outputs/gpt2cocoSAD/tokenized_data_SAD/dataset_dict.json

{"splits": ["train", "test"]}