model:
  arch: minigpt_v2
  model_type: pretrain
  max_txt_len: 500
  bos_token_id: "###"
  low_resource: False
  prompt_template: '###Human: {} ###Assistant: '
  ckpt: "checkpoint_49.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"

run:
  task: image_text_pretrain