{
  "_class_name": "GAECodec",
  "latent_dim": 64,
  "hidden_dims": [
    3072,
    2048,
    1024
  ],
  "num_attn_blocks": 4,
  "attn_heads": 8,
  "kl_weight": 1e-06,
  "use_spatial": true,
  "level_weights": [
    2.0,
    1.0,
    1.0,
    1.0
  ],
  "feature_level_dim": 3072,
  "rgb_decoder": {
    "hidden_dim": 1024,
    "depth": 12,
    "num_heads": 16,
    "ffn_ratio": 4.0,
    "patch_size": 14,
    "temporal": true,
    "temporal_mode": "divided",
    "temporal_num_heads": 16,
    "rope_theta": 10000.0
  }
}