model:
  arch: minigpt_v2
  model_type: pretrain
  max_txt_len: 500
  end_sym: "###"
  low_resource: False
  prompt_template: 'Instruct: {} /n Output: '
  llama_model: ""
  ckpt: ""
  lora_r: 64
  lora_alpha: 16



datasets:
  cc_sbu_align:
    vis_processor:
      train:
        name: "blip2_image_eval"
        image_size: 448
    text_processor:
      train:
        name: "blip_caption"

evaluation_datasets:
  gqa:
    eval_file_path: /root/autodl-tmp/evaluation/gqa/annotations/testdev_balanced_questions.json
    img_path: /root/autodl-tmp/evaluation/gqa/images  
    max_new_tokens: 20
    batch_size: 10
  vizwiz:
    eval_file_path: /root/autodl-tmp/evaluation/vizwiz/val.json
    img_path: /root/autodl-tmp/evaluation/vizwiz/val  
    max_new_tokens: 20
    batch_size: 10
  iconvqa:
    eval_file_path: /root/autodl-tmp/evaluation/iconqa/iconqa_data/problems.json
    img_path: /root/autodl-tmp/evaluation/iconqa/iconqa_data/iconqa 
    max_new_tokens: 20
    batch_size: 1
  vsr:
    eval_file_path: /root/autodl-tmp/evaluation/vsr/dev.jsonl
    img_path: /root/autodl-tmp/coco2017/train  
    max_new_tokens: 20
    batch_size: 10
  hm:
    eval_file_path: /root/autodl-tmp/evaluation/Hateful_Memes/data/dev.jsonl  
    img_path: /root/autodl-tmp/evaluation/Hateful_Memes/data
    max_new_tokens: 20
    batch_size: 10

run:
  task: image_text_pretrain
  name: minigptv2_evaluation
  save_path: /root/MiniGPT-4/save_evalution