group: mmlu_pro
task:
  - mmlu_pro_biology
  - mmlu_pro_business
  - mmlu_pro_chemistry
  - mmlu_pro_computer_science
  - mmlu_pro_economics
  - mmlu_pro_engineering
  - mmlu_pro_health
  - mmlu_pro_history
  - mmlu_pro_law
  - mmlu_pro_math
  - mmlu_pro_other
  - mmlu_pro_philosophy
  - mmlu_pro_physics
  - mmlu_pro_psychology
aggregate_metric_list:
  - aggregation: mean
    metric: exact_match
    weight_by_size: true
    filter_list: custom-extract
metadata:
  version: 2.0