medversa.yaml

model:
  arch: medomni
  model_type: medomni
  freeze_vit: True
  freeze_qformer: True
  max_txt_len: 256
  end_sym: "###"
  low_resource: False

datasets:
  med:
    vis_processor:
      train:
        name: "blip2_image_eval"
        image_size: 224
    text_processor:
      train:
        name: "blip_caption"

run:
  task: image_text_pretrain