decoder/config.json

{
  "_class_name": "StableCascadeUNet",
  "_diffusers_version": "0.28.2",
  "block_out_channels": [
    320,
    640,
    1280,
    1280
  ],
  "block_types_per_layer": [
    [
      "SDCascadeResBlock",
      "SDCascadeTimestepBlock"
    ],
    [
      "SDCascadeResBlock",
      "SDCascadeTimestepBlock"
    ],
    [
      "SDCascadeResBlock",
      "SDCascadeTimestepBlock",
      "SDCascadeAttnBlock"
    ],
    [
      "SDCascadeResBlock",
      "SDCascadeTimestepBlock",
      "SDCascadeAttnBlock"
    ]
  ],
  "clip_image_in_channels": null,
  "clip_seq": 4,
  "clip_text_in_channels": null,
  "clip_text_pooled_in_channels": 1280,
  "conditioning_dim": 1280,
  "down_blocks_repeat_mappers": [
    1,
    1,
    1,
    1
  ],
  "down_num_layers_per_block": [
    2,
    6,
    28,
    6
  ],
  "dropout": [
    0,
    0,
    0.1,
    0.1
  ],
  "effnet_in_channels": 16,
  "in_channels": 4,
  "kernel_size": 3,
  "num_attention_heads": [
    0,
    0,
    20,
    20
  ],
  "out_channels": 4,
  "patch_size": 2,
  "pixel_mapper_in_channels": 3,
  "self_attn": true,
  "switch_level": null,
  "timestep_conditioning_type": [
    "sca"
  ],
  "timestep_ratio_embedding_dim": 64,
  "up_blocks_repeat_mappers": [
    3,
    3,
    2,
    2
  ],
  "up_num_layers_per_block": [
    6,
    28,
    6,
    2
  ]
}