{
  "model_type": "autoencoder_kl",
  "sample_size": 256,
  "in_channels": 3,
  "out_channels": 3,
  "latent_channels": 8,
  "base_channels": 128,
  "channel_multipliers": [1, 2, 4, 4],
  "num_res_blocks": 3,
  "dropout": 0.0,
  "use_attention": true,
  "attention_resolutions": [32],
  "attention_heads": 4,
  "scaling_factor": 1.0,
  "recon_loss_type": "l1",
  "trained_on": "coco_captions",
  "resolution": 256
}