Medical Triage Alignment (MTA) dataset use case#16
Open
barry-ravichandran wants to merge 6 commits into
Open
Conversation
Add evaluation use case for the MTA (Medical Triage Alignment) dataset. Supports baseline (neutral) and attribute-aligned system prompts using chat template format (system + user roles). Includes deterministic answer shuffling for position bias measurement. System prompts sourced from ITM-Kitware/align-system. Signed-off-by: Bharadwaj Ravichandran (Barry) <barry.ravichandran@kitware.com>
Add shared utilities for the Medical Triage Alignment dataset: - download_raw(): fetch NAACL24 data from ITM-Kitware/align-system - convert_to_eval(): convert to evaluation format using highest KDMA score as ground truth - convert_to_semantic_pairs(): build contrastive training pairs - augment_semantic(): augment training data (7 types x 2 orderings) Signed-off-by: Bharadwaj Ravichandran (Barry) <barry.ravichandran@kitware.com>
Add notebook comparing zero-shot baseline vs attribute-aligned prompts across 6 MTA ethical attributes. Fetches prompts directly from ITM-Kitware/align-system via AST parsing. Runs 10 stochastic trials per condition to report mean +/- std accuracy. Signed-off-by: Bharadwaj Ravichandran (Barry) <barry.ravichandran@kitware.com>
Return {"mean": 0.0, "std": 0.0} when all responses are None instead
of raising ZeroDivisionError. This occurs when the model outputs
unparseable text at extreme steering strengths.
Signed-off-by: Bharadwaj Ravichandran (Barry) <barry.ravichandran@kitware.com>
Allow mergekit>=0.0.5.1 instead of pinning to exactly 0.0.5.1, which caused dependency resolution failures during pip install. Signed-off-by: Bharadwaj Ravichandran (Barry) <barry.ravichandran@kitware.com>
barry-ravichandran
force-pushed
the
feat/medical-triage-prompt-baseline
branch
from
36d2856 to
48e3dc9
Compare
Collaborator
|
Thanks @barry-ravichandran I'll take a look! |
Signed-off-by: Bharadwaj Ravichandran (Barry) <barry.ravichandran@kitware.com>
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.
This PR adds in the use case for the Medical Triage Alignment dataset that focuses on multiple choice decision-making in hig stakes medical triage scenarios.
Related paper: https://aclanthology.org/2024.naacl-industry.18/
Below is a summary of the feature additions and misc fixes: