[Bugfix] Avoid truncating the outputs based on string lengths #201
Conversation
This file contains 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
anton-l
commented
Jun 11, 2024
| hlog_warn( | ||
| f"The smallest context of your batch ({smallest_context}) is bigger than the maximum context size allowed by the model ({self.max_length}) for a task in" | ||
| f"The context size of your batch ({context_size}) is bigger than the maximum context size allowed by the model ({self.max_length}) for a task in" |
There was a problem hiding this comment.
Replaced smallest_context here because the real issues are caused by biggest_context (renamed to context_size)
clefourrier
reviewed
Jun 12, 2024
clefourrier
reviewed
Jun 12, 2024
There was a problem hiding this comment.
It overall LGTM, I'd welcome a second look by @NathanHB
Side note - we need to add a long context eval to the test suite, though gpt2 has so small a context that anything is long context I guess XD
clefourrier
approved these changes
Jul 3, 2024
hynky1999
pushed a commit
to hynky1999/lighteval
that referenced
this pull request
Jul 12, 2024
…gface#201) * Fix context size * - redundant condition --------- Co-authored-by: Clémentine Fourrier <[email protected]>
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Outdated suggestions cannot be applied.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.
This fixes a bug where
max_new_tokenswas set to1due tobiggest_context = max(len(c) for c in context)instead ofbiggest_context = <max number of tokens>.This meant that a 4-shot gsm8k eval was generating single-token predictions if the model's context size is
2048.To reproduce:
And check for single-token predictions in
$OUTPUT_DIR/details/HuggingFaceFW/ablation-model-fineweb-edu/2024-06-11T15-34-10.689447/details_leaderboard|gsm8k|4_2024-06-11T15-34-10.689447.parquet