{
  "embed_frequency": 8,
  "embed_include_pi": false,
  "framestep_encoding_strategy": "linear",
  "in_channels": 3,
  "in_extra_channels": 3,
  "latent_channels": 64,
  "num_attention_heads": 8,
  "num_freqs_ts": 128,
  "num_layers": 16,
  "out_dim": 3,
  "prediction_mode": "direct",
  "temporal_context_size": 16,
  "width": 1024
}