haowu89/

math-ai-bench-sources-low

Question-level subsets built from haowu89/math-ai-bench-sources-final. Each row keeps one benchmark question and 8 selected reasoning trajectories. Fields:

problem, original_solution, answer, source, index model: mixed quality bucket identifier generated_solutions: 8 selected trajectories count: always 8 context_metadata: token counts, correctness labels, selected source models, and selected trajectory indices

Available configs:

aime25 aime26 apex_2025… See the full description on the dataset page: https://huggingface.co/datasets/haowu89/math-ai-bench-sources-low.