model:
  arch: minigpt_v2
  model_type: pretrain
  max_txt_len: 500
  end_sym: "</s>"
  low_resource: False
  prompt_template: '[INST] {} [/INST]'
  llama_model: "meta-llama/Llama-2-7b-chat-hf"
  ckpt: "./ckpt/checkpoint_stage3.pth"
  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:
  mvtec_ad:
    batch_size: 4
    eval_file_path: ./data/MVTEC_det/val_data.json
    max_new_tokens: 40

run:
  task: image_text_pretrain
  name: minigptv2_evaluation
  save_path: eval_outputs