llm_trainer/data/dataset_info.json

{
  "sample_instruct": {
    "file_name": "sample_aq.json",
    "columns": {
      "prompt": "instruction",
      "response": "response"
    }
  }
}