tydiqa_bengali.yaml

task: tydiqa_bengali
dataset_path: google-research-datasets/tydiqa
dataset_name: secondary_task
output_type: generate_until
process_docs: !function util.filter_bengali
training_split: train
validation_split: validation
doc_to_text: "Answer the following question based on Context. You can extract an answer span from Context.\nContext: {{context}} Question: {{question}}?\nAnswer:"
doc_to_target: !function util.doc_to_target
process_results: !function util.process_results
num_fewshot: 4
generation_kwargs:
  until:
    - "."
    - "\n"
filter_list:
  - name: remove_whitespace
    filter:
      - function: remove_whitespace
metric_list:
  - metric: em
    aggregation: mean
    higher_is_better: true
  - metric: f1
    aggregation: mean
    higher_is_better: true
metadata:
  version: 1.0