| { | |
| "num_attention_heads": 16, | |
| "attention_head_dim": 72, | |
| "in_channels": 4, | |
| "cond_channels": 9, | |
| "out_channels": 8, | |
| "num_layers": 28, | |
| "dropout": 0.0, | |
| "norm_num_groups": 32, | |
| "cross_attention_dim": 1152, | |
| "attention_bias": true, | |
| "sample_size": 128, | |
| "patch_size": 2, | |
| "activation_fn": "gelu-approximate", | |
| "num_embeds_ada_norm": 1000, | |
| "upcast_attention": false, | |
| "norm_type": "ada_norm_single", | |
| "norm_elementwise_affine": false, | |
| "norm_eps": 1e-06, | |
| "caption_channels": 4096, | |
| "attention_type": "default" | |
| } |