This dataset is converted from Open-Reasoner-Zero's math dataset.
Check out
https://github.com/allenai/open-instruct/blob/main/scripts/data/rlvr/open_reasoner.py for the conversion script.
The dataset contains math problems and their solutions in a conversational format:
messages: List of message dictionaries with user questions and assistant answers
ground_truth: The correct solution for each problem
dataset: Always "math" to… See the full description on the dataset page:
https://huggingface.co/datasets/vwxyzjn/rlvr_orz_math_57k_collected.