model:
  arch: bind_gpt4
  model_type: pretrain_vicuna
  freeze_imagebind: True
  freeze_qformer: False
  max_txt_len: 160
  end_sym: "###"
  low_resource: False
  prompt_path: "prompts/alignment.txt"
  prompt_template: '###Human: {} ###Assistant: '
  ckpt: 'minigpt4/output/minigpt4_stage1_pretrain/20230524192/checkpoint_0.pth'


datasets:
  cc12m:  # Double check
    vis_processor:
      train:
        name: "imagebind_vision_eval"
        image_size: 224
    text_processor:
      train:
        name: "imagebind_caption"

run:
  task: image_text_pretrain