This dataset provides Magpie-annotated metadata for the original Ultrafeedback preference corpus, including task category, difficulty, input quality, language, and independent preference reward scores.
It serves as one of the five foundational components used in the UltraMix curation pipeline.
🏆 This annotated release was created as part of our paper (2026)