transformer/config.json

{
  "_class_name": "Transformer3DModel",
  "_diffusers_version": "0.33.1",
  "activation_fn": "gelu-approximate",
  "adaln_norm_type": "layernorm",
  "add_motion_score": true,
  "attention_bias": true,
  "attention_head_dim": 64,
  "attention_type": "default",
  "caption_channels": null,
  "cfg_parallel_group": null,
  "class_dropout_prob": 0.1,
  "compile_mem_gc_threshold": 60,
  "cond_channels": 0,
  "condition_type": "inpaint_mask",
  "cross_attention_dim": 2048,
  "desired_step": false,
  "double_self_attention": false,
  "dropout": 0.0,
  "enable_cached_compile": false,
  "enable_cfg_parallel": false,
  "gradient_checkpointing": "full-block",
  "in_channels": 16,
  "norm_elementwise_affine": false,
  "norm_eps": 1e-05,
  "norm_num_groups": 32,
  "norm_type": "ada_norm_zero_nolabel",
  "num_attention_heads": 48,
  "num_block_chunks": 40,
  "num_embeds_ada_norm": null,
  "num_latent_frames": 24,
  "num_layers": 40,
  "only_cross_attention": false,
  "out_channels": 16,
  "patch_size": 2,
  "patch_size_t": 1,
  "sample_size": 32,
  "sequence_parallel_size": 0,
  "task_type": "i2v+++",
  "time_embed_type": "rope",
  "upcast_attention": false,
  "use_depth_cond": true,
  "use_depth_loss": false,
  "use_navit": false,
  "use_rope": true,
  "vae_stride": 8
}