{
  "cross_attention_dim": 1024,
  "in_channels": 64,
  "inflation_end": -1,
  "inflation_start": 0,
  "mlp_ratio": 4.0,
  "num_attention_heads": 16,
  "num_layers": 21,
  "num_tokens_nominal": 2048,
  "temporal_context_size": 16,
  "width": 2048
}