ctrl: Mix Alpaca + Dolly + Beavertails datasets rather than only using Alpaca#111
Merged
ctrl: Mix Alpaca + Dolly + Beavertails datasets rather than only using Alpaca#111
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Changes
ablation_uncurated.py: Investigation of why Llama-3-8B CTRL has such MMLU-Pro score (0.08). Turns out just fine-tuning Llama-3-8B on CTRL's original choice of datasets, without CTRL's special curation/rewriting of the dataset, also causes the same drop in MMLU-Pro score, which comes from the model always repeating the reasoning + answer from one of the few-shot examples in the MMLU-Pro query. So CTRL is maybe just broken — though we could still run our other non-MMLU-Pro evals on it, maybe it fares better in zero-shot settingsctrl.py: I noticed our implementation only uses the Alpaca dataset rather than the mix of Alpaca + Dolly + Beavertails that the original paper uses. I've changed it to use the mix since I wondered if it would fix the poor MMLU-Pro score (it doesn't)