Multi Subject RLVR is a dataset of 638,000 college-level question-answer pairs derived from ExamQA. The dataset was originally written in Chinese and covers at least 48 first-level subjects, with questions and answers authored by domain experts for examination purposes. It was created by virtuoussy and last updated on April 16, 2025.
Use Cases
- Training question-answering models based on the 638k college-level QA pairs.
- Evaluating model performance across diverse academic subjects based on the coverage of at least 48 first-level subjects.
- Benchmarking reinforcement learning agents on tasks requiring domain expert knowledge based on the examination-purpose content.
Strengths
- Contains 638,000 college-level QA instances, providing substantial volume.
- Questions and answers are written by domain experts for examination purposes, suggesting high-quality content.
- Covers at least 48 first-level subjects, indicating broad domain diversity.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- The dataset was originally in Chinese and has been converted; potential translation artifacts may exist.
- Last updated 2025-04-16 09:47:25; freshness should be verified.
Provenance
- Source
- ExamQA (Yu et al., 2021)
- Collection Method
- Distractors were removed and each instance was converted into a free-form QA pair.
- Freshness
- 2025-04-16 09:47:25