The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
DebateQA Benchmark Dataset
Official Dataset for "Pretraining on the Test Set Is No Longer All You Need: A Debate-Driven Approach to QA Benchmarks" (COLM 2025)
Paper: [Link to arXiv will be added here once available]
Dataset Description
This repository will host the official dataset artifacts for our COLM 2025 paper. This includes:
- The 50 MMLU-Pro questions used in our experiments.
- The complete logs from our 11-model reference tournament, comprising over 11,000 debate transcripts.
The full dataset files are being prepared and will be uploaded here soon. Thank you for your interest!
- Downloads last month
- 2