Improve resumable single-step eval#155
Merged
Merged
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR continues after #149 by improving the robustness of single-step eval, especially in cases when the storage backend does not save things immediately (e.g. mounted remote container).
The most important change is that the current version of resumable evals rewrites the results file upon a restart, which is potentially a very heavy operation, and one that can take a minute or so to get flushed to storage. I saw cases where results from the next couple of batches would be lost because of a race condition between the file rewrite and subsequent appends. This PR proposes to instead truncate the file to remove the potential broken lines instead of fully rewriting it. Moreover, I found that reopening the results file after each batch instead of keeping it open throughout can lead to more frequent flushing to underlying storage. Finally, if running in resumable mode,
all_predictionsandall_back_translation_predictionswere keeping the results in memory unnecessarily, as there is no downstream consumer of that data and it is already stored in the results file itself; this is also cleaned up here.