# 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: blip2_vicuna_instruct
  load_finetuned: False
  load_pretrained: True

  pretrained: "/mnt/pfs-guan-ssai/nlu/wanghanzi/models/blip2/blip2_vicuna7b/blip2_pretrained_vicuna7b.pth"
  finetuned: ""
  q_former_model: "/mnt/pfs-guan-ssai/nlu/wanghanzi/models/blip2/blip2_vicuna7b/blip2_pretrained_vicuna7b.pth"

  # vit encoder
  image_size: 224
  drop_path_rate: 0
  use_grad_checkpoint: False
  vit_precision: "fp16"

  # Q-Former
  num_query_token: 32
  qformer_text_input: True

  # path to Vicuna checkpoint
  llm_model: "/mnt/pfs-guan-ssai/nlu/wanghanzi/models/vicuna-7b-v1.1"
  prompt: ""
  max_txt_len: 256
  max_output_txt_len: 256

  # freeze
  freeze_vit: True
  freeze_llm: True
  freeze_qformer: False
  freeze_t5_proj: False

  # moe
  general_version: 'route_moe'
  moebert_route_method: "post-route"
  moebert_load_balance: 0.05
  moebert_expert_num: 3
  moebert_num_beams: 3
  moe_weight_type: 'ffn_prob'
  use_balance_loss: False
  ln_position: "out"

preprocess:
    vis_processor:
        train:
          name: "blip2_image_train"
          image_size: 224
        eval:
          name: "blip_image_eval"
          image_size: 224
    text_processor:
        train:
          name: "blip_caption"
        eval:
          name: "blip_caption"