env:
  debug: false
  cuda: true
  dist_backend: nccl
  matplotlib_mode: agg
  log_root_dir: log
  rnd_seed: 20

model: MODEL(vd_noema)

eval:
  main: lib.experiments.vd_default.eval
  stage: lib.experiments.vd_default.eval_stage

  dataset: null

  conditioning:
    - ["a dream of a village in china, by Caspar David Friedrich, matte painting trending on artstation HQ", "image"]
    - ["a beautiful grand nebula in the universe", "image"]
    - ["area of rocks that deep inside the forest, divine domain", "image"]
    - ["heavy arms gundam penguin mech", "image"]
    - ["assets/boy_and_girl.jpg", "image"]
    - ["assets/house_by_lake.jpg", "image"]
    - ["assets/ghibli.jpg", "image"]
    - ["assets/scream.jpg", "image"]

    - ["a dream of a village in china, by Caspar David Friedrich, matte painting trending on artstation HQ", "text"]
    - ["a beautiful grand nebula in the universe", "text"]
    - ["area of rocks that deep inside the forest, divine domain", "text"]
    - ["heavy arms gundam penguin mech", "text"]
    - ["assets/boy_and_girl.jpg", "text"]
    - ["assets/house_by_lake.jpg", "text"]
    - ["assets/ghibli.jpg", "text"]
    - ["assets/scream.jpg", "text"]

  replicate: 1

  sample:
    image_output_dim: [512, 512]
    text_latent_dim: 768
    n_samples: 4
    ddim_steps: 50
    ddim_eta: 0.0
    scale: 7.5

  # Some useful post processing
  prompt_temperature: 1.0
  prompt_merge_same_adj_word: true
  color_adj: true
  color_adj_keep_ratio: 0.5
  color_adj_simple: true

  batch_size_per_gpu: 0
  batch_size: null
  dataset_num_workers_per_gpu: 0
  dataset_num_workers: null

  pretrained_pth: pretrained/vd-four-flow-v1-0.pth
  strict_sd: true

  is_lite: true
  fix_seed: true
  eval_subdir: vd-four-flow-v1-0-results