seq2seq.yaml

group:
  - super-glue-lm-eval-v1-seq2seq
task: "boolq-seq2seq"
dataset_path: super_glue
dataset_name: boolq
output_type: greedy_until
training_split: train
validation_split: validation
doc_to_text: "{{passage}}\nQuestion: {{question}}\nAnswer:"
doc_to_target: "{{answer_choices[label]}}"
gold_alias: " {{answer_choices[label]}}" # this will be cast to an int.
generation_kwargs:
  until:
    - "\n\n"
    - "\n"
  do_sample: false
  temperature: 0.0
template_aliases: "{% set answer_choices = ['no', 'yes'] %}"
metric_list:
  - metric: exact_match
    aggregation: mean
    higher_is_better: true
    ignore_case: true
    ignore_punctuation: true