Fix: Refactor BpeTrainer to use chunked pair counting #1805
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
This commit refactors the `BpeTrainer` to address an Out-of-Memory (OOM) bug that occurred when training on large-alphabet corpora. The `count_pairs` function previously loaded all unique token pairs and their counts into a single `AHashMap`, leading to excessive memory usage with large alphabets. This fix introduces a new `count_pairs_chunked` function that processes words in chunks, significantly reducing memory consumption. The `do_train` function has been updated to use this new chunked approach. The library compiles successfully with these changes, and this new strategy should prevent OOM errors during training on large datasets.
|
sorry did not have time to go though it, did you end up finding a fix ? |
|
@ArthurZucker Hi Arthur, this commit was actually a test involving Google Jules in a new version. I'm very sorry for having opened this PR without definitively correcting the problem. This PR is not complete, so I closed it. Sorry for the delay in responding and for any inconvenience. |
|
No worries 🤗 |
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Outdated suggestions cannot be applied.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 commit refactors the
BpeTrainerto address an Out-of-Memory (OOM) bug that occurred when training on large-alphabet corpora.The
count_pairsfunction previously loaded all unique token pairs and their counts into a singleAHashMap, leading to excessive memory usage with large alphabets.This fix introduces a new
count_pairs_chunkedfunction that processes words in chunks, significantly reducing memory consumption. Thedo_trainfunction has been updated to use this new chunked approach.The library compiles successfully with these changes, and this new strategy should prevent OOM errors during training on large datasets.