This is a math dataset mixed from four open-source data. It was used to analyze the contamination test on the MATH and contains 1M samples.
question
question from open-source data
solution
the answer corresponding to question
5grams
5-gram list of f"{question} {answer}"
test_question
the most relevant question from MATH
test_solution
the answer corresponding to test_question
test_5grams
5-gram list of… See the full description on the dataset page:
https://huggingface.co/datasets/newsbang/math_benbench_data_leak_analysis.