dataset_path: TIGER-Lab/MMLU-Pro
test_split: test
fewshot_split: validation
fewshot_config:
  sampler: first_n
  doc_to_text: !function utils.fewshot_to_text
  doc_to_target: !function utils.empty_str
output_type: generate_until
doc_to_text: !function utils.doc_to_text
doc_to_target: answer
filter_list:
  - name: "tigerlab-extract"
    filter:
      - function: "fallback-regex"
        regex_patterns:
          - 'answer is \(?([A-J])\)?'
          - '.*[aA]nswer:\s*([A-J])'
          - '\b[A-J]\b(?!.*\b[A-J]\b)'
        group_select: [1, 1, 0]
        fallback: "[invalid]"
      - function: "take_first"
generation_kwargs:
  max_gen_toks: 1024
  until:
    - "</s>"
    - "Q:"
    - "Question:"
    - "<|im_end|>"
  do_sample: false
  temperature: 0.0
num_fewshot: 5
metric_list:
  - metric: exact_match
    aggregation: mean
    higher_is_better: true
    ignore_case: true
    ignore_punctuation: true
metadata:
  version: 1.0