dataset_path: HAERAE-HUB/KMMLU-HARD
output_type: generate_until
validation_split: dev # not meant to be used, only here to silence warnings
test_split: test
doc_to_target: "{{['A', 'B', 'C', 'D'][answer-1] if answer is defined else target}}" # answer undefined: cot few shot example, else normal sample
metric_list:
  - metric: exact_match
    aggregation: mean
    higher_is_better: true
    ignore_case: true
    ignore_punctuation: true
generation_kwargs:
  until:
    - "질문:"
  do_sample: false
  max_gen_toks: 2048
  temperature: 0.0
num_fewshot: 5
filter_list:
  - name: "get-answer"
    filter:
      - function: "regex"
        regex_pattern: "(?<=정답은)(.*)(?=입니다.)"
        regex_pattern: "(?<=정답은)\\s*[(]?(A|B|C|D)[)]?.*(?=입니다.)"
      - function: "take_first"
metadata:
  version: 2.0
