterministic-algorithmsflag to the scriptslen(dataset).Out of these, I think only the first one is really important. I don't feel strongly about the rest, LMK what you think.
cc @datumbox