Avoid using skip() in hf_datasets #835
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
This was referenced Feb 12, 2025
Merged
facebook-github-bot
added
the
CLA Signed
tianyu-l
reviewed
Feb 12, 2025
| if isinstance(self._data, Dataset) and self._sample_idx == len(self._data): | ||
| return iter([]) | ||
|
|
||
| return iter(self._data.skip(self._sample_idx)) |
There was a problem hiding this comment.
two questions:
- The wrong behavior is not captured by unit test /~https://github.com/pytorch/torchtitan/blob/main/tests/unit_tests/test_dataset_checkpointing.py. Is it because the error only happens in distributed environment, on some rank but not others?
- Does skip fail for both map-style datasets (
c4_test) and IterableDatasets (c4)? Asking because from my PR uniformly use skip for both (map-style) Dataset and IterableDataset #521 I recallskipfor IterableDataset is a "new" feature.
There was a problem hiding this comment.
I am working on testing loss issue(whether still happens) and whether difference between datasets c4/c4_test right now.
unit_test would be the next PR after fix.
There was a problem hiding this comment.
- It only happens in distributed environment and on certain ranks only.
- I could not reproduce this with
c4_test.
I don't know the detail of the root cause though.
There was a problem hiding this comment.
I would not land this PR but I verified all ranks output the expected sample even after resuming from checkpoints, @mori360 would work on loss parity with unittest verifcation.
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Outdated suggestions cannot be applied.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.
Stack from ghstack (oldest at bottom):
We found out that skip() may not behave as expected. This is a workaround solution while we are investiging the root cause.