durgeshbhirud/MATH

--- pretty_name: LLaDA Math Responses license: apache-2.0 language: - en task_categories: - text-generation size_categories: - 1K<n<10K --- # Dataset Card: MATH This dataset is converted from model generation outputs. ## Split - `train`: 7500 rows ## Data Fields - `solution`: Ground-truth answer from evaluation file - `question`: Input question - `cot_type`: `math` - `source_type`: Source identifier (`lift/d1_eval/eval_results_512_train`) - `metadata`: JSON string with source file, row index, prompt input, and run config - `cot`: `null` - `thinking_trajectories`: List with one reasoning trace - `attempt`: Extracted final answer from generated output ## Source Files Built from: - `math_GSAI-ML-LLaDA-1-5_512_512_0_20260401-120823_generations.json` - `math_GSAI-ML-LLaDA-1-5_512_512_1_20260401-120823_generations.json` - `math_GSAI-ML-LLaDA-1-5_512_512_2_20260401-120822_generations.json` ## Notes This is an automatically converted dataset from model responses. Some examples may contain malformed XML-like tags from generation output.