lm-evaluation/build/lib/lm_eval/tasks/babi/babi.yaml

task: babi
dataset_path: Muennighoff/babi
dataset_name: null
output_type: generate_until
training_split: train
validation_split: valid
test_split: test
doc_to_text: "Passage: {{passage}}Question: {{question}}\nAnswer:"
doc_to_target: " {{answer}}"
target_delimiter: ""
generation_kwargs:
  until:
    - "\n"
    - "Passage:"
metric_list:
  - metric: exact_match
    aggregation: mean
    higher_is_better: true
metadata:
  version: 1.0