config.json

{
  "num_attention_heads": 30,
  "attention_head_dim": 64,
  "vae_channels": 16,
  "in_channels": 6,
  "downscale_coef": 8,
  "flip_sin_to_cos": true,
  "freq_shift": 0,
  "time_embed_dim": 512,
  "num_layers": 6,
  "dropout": 0.0,
  "attention_bias": true,
  "sample_width": 90,
  "sample_height": 60,
  "sample_frames": 49,
  "patch_size": 2,
  "temporal_compression_ratio": 4,
  "max_text_seq_length": 226,
  "activation_fn": "gelu-approximate",
  "timestep_activation_fn": "silu",
  "norm_elementwise_affine": true,
  "norm_eps": 1e-05,
  "spatial_interpolation_scale": 1.875,
  "temporal_interpolation_scale": 1.0,
  "use_rotary_positional_embeddings": false,
  "use_learned_positional_embeddings": false,
  "out_proj_dim": 1920,
  "out_proj_dim_zero_init": true,
  "camera_scale_fourier_bands": 6,
  "camera_scale_fourier_min_freq": 1.0,
  "camera_scale_fourier_max_freq": 64.0,
  "camera_scale_embed_dim": null,
  "_use_default_values": [
    "freq_shift",
    "timestep_activation_fn",
    "sample_height",
    "dropout",
    "camera_scale_fourier_bands",
    "temporal_compression_ratio",
    "max_text_seq_length",
    "camera_scale_embed_dim",
    "attention_bias",
    "time_embed_dim",
    "camera_scale_fourier_min_freq",
    "sample_frames",
    "use_rotary_positional_embeddings",
    "sample_width",
    "norm_elementwise_affine",
    "norm_eps",
    "spatial_interpolation_scale",
    "use_learned_positional_embeddings",
    "activation_fn",
    "flip_sin_to_cos",
    "camera_scale_fourier_max_freq",
    "temporal_interpolation_scale",
    "patch_size"
  ]
}