finetune-db-dog.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: blip_diffusion

  model_type: base

  qformer_train: False
  sd_train_text_encoder: False

  vae_half_precision: True
  # load_finetuned: False

datasets:
  blip_diffusion_finetune: # name of the dataset builder
    text_processor:
        train:
          name: "blip_caption"
        eval:
          name: "blip_caption"
    # use keyword processors
    kw_processor:
      inp_vis_processor:
          name: "blip_diffusion_inp_image_train"
      tgt_vis_processor:
          name: "blip_diffusion_tgt_image_train"
    
    build_info:
      subject_text: "dog"
      images:
        storage: "/export/home/workspace/LAVIS-Diffusion/LAVIS/projects/blip-diffusion/images/dreambooth/dog8" 

run:
  runner: runner_iter

  max_iters: 40
  iters_per_inner_epoch: 20

  task: text-to-image-generation
  # optimizer
  lr_sched: "constant_lr"
  init_lr: 5e-6
  min_lr: 0
  weight_decay: 0.01

  # batch_size_train: 2
  # batch_size_train: 16
  batch_size_train: 3
  batch_size_eval: 1
  num_workers: 4

  seed: 42
  # output_dir: "output/BLIP-diffusion/finetune"
  output_dir: "output/debug/BLIP-diffusion/finetune"

  amp: True
  # amp: False
  resume_ckpt_path: null

  evaluate: False 
  train_splits: ["train"]

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