fix: prevent duplicate data when using multiple DataLoader workers#169
Open
AmSach wants to merge 1 commit into
Open
fix: prevent duplicate data when using multiple DataLoader workers#169AmSach wants to merge 1 commit into
AmSach wants to merge 1 commit into
Conversation
When using num_workers > 0 in DataLoader, each worker iterates over ALL batches instead of a distinct subset, causing the model to train on duplicate data multiple times per epoch. Fix: - Dataset now accepts num_workers parameter to partition batches - DataLoader passes num_workers to Dataset and creates worker_init_fn that sets worker_id on the Dataset for each subprocess - Dataset.__iter__ now partitions batches across workers so each worker processes a distinct subset This ensures each worker handles 1/num_workers of the batches, eliminating duplicate training data.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixed the bug described in issue #150.
What was wrong
When using num_workers > 0 in DataLoader, each worker process iterates over ALL batches instead of a distinct subset. This causes the model to train on duplicate data multiple times per epoch (e.g., with 30 workers, each sample is processed 30 times per epoch instead of once).
How I fixed it
Testing
Closes #150