{
  "model_type": "spatial_dit",
  "architectures": [
    "SpatialDiT"
  ],
  "sample_rate": 48000,
  "latent_fps": 50,
  "channel_layout": "7.1.4",
  "channel_order": [
    "FL",
    "FR",
    "FC",
    "LFE",
    "BL",
    "BR",
    "SL",
    "SR",
    "TFL",
    "TFR",
    "TBL",
    "TBR"
  ],
  "target_channels": 12,
  "cond_channels": 1,
  "latent_dim": 64,
  "hidden_dim": 1024,
  "num_layers": 12,
  "num_heads": 16,
  "mlp_ratio": 4.0,
  "timestep_embed_dim": 1024,
  "timestep_scale": 1000.0,
  "max_period": 10000.0,
  "num_memory_tokens": 32
}