williamberman/consistency-decoder

1{2  "_class_name": "ConsistencyDecoderVae",3  "_diffusers_version": "0.23.0.dev0",4  "block_out_channels": [5    128,6    256,7    512,8    5129  ],10  "decoder_args": {11    "_use_default_values": [12      "downsample_type",13      "flip_sin_to_cos",14      "mid_block_scale_factor",15      "num_class_embeds",16      "attention_head_dim",17      "sample_size",18      "act_fn",19      "freq_shift",20      "class_embed_type",21      "attn_norm_num_groups",22      "downsample_padding",23      "upsample_type",24      "dropout",25      "center_input_sample"26    ],27    "act_fn": "silu",28    "add_attention": false,29    "attention_head_dim": 8,30    "attn_norm_num_groups": null,31    "block_out_channels": [32      320,33      640,34      1024,35      102436    ],37    "center_input_sample": false,38    "class_embed_type": null,39    "down_block_types": [40      "ResnetDownsampleBlock2D",41      "ResnetDownsampleBlock2D",42      "ResnetDownsampleBlock2D",43      "ResnetDownsampleBlock2D"44    ],45    "downsample_padding": 1,46    "downsample_type": "conv",47    "dropout": 0.0,48    "flip_sin_to_cos": true,49    "freq_shift": 0,50    "in_channels": 7,51    "layers_per_block": 3,52    "mid_block_scale_factor": 1,53    "norm_eps": 1e-05,54    "norm_num_groups": 32,55    "num_class_embeds": null,56    "num_train_timesteps": 1024,57    "out_channels": 6,58    "resnet_time_scale_shift": "scale_shift",59    "sample_size": null,60    "time_embedding_type": "learned",61    "up_block_types": [62      "ResnetUpsampleBlock2D",63      "ResnetUpsampleBlock2D",64      "ResnetUpsampleBlock2D",65      "ResnetUpsampleBlock2D"66    ],67    "upsample_type": "conv"68  },69  "encoder_args": {70    "act_fn": "silu",71    "block_out_channels": [72      128,73      256,74      512,75      51276    ],77    "double_z": true,78    "down_block_types": [79      "DownEncoderBlock2D",80      "DownEncoderBlock2D",81      "DownEncoderBlock2D",82      "DownEncoderBlock2D"83    ],84    "in_channels": 3,85    "layers_per_block": 2,86    "norm_num_groups": 32,87    "out_channels": 488  },89  "latent_channels": 4,90  "scaling_factor": 0.1821591}92