config datasets help

[TousenKaname]

May I ask if you have a more detailed document, or will you explain the meaning of the fields carefully in the future? I am a little confused about how the template field in this file is used, and why HUMAN and BOT are needed

chatmed_eval_cfg = dict(
    evaluator=dict(type=EMEvaluator))

chatmed_reader_cfg = dict(
    input_columns=['query'], 
    output_column='response'
    )

chatmed_infer_cfg = dict(
    # Prompt 生成配置
    prompt_template=dict(
        type=PromptTemplate,
        template=dict(
                round=[
                        dict(role="HUMAN", prompt='{query}')
                       ]
                    )),
    retriever=dict(type=ZeroRetriever),
    inferencer=dict(type=PPLInferencer)
    )

chatmed_datasets = [
    dict(
        type=ChatMedDataset,
        path='/mnt/petrelfs/wangguoan/git_repo/opencompass/data/ChatMed_Consult_Dataset/test.json',
        reader_cfg=chatmed_reader_cfg,
        infer_cfg=chatmed_infer_cfg,
        eval_cfg=chatmed_eval_cfg
        )
]

below is chatmed data

{"query": "这段时间去上厕所本来想小便的可是每次都会拉大便", "response": "这可能是因为你的饮食习惯或者消化系统的问题导致的。建议你试着调整一下饮食，增加膳食纤维的摄入量，多喝水，避免过度依赖泻药。同时，如果问题持续存在，建议去医院检查一下，排除肠道疾病等可能性。"}

[bittersweet1999]

We set HUMAN and BOT to differentiate the roles in a conversation. In a single-turn dialogue dataset, you only need to use HUMAN. For example:

template = dict(
    round=[
        dict(
            role='HUMAN',
            prompt="{question}"
        ),
    ]
)

In the case of multi-turn dialogues, you might need to specify templates for multiple rounds of HUMAN and BOT interactions. In the chatmed data example you provided, since it's a single-turn dialogue, using HUMAN as the template is sufficient.