lm-evaluation-harness/lm_eval/tasks/ceval/_default_ceval_yaml

group: ceval-valid
dataset_path: ceval/ceval-exam
validation_split: val
fewshot_split: dev
fewshot_config:
  sampler: first_n
output_type: multiple_choice
doc_to_text: "{{question.strip()}}\nA. {{A}}\nB. {{B}}\nC. {{C}}\nD. {{D}}\n答案："
doc_to_choice: ["A", "B", "C", "D"]
doc_to_target: "{{['A', 'B', 'C', 'D'].index(answer)}}"
metric_list:
  - metric: acc
    aggregation: mean
    higher_is_better: true
  - metric: acc_norm
    aggregation: mean
    higher_is_better: true
metadata:
  version: 1.0