miria0/EduFeedback
EduFeedback Alternating dataset example: a single curated multi-turn conversation yields a complete (prompt, chosen, rejected) triplet on its own — the direct early response becomes chosen and a later, less-direct response becomes rejected. Both sides come from the same real dialog, so no synthetic LLM generation is needed to fill in the rejected side. EduFeedback is a synthetically generated, multi-turn conversational preference dataset in an educational tutoring… See the full description on the dataset page: https://huggingface.co/datasets/miria0/EduFeedback.
Frame Alternating Population around real-world dialog scarcity and removing synthetic rejected responses
Add caption to Alternating Population figure
Expand dataset card: headline figure + Alternating Population Strategy details
Add Alternating Population Strategy figure
Initial release of EduFeedback (COALA, ICML 2026): conversations, pairs-single, pairs-alternating
initial commit
