{ "num_attention_heads": 48, "attention_head_dim": 64, "in_channels": 32, "out_channels": 16, "flip_sin_to_cos": true, "freq_shift": 0, "time_embed_dim": 512, "ofs_embed_dim": null, "text_embed_dim": 4096, "num_layers": 42, "dropout": 0.0, "attention_bias": true, "sample_width": 90, "sample_height": 60, "sample_frames": 49, "patch_size": 2, "patch_size_t": null, "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": true, "use_learned_positional_embeddings": true, "patch_bias": true, "_class_name": "CogVideoXTransformer3DModelTracking", "_diffusers_version": "0.32.2", "num_tracking_blocks": 18 }