Fit reporters on input embeddings as a sanity check #209
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Outdated suggestions cannot be applied.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.
I noticed that on some (model, dataset) pairs the VINC and/or logistic regression AUROC is rather high after the very first layer, which seemed moderately implausible to me. It struck me that we can fit reporters to the input embeddings to sanity check our results. The idea is that if the input embedding AUROC is significantly higher than 0.5 there must be something wrong with the code or the prompt templates or both, since you can't classify a statement as true or false only by looking at its very last token and nothing else.