{ "split": "train", "input_root": "data_dpo_v2\\train_pref_v1", "total_pairs": 2500, "assembly_counts": { "chosen_rule_fallback": 2500, "rejected_constructed_fallback": 2500 }, "strong_answer_summary": { "label": "strong_answers", "count": 0 }, "weak_answer_summary": { "label": "weak_answers", "count": 0 } }