- data_name: your_dataset_name  # Data set name. This field is associated with the name entered on the evaluation set configuration page.
  metrics:  # Configuration for one or more metrics included in this data set.
      # name: metric name.
    - name: judge model scoring.  
      # The pipeline field is used to define the scoring process, where each element in this array corresponds to a processing node in sequence.
      pipeline:
        # The type field is set to PREPROCESS, which indicates that this node is a preprocessing node.
        - type: PREPROCESS
          # filename specifies the name of the Python script on which preprocessing or postprocessing depends.
          # This field should be filled with the relative path of the corresponding file to the root directory within the compressed package.
          filename: scripts/preprocessor.py
          # file_content: For preprocessing or postprocessing steps, you can directly enter the script content in this field.
          file_content: |-
            def preprocess(data, resp, **kwargs) -> bool | int | str | float | None:
                pass
        
        # type is JUDGE_MODEL, which indicates that this node is a scoring node based on a judge model.                                     
        - type: JUDGE_MODEL
          # Configure the judge model information by specifying the judge_model field.
          judge_model:
            name: DeepSeek-R1  # Name of a judge model.
            # Method for calling a judge model:
            #   If MS is entered, it indicates using the Online Services module of Tencent Cloud TI-ONE Platform (TI-ONE).
            #   If a URL is entered, it indicates using a third-party URL.
            source: MS
            # ti_model_service_api: This field is required if the Online Services module of TI-ONE is used.
            ti_model_service_api:  
              # service_group_id: ID of an online service, for example, ms-45mrs4rv.
              service_group_id: ms-xxxx
              # service_group_name: Name of an online service.
              service_group_name: "DeepSeek-R1 judge model"  
              # service_id: ID of an online service. You can choose Online Service Details > Service Management to view the ID of a version, for example, ms-45mrs4rv-1.
              # If this field is left unspecified, service_group_id with the suffix -1 will be generated by default.
              service_id: ms-xxxx-1
              # url_prefix: address of Regular Service Calling. You can choose Online Service Details > Service Call > Regular Service Calling to view it.
              url_prefix: http://ms-xxxx-uuuu-sw.gw.ap-region.ti.tencentcs.com/ms-xxxx
              # path_suffix: path for the calling chat API exposed by an online service.
              path_suffix: /v1/chat/completions
              # auth_token: authentication token. You can choose Service Authentication > Online Service Details to view it.
              auth_token: "aaaaaaaa"
            # third_party_api: If URL is specified for calling, this field is required.
            third_party_api:
              # url: calling URL.
              url: http://ms-xxxx-uuuu-sw.gw.ap-region.ti.tencentcs.com/ms-xxxx/v1/chat/completions
              # authorization_header: authentication HTTP header.
              authorization_header: 
                # key HTTP Header Key
                key: authorization
                # value: HTTP header content.
                value: your_token
            # generation_params: parameters for calling a judge model.
            # Note: According to restrictions in Tencent Cloud API specifications, this field should be a string.
            # Namely, add `|-` at the end of this field to limit the field type to a string.
            # During subsequent parsing, we will convert this field into one of the parameters in the judge model request body.
            generation_params: |-
              temperature: 0.8
              top_p: 0.85
 
            # judge_template_filename: scoring template file of a judge model.
            judge_template_filename: template.jinja
            # judge_template_content: If no file is specified, you can use this field to specify the scoring template content of a judge model.
            judge_template_content: |-
              You are a judge. Please rate the answers, with 5 as the highest score and 1 as the lowest score.
 
              [Question]
              {{ data.question }}
 
              [Answer from the to-be-evaluated model]
              {{ response.content }}
 
              [Reference answer]
              {{ data.ref_answer }}
 
              Now give your score.
 
        # type is POSTPROCESS, which indicates that this node is a postprocessing node.
        - type: POSTPROCESS
          filename: scripts/postprocessor.py
- data_name: another_dataset_name    # You can configure multiple data sets in a YAML configuration file.
  metrics:
    - name: judge model scoring.
      pipeline:
        - type: PREPROCESS
          filename: preprocessor-at-root.py
        - type: JUDGE_MODEL
          filename: judge.yaml  # You can also configure a judge model separately through a YAML file.
        - type: POSTPROCESS
          filename: postprocessor-at-root.py