vir_vlfm_clevr_change.yaml

model:
  arch: vir_vlfm
  vit_model: "eva_clip_g"
  model_type: pretrain_vicuna
  vit_precision: ""
  freeze_vit: True
  freeze_qformer: True
  use_adapter: True
  max_txt_len: 50
  end_sym: "###"
  prompt_path: "prompts/clevr_change.txt"
  prompt_template: '###Human: {} ###Assistant: '
  ckpt: './prerained_minigpt4_7b.pth'

datasets:
  clevr_change:
    vis_processor:
      train:
        name: "blip2_image_train"
        image_size: 224
    text_processor:
      train:
        name: "blip_caption"

run:
  task: image_text_pretrain
  # optimizer
  lr_sched: "linear_warmup_cosine_lr"
  init_lr: 5e-5
  min_lr: 1e-5
  warmup_lr: 1e-6

  weight_decay: 0.05
  max_epoch: 40
  iters_per_epoch: 4500
  batch_size_train: 1
  batch_size_eval: 1
  num_workers: 16
  warmup_steps: 4500

  seed: 42
  output_dir: "./output/vir_vlfm"

  amp: True
  resume_ckpt_path: null

  evaluate: False 
  train_splits: ["train"]

  device: "cuda"
  world_size: 1
  dist_url: "env://"
  distributed: True