transformer/config.json

{
  "_class_name": "SanaVideoTransformer3DModel",
  "_diffusers_version": "0.36.0.dev0",
  "attention_bias": false,
  "attention_head_dim": 112,
  "caption_channels": 2304,
  "cross_attention_dim": 2240,
  "cross_attention_head_dim": 112,
  "dropout": 0.0,
  "guidance_embeds": false,
  "guidance_embeds_scale": 0.1,
  "in_channels": 16,
  "interpolation_scale": null,
  "mlp_ratio": 3.0,
  "norm_elementwise_affine": false,
  "norm_eps": 1e-06,
  "num_attention_heads": 20,
  "num_cross_attention_heads": 20,
  "num_layers": 20,
  "out_channels": 16,
  "patch_size": [
    1,
    2,
    2
  ],
  "qk_norm": "rms_norm_across_heads",
  "rope_max_seq_len": 1024,
  "sample_size": 30
}