dynamic scoring for LM #188
Open
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
vince62s
reviewed
Jan 22, 2025
vince62s
reviewed
Jan 22, 2025
| kwargs = {} | ||
| string = string.replace(DefaultTokens.SEP, "\n").replace( | ||
| DefaultTokens.MASK_BEFORE, self.tokenizers[side].pad_token | ||
| ) |
Contributor
There was a problem hiding this comment.
shouldn't we handle it the same way for other tokenizers ?
Contributor
Author
There was a problem hiding this comment.
I have only tested for that one (with eurollm) at this point; an error will be raised with the others.
vince62s
reviewed
Jan 22, 2025
vince62s
reviewed
Jan 22, 2025
vince62s
reviewed
Jan 22, 2025
vince62s
reviewed
Jan 22, 2025
vince62s
reviewed
Jan 22, 2025
Member
francoishernandez
left a comment
francoishernandez
left a comment
There was a problem hiding this comment.
It seems a bit weird to have such very specific changes in scoring_utils (response_patterns handling mostly), but I'm not sure how this can be properly factorized with the current structure. Let's keep it for now and we might reconsider later.
Comment on lines
+80
to
+103
| if is_seq2seq: | ||
| predictor = Translator.from_config( # we need to review opt/config stuff in translator | ||
| model, | ||
| self.vocabs, | ||
| predict_config, | ||
| model_config, | ||
| device_id=gpu_rank, | ||
| global_scorer=scorer, | ||
| report_align=predict_config.report_align, | ||
| report_score=False, | ||
| logger=None, | ||
| ) | ||
| else: | ||
| predictor = GeneratorLM.from_config( | ||
| model, | ||
| self.vocabs, | ||
| predict_config, | ||
| model_config, | ||
| device_id=gpu_rank, | ||
| global_scorer=scorer, | ||
| report_align=predict_config.report_align, | ||
| report_score=False, | ||
| logger=None, | ||
| ) |
Member
There was a problem hiding this comment.
Maybe cleaner to just define a predictor_class in the condition, and call predictor_class(*) once, since they should have the same signature.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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 PRs adapts the ScoringPreparator for LM architecture and fixes the ignore_prompt method of the loss for LM validation with left padding.
We noticed that the
filtertoolongtransform should not be used along with thehuggingface_tokenizetransform. See#191