MPB_Missing-Premise-Benchmark

简介

HRB: Missing-Premise Reasoning Benchmark HRB is the Hugging Face release of the Missing-Premise Benchmark (MPB) introduced in Ask, Condition or Abstain: Reinforcement Learning for Missing-Premise Reasoning. The benchmark contains 274 human-verified instances designed to evaluate whether a language model can recognize when a problem is underdetermined and respond constructively—by asking for the missing premise, conditioning its answer on an unknown variable, or… See the full description on the dataset page: https://huggingface.co/datasets/ant-intl/MPB_Missing-Premise-Benchmark.