Fix sampling, add timeouts for test suprocess and data loaders #221
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
jlamypoirier
requested review from
tscholak and
sohamparikh
and removed request for
tscholak
sohamparikh
reviewed
Apr 3, 2025
fast_llm/data/dataset/gpt/sampled.py
Outdated
| if unshuffled_tokens := data.get("unshuffled_tokens") is not None: | ||
| self._unshuffled_tokens = unshuffled_tokens | ||
| else: | ||
| self._unshuffled_tokens = data["unshuffled_epochs"] * data["dataset"]["tokens_per_epoch"] | ||
| if "unshuffled_tokens" in data: | ||
| self._unshuffled_tokens = data["unshuffled_tokens"] |
There was a problem hiding this comment.
should we care about backwards compatibility?
We could throw an error if unshuffled_tokens is not present in loaded_yaml_data. And speaking of backwards compatibility, I think we should also add truncate_documents: True in loaded_yaml_data if not present, else it will fail loading previous datasets.
There was a problem hiding this comment.
The if was because of the method being called before "unshuffled_tokens" is added to data, I reorganized things so it's not needed.
We don't really care about backward compatibility for dataset cache, but I added a simple one.
sohamparikh
approved these changes
Apr 3, 2025
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Outdated suggestions cannot be applied.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.
✨ Description
#186 Broke sampling because of an incorrect walrus which always set
unshuffled_tokensto 1 (True). I fixed it and simplified related code.This bug only showed in slow tests (@sohamparikh please make sure these pass before merging) which didn't terminate because of a data loader crash somehow not being caught. I added some timeouts to ensure this doesn't happen again.
🔍 Type of change
Select all that apply: