Persist IterableDataset epoch in workers #6710
Open
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Outdated suggestions cannot be applied.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.
Use shared memory for the IterableDataset epoch.
This way calling
ds.set_epoch()
in the main process will update the epoch in the DataLoader workers as well.This is useful especially because the epoch is used to compute the
effective_seed
used for shuffling.I used torch's shared memory in case users want to send dataset copies without shared memory using pickle. I also find it easier to use than using
multiprocessing.shared_memory
than requires unlinking only in the main process, ormp.Value
that is not picklable.close #6673
cc @rwightman