lmms_eval/tasks/multidocvqa/multidocvqa_val.yaml

dataset_path: lmms-lab/MP-DocVQA
task: "multidocvqa_val"
test_split: val
output_type: generate_until
doc_to_visual: !function utils.multidocvqa_doc_to_visual
doc_to_text: !function utils.multidocvqa_doc_to_text
doc_to_target: "answers"
generation_kwargs:
  max_new_tokens: 32
  temperature: 0
  do_sample: False
process_results: !function utils.multidocvqa_process_results
metric_list:
  - metric: anls
    aggregation: !function utils.multidocvqa_aggregate_results_anls
    higher_is_better: true
  - metric: accuracy
    aggregation: !function utils.multidocvqa_aggregate_results_accuracy
    higher_is_better: true
model_specific_prompt_kwargs:
  default:
    pre_prompt: ""
    post_prompt: "\nAnswer the question using a single word or phrase."