group: mmlu_pro_plus
task:
  - mmlu_pro_plus_biology
  - mmlu_pro_plus_business
  - mmlu_pro_plus_chemistry
  - mmlu_pro_plus_computer_science
  - mmlu_pro_plus_economics
  - mmlu_pro_plus_engineering
  - mmlu_pro_plus_health
  - mmlu_pro_plus_history
  - mmlu_pro_plus_law
  - mmlu_pro_plus_math
  - mmlu_pro_plus_other
  - mmlu_pro_plus_philosophy
  - mmlu_pro_plus_physics
  - mmlu_pro_plus_psychology
aggregate_metric_list:
  - aggregation: mean
    metric: exact_match
    weight_by_size: true
    filter_list: custom-extract
metadata:
  version: 1.0