config.json

{
  "auto_map": {
    "AutoConfig": "configuration_dualvitok.DualViTokConfig",
    "AutoModel": "modeling_dualvitok.DualViTok"
  },
  "architectures": [
    "DualViTok"
  ],
  "semantic_encoder": {
    "pretrained_semantic_encoder": "Emova-ollm/qwen2vit600m",
    "z_channels": 32,
    "num_blocks": 4,
    "out_layer": "linear",
    "embed_dim": 1280,
    "target_mlp": "norm"
  },
  "semantic_decoder": {
    "z_channels": 32,
    "num_blocks": 4,
    "embed_dim": 1280,
    "out_layer": "linear_norm",
    "out_channels": 3584
  },
  "semantic_quantizer_type": "simvq",
  "pixel_quantizer_type": "simvq",
  "semantic_quantizer_codebook_size": 32768,
  "pixel_quantizer_codebook_size": 98304,
  "attn_implementation": "eager",
  "pixel_encoder": {
    "codebook_size": 98304,
    "embed_dim": 32,
    "z_channels": 32,
    "double_z": false,
    "in_channels": 3,
    "out_channels": 3,
    "ch": 128,
    "ch_mult": [
      1,
      1,
      2,
      2,
      4
    ],
    "num_res_blocks": 2,
    "attn_resolutions": [
      4
    ],
    "dropout": 0.0,
    "use_dc_up_down_blocks": true
  },
  "pixel_decoder": {
    "codebook_size": 98304,
    "embed_dim": 64,
    "z_channels": 64,
    "double_z": false,
    "in_channels": 3,
    "out_channels": 3,
    "ch": 384,
    "ch_mult": [
      1,
      1,
      2,
      2,
      4
    ],
    "num_res_blocks": 2,
    "attn_resolutions": [
      4
    ],
    "dropout": 0.0,
    "use_dc_up_down_blocks": true
  },
  "torch_dtype": "float16",
  "transformers_version": "4.44.2"
}