Fixing leaking autograd graph memory in Trainer - #292
Merged
Merged
Conversation
… trainer process. Two losses are stored without .detach() being called, the self.best_devloss = output[self.eval_metric] is the big one, and self.loss_history[train].append(mean_loss) is the smaller
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Trainer has two memory leaks, both from the same source, we're not detaching the grad graphs from the metrics before we store them. This fixes both of those, as well as including a benchmark so you can see the difference. Check out the second commit, run the benchmark, then check out the third commit and run the benchmark.
No longer storing hundreds of MB or a few GB of autograd graph on accident per long training run!