lavis/configs/models/albef_vqav2.yaml

 # Copyright (c) 2022, salesforce.com, inc.
 # All rights reserved.
 # SPDX-License-Identifier: BSD-3-Clause
 # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause

model:
  arch: albef_vqa
  load_finetuned: True

  pretrained: "https://storage.googleapis.com/sfr-pcl-data-research/ALBEF/ALBEF.pth"
  finetuned: "https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/ALBEF/albef_vqav2_lavis.pt"

  use_distill: True
  momentum: 0.995
  alpha: 0.4

  # vit encoder
  vit_type: "base"
  vit_grad_ckpt: False
  vit_ckpt_layer: 0
  vit_layer_norm_epsilon: 1e-6

  image_size: 384

  # bert config
  med_config_path: "configs/models/med_config_albef.json"

preprocess:
  vis_processor:
      train:
        name: "blip_image_train"
        image_size: 384
      eval:
        name: "blip_image_eval"
        image_size: 384
  text_processor:
      train:
        name: "blip_question"
      eval:
        name: "blip_question"