sungyub/qa-verl-unified

统一问答VERL数据集是一个包含5个高质量问答和推理数据集的统一集合，已去重并优化用于强化学习训练。数据集包含84,373个唯一问题，涵盖逻辑谜题、表格推理、长文本问答等多种问答和推理任务。数据集采用VERL格式，语言为英语（部分包含中文），遵循Apache 2.0许可证。数据集经过两个阶段的去重处理，保留了最小的数据集以保持问题的多样性。

Unified QA VERL Dataset is a collection of 5 high-quality question-answering and reasoning datasets unified in VERL format, deduplicated and optimized for reinforcement learning training. The dataset contains 84,373 unique problems covering a variety of QA and reasoning tasks such as logic puzzles, table reasoning, long-text QA, etc. The dataset is in VERL format, in English (with some Chinese in docqa-rl), and is licensed under Apache 2.0. The dataset has undergone a two-phase deduplication process to preserve diversity by retaining the smallest datasets.