Fix wrong world vector range in training loss calculation#84
Open
jonathansalzer wants to merge 1 commit intogoogle-deepmind:mainfrom
Open
Fix wrong world vector range in training loss calculation#84jonathansalzer wants to merge 1 commit intogoogle-deepmind:mainfrom
jonathansalzer wants to merge 1 commit intogoogle-deepmind:mainfrom
Conversation
AoqunJin
reviewed
Jan 1, 2025
There was a problem hiding this comment.
I found this due to the poor performance of the trained model when tested. But it's solved now, took days to debug 😂😂
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Throughout training and inference, a world vector range of (-2; 2) is consistently used. However, during loss calculation, the world vector range is not explicitly passed to the action tokenization function, causing it to default to a range of (-1; 1). This mismatch leads to scaling issues: values are doubled, as well as clipped if they exceed the range (-1; 1).
By passing the model world vector range to the action tokenization, this issue is resolved.