dit/config.json

{
  "_class_name": "LongCatVideoTransformer3DModel",
  "_diffusers_version": "0.32.0",
  "adaln_tembed_dim": 512,
  "bsa_params": {
    "sparsity": 0.9375,
    "chunk_3d_shape_q": [
      4,
      4,
      4
    ],
    "chunk_3d_shape_k": [
      4,
      4,
      4
    ]
  },
  "caption_channels": 4096,
  "cp_split_hw": null,
  "depth": 48,
  "enable_bsa": false,
  "enable_flashattn3": false,
  "enable_flashattn2": true,
  "frequency_embedding_size": 256,
  "hidden_size": 4096,
  "in_channels": 16,
  "text_tokens_zero_pad": true,
  "mlp_ratio": 4,
  "num_heads": 32,
  "out_channels": 16,
  "patch_size": [
    1,
    2,
    2
  ],
  "quantization": {
    "method": "mlx.nn.quantize",
    "bits": 8,
    "group_size": 64,
    "skip_patterns": [
      "final_layer.linear",
      "t_embedder.",
      "y_embedder.",
      "adaLN_modulation."
    ]
  }
}