dataset_path: sming256/MinimalVideoPairs
dataset_kwargs:
  token: True
  cache_dir: minimal_video_pairs
  video: True
lmms_eval_specific_kwargs:
  default:
    pre_prompt: "You are an expert video understanding AI system. Carefully watch the video and pay attention to the cause and sequence of events, the details and movements of objects, and actions of people. Based on your observations, select the best option that accurately addresses the following question: \n"
    post_prompt: " \nEven when unsure, always answer with a single letter from A or B, format exactly like: 'Answer: A' or 'Answer: B'."
test_split: train
output_type: generate_until
doc_to_visual: !function utils.mvp_doc_to_visual
doc_to_text: !function utils.mvp_doc_to_text
doc_to_target: !function utils.mvp_doc_to_answer
process_results: !function utils.mvp_process_results
