dataset_path: LongVideo-Reason/longvideo-reason
dataset_kwargs:
  cache_dir: longvideo_reason
  video: true
  create_link: true
test_split: test
output_type: generate_until
cluster_key: videos
doc_to_visual: !function utils.longvideo_reason_doc_to_visual
doc_to_text: !function utils.longvideo_reason_doc_to_text
doc_to_target: !function utils.longvideo_reason_doc_to_target
generation_kwargs:
  max_new_tokens: 2048
  temperature: 0
  top_p: 1.0
  num_beams: 1
  do_sample: false
process_results: !function utils.longvideo_reason_process_results
metric_list:
  - metric: longvideo_reason_overall_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_overall
    higher_is_better: true
  - metric: longvideo_reason_strict_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_strict
    higher_is_better: true
  - metric: longvideo_reason_wellformed_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_wellformed
    higher_is_better: true
  - metric: longvideo_reason_format_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_format
    higher_is_better: true
  - metric: longvideo_reason_temporal_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_temporal
    higher_is_better: true
  - metric: longvideo_reason_goal_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_goal
    higher_is_better: true
  - metric: longvideo_reason_spatial_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_spatial
    higher_is_better: true
  - metric: longvideo_reason_plot_accuracy
    aggregation: !function utils.longvideo_reason_aggregate_plot
    higher_is_better: true
lmms_eval_specific_kwargs:
  default:
    pre_prompt: ""
    post_prompt: ""
metadata:
  version: 1.0
