{ "format_version": 1, "num_classes": 1000, "class_tokens": 4, "class_token_width": 256, "width": 896, "depth": 16, "prefix_depth": 2, "suffix_depth": 2, "heads": 14, "mlp_ratio": 4, "time_embed_width": 256, "time_embed_hidden": 1024, "cross_heads": 4, "cross_head_dim": 64, "register_tokens": 4, "pixel_patch": 16, "semantic_channels": 32, "pixel_noise_std": 0.558, "tail_coarse_patch": 8, "tail_coarse_width": 768, "tail_fine_patch": 4, "tail_fine_width": 256, "tail_mlp_ratio": 3, "tail_kernel_size": 3 }