language:
en
license: mit
task_categories:
text-generation
pretty_name: project14-dpo
tags:
medical
triage
project14
openclassrooms
dpo
preference-learning
rlhf---
Direct Preference Optimisation dataset for a medical triage agent. ~1,000 prompt/chosen/rejected pairs sourced from UltraMedical-Preference. Human-annotated pairs prioritised during undersampling. PII anonymised with Presidio (RGPD compliant).
