dataset_path: TencentARC/Video-Holmes
dataset_kwargs:
  token: True
  cache_dir: video_holmes
  video: True
  data_files:
    test: test_Video-Holmes.json
    train: train_Video-Holmes.json
test_split: test
output_type: generate_until
doc_to_visual: !function utils.video_holmes_doc_to_visual
doc_to_text: !function utils.video_holmes_doc_to_text
doc_to_messages: !function utils.video_holmes_doc_to_messages
doc_to_target: "Answer"
generation_kwargs:
  max_new_tokens: 64
  temperature: 0
  top_p: 1.0
  do_sample: false
process_results: !function utils.video_holmes_process_results
metric_list:
  - metric: video_holmes_accuracy
    aggregation: !function utils.video_holmes_aggregate_accuracy
    higher_is_better: true
  - metric: video_holmes_SR
    aggregation: !function utils.video_holmes_aggregate_SR
    higher_is_better: true
  - metric: video_holmes_IMC
    aggregation: !function utils.video_holmes_aggregate_IMC
    higher_is_better: true
  - metric: video_holmes_TCI
    aggregation: !function utils.video_holmes_aggregate_TCI
    higher_is_better: true
  - metric: video_holmes_TA
    aggregation: !function utils.video_holmes_aggregate_TA
    higher_is_better: true
  - metric: video_holmes_MHR
    aggregation: !function utils.video_holmes_aggregate_MHR
    higher_is_better: true
  - metric: video_holmes_PAR
    aggregation: !function utils.video_holmes_aggregate_PAR
    higher_is_better: true
  - metric: video_holmes_CTI
    aggregation: !function utils.video_holmes_aggregate_CTI
    higher_is_better: true
lmms_eval_specific_kwargs:
  default:
    pre_prompt: ""
    post_prompt: "\nAnswer with the option's letter from the given choices directly."
  qwen3_vl:
    format: "qwen3_vl"
    pre_prompt: "Question: "
    post_prompt: "Answer with the option letter only."
metadata:
  - version: 0.0
