| { | |
| "model_class": "SanaModel", | |
| "input_size": 32, | |
| "patch_size": 1, | |
| "in_channels": 32, | |
| "hidden_size": 2240, | |
| "depth": 20, | |
| "num_heads": 20, | |
| "mlp_ratio": 2.5, | |
| "caption_channels": 2304, | |
| "model_max_length": 300, | |
| "attn_type": "linear", | |
| "cross_attn_type": "flash", | |
| "linear_head_dim": 32, | |
| "qk_norm": true, | |
| "cross_norm": true, | |
| "ffn_type": "glumbconv", | |
| "mlp_acts": [ | |
| "silu", | |
| "silu", | |
| null | |
| ], | |
| "pred_sigma": false, | |
| "learn_sigma": false, | |
| "use_pe": false, | |
| "pe_interpolation": 1.0, | |
| "y_norm": true, | |
| "y_norm_scale_factor": 1.0, | |
| "use_connector": true, | |
| "y_embedding_init_method": "zeros", | |
| "connector_channels": 4096, | |
| "norm_eps": 1e-06, | |
| "class_dropout_prob": 0.1, | |
| "drop_path": 0.0, | |
| "vae_type": "AutoencoderDC", | |
| "vae_pretrained": "mit-han-lab/dc-ae-f32c32-sana-1.1-diffusers", | |
| "vae_downsample_rate": 32, | |
| "text_encoder_name": "google/gemma-2-2b-it", | |
| "__class__": "SanaConfig" | |
| } | |