lavis/configs/models/pnp-vqa/pnp_vqa_3b.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: pnp_vqa
  model_type: 3b

  image_question_matching_model:
    arch: blip_image_text_matching
    load_finetuned: True

    finetuned: "https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_large_retrieval_coco_train2014.pth"

    # vit encoder
    vit_type: "large"
    vit_grad_ckpt: False
    vit_ckpt_layer: 0

    image_size: 384

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

    embed_dim: 256

  image_captioning_model:
    arch: blip_caption
    load_finetuned: True

    finetuned: "https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_large_caption_coco_train2014.pth"

    vit_type: "large"
    vit_grad_ckpt: True
    vit_ckpt_layer: 5

    image_size: 384

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

    # generation configs
    prompt: "a picture of "

  question_answering_model:
    arch: pnp_unifiedqav2_fid

    pretrained: "allenai/unifiedqa-v2-t5-3b-1363200"

    t5_config_path: "configs/models/pnp-vqa/unifiedqav2_3b_config.json"

preprocess:
  vis_processor:
      eval:
        name: "blip_image_eval"
        image_size: 384
  text_processor:
      eval:
        name: "blip_caption"