swadeshb/

openmath-cot-100k-filtered

Filtered from nvidia/OpenMathReasoning split cot in streaming mode. Filters:

problem_type == has_answer_extracted: True excluded problem_source == MATH_training_set: True max problem chars: 2500 min/max solution chars: 200/240000 difficulty mix: 0.3,0.4,0.3 using pass_rate_72b_tir

Counts: { "hard": 30000, "medium": 28296, "easy": 30000 }

This dataset is intended for student-native reasoning-trace compression experiments.