dataset_path: Video-Index/Video-Index
test_split: test
output_type: generate_until
doc_to_target: !function utils.video_index_doc_to_target
generation_kwargs:
  max_new_tokens: 16
  temperature: 0
  top_p: 1.0
  num_beams: 1
  do_sample: false
# The return value of process_results will be used by metrics
process_results: !function utils.video_index_process_results
metric_list:
  - metric: video_index_acc
    aggregation: !function utils.video_index_aggregate_overall
    higher_is_better: true
  - metric: video_index_perception
    aggregation: !function utils.video_index_aggregate_perception
    higher_is_better: true
  - metric: video_index_temporal
    aggregation: !function utils.video_index_aggregate_temporal
    higher_is_better: true
  - metric: video_index_spatial
    aggregation: !function utils.video_index_aggregate_spatial
    higher_is_better: true
  - metric: video_index_reasoning
    aggregation: !function utils.video_index_aggregate_reasoning
    higher_is_better: true
metadata:
  - version: 0.0
