{
  "model_type": "inl-vae",
  "image_size": 256,
  "base_channels": 128,
  "latent_dim": 4,
  "channel_mult": [
    1,
    2,
    4,
    4
  ],
  "num_res_blocks": 2,
  "training_steps": 15000,
  "torch_dtype": "float16"
}