input: ./dataset/dapo-math-17k.parquet
tokenizer_name_or_path: ./model_from_hf/qwen25-7b
output_prefix: ./dataset/data
handler_name: Math17kAlpacaStyleInstructionHandler
tokenizer_type: HuggingFaceTokenizer
workers: 8
log_interval: 1000
prompt_type: empty
dataset_additional_keys: [labels]
map_keys:  {"prompt":"prompt", "query":"", "response": "reward_model", "system":""}