This is a preference dataset of 80k samples generated for the paper ActiveUltraFeedback: Efficient Preference Data Generation using Active Learning (Melikidze et al., 2026).
The prompts are from Skywork Reward Preference 80k v0.2 (Liu et al., 2024). The response pairs were generated with the ActiveUltraFeedback pipeline, which calls a large pool of open-weight LLMs to first generate candidate responses, then uses various active selection strategies… See the full description on the dataset page:
https://huggingface.co/datasets/ActiveUltraFeedback/skywork.
