Instructions to use stabilityai/SAME-L with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Stable Audio 3
How to use stabilityai/SAME-L with Stable Audio 3:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Update model_config.json
Browse files- model_config.json +128 -0
model_config.json
CHANGED
|
@@ -63,5 +63,133 @@
|
|
| 63 |
"latent_dim": 256,
|
| 64 |
"downsampling_ratio": 4096,
|
| 65 |
"io_channels": 2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
}
|
| 67 |
}
|
|
|
|
| 63 |
"latent_dim": 256,
|
| 64 |
"downsampling_ratio": 4096,
|
| 65 |
"io_channels": 2
|
| 66 |
+
},
|
| 67 |
+
"training": {
|
| 68 |
+
"learning_rate": 5e-5,
|
| 69 |
+
"warmup_steps": 0,
|
| 70 |
+
"use_ema": true,
|
| 71 |
+
"decoder_finetune": true,
|
| 72 |
+
"decoder_loss": false,
|
| 73 |
+
"clip_grad_norm": 20.0,
|
| 74 |
+
"num_synthetic_chirps": 2,
|
| 75 |
+
"optimizer_configs": {
|
| 76 |
+
"autoencoder": {
|
| 77 |
+
"optimizer": {
|
| 78 |
+
"type": "CAdamW",
|
| 79 |
+
"config": {
|
| 80 |
+
"betas": [0.9, 0.95],
|
| 81 |
+
"lr": 1e-5,
|
| 82 |
+
"weight_decay": 1e-4
|
| 83 |
+
}
|
| 84 |
+
},
|
| 85 |
+
"scheduler": {
|
| 86 |
+
"type": "InverseLR",
|
| 87 |
+
"config": {
|
| 88 |
+
"inv_gamma": 100000,
|
| 89 |
+
"power": 0.5,
|
| 90 |
+
"warmup": 0.999
|
| 91 |
+
}
|
| 92 |
+
}
|
| 93 |
+
},
|
| 94 |
+
"discriminator": {
|
| 95 |
+
"optimizer": {
|
| 96 |
+
"type": "CAdamW",
|
| 97 |
+
"config": {
|
| 98 |
+
"betas": [0.8, 0.99],
|
| 99 |
+
"lr": 1e-5,
|
| 100 |
+
"weight_decay": 1e-4
|
| 101 |
+
}
|
| 102 |
+
},
|
| 103 |
+
"scheduler": {
|
| 104 |
+
"type": "InverseLR",
|
| 105 |
+
"config": {
|
| 106 |
+
"inv_gamma": 50000,
|
| 107 |
+
"power": 0.5,
|
| 108 |
+
"warmup": 0.999
|
| 109 |
+
}
|
| 110 |
+
}
|
| 111 |
+
}
|
| 112 |
+
},
|
| 113 |
+
"loss_configs": {
|
| 114 |
+
"semantic_regressors": {
|
| 115 |
+
"decay": 1.0,
|
| 116 |
+
"weights": {
|
| 117 |
+
"semantic_regressors_l1": 1.0
|
| 118 |
+
}
|
| 119 |
+
},
|
| 120 |
+
"contrastive": {
|
| 121 |
+
"config": {
|
| 122 |
+
"depth": 4,
|
| 123 |
+
"transformer_dim": 1024,
|
| 124 |
+
"use_text": true
|
| 125 |
+
},
|
| 126 |
+
"weights": {
|
| 127 |
+
"contrastive": 1.0
|
| 128 |
+
}
|
| 129 |
+
},
|
| 130 |
+
"generative":{
|
| 131 |
+
"type": "diffusion",
|
| 132 |
+
"warmup": 5000,
|
| 133 |
+
"config" : {
|
| 134 |
+
"embed_dim": 768,
|
| 135 |
+
"depth": 4
|
| 136 |
+
},
|
| 137 |
+
"weights": {
|
| 138 |
+
"diffusion": 1e-3
|
| 139 |
+
}
|
| 140 |
+
},
|
| 141 |
+
"discriminator": {
|
| 142 |
+
"type": "transformer",
|
| 143 |
+
"config": {
|
| 144 |
+
"normalize_losses": true,
|
| 145 |
+
"add_noise": false,
|
| 146 |
+
"mfb_kwargs": {"enabled": true, "use_HIL": true},
|
| 147 |
+
"stft_kwargs": {"enabled": true, "depths": [1,1,1]},
|
| 148 |
+
"chroma_kwargs": {"enabled": true, "depths": [1,1,1]},
|
| 149 |
+
"patched_kwargs": {"enabled": true, "depths": [1,1,1]}
|
| 150 |
+
},
|
| 151 |
+
"weights": {
|
| 152 |
+
"adversarial": 0.1,
|
| 153 |
+
"feature_matching": 1.0
|
| 154 |
+
}
|
| 155 |
+
},
|
| 156 |
+
"spectral": {
|
| 157 |
+
"type": "mrstft",
|
| 158 |
+
"decay": 1.0,
|
| 159 |
+
"config": {
|
| 160 |
+
"fft_sizes": [2048, 1024, 512, 256, 128, 64, 32],
|
| 161 |
+
"hop_sizes": [512, 256, 128, 64, 32, 16, 8],
|
| 162 |
+
"win_lengths": [2048, 1024, 512, 256, 128, 64, 32],
|
| 163 |
+
"perceptual_weighting": true,
|
| 164 |
+
"w_phs": 1.0,
|
| 165 |
+
"w_sc": 1.0
|
| 166 |
+
},
|
| 167 |
+
"weights": {
|
| 168 |
+
"mrstft": 1.0
|
| 169 |
+
}
|
| 170 |
+
},
|
| 171 |
+
"time": {
|
| 172 |
+
"type": "l1",
|
| 173 |
+
"decay": 1.0,
|
| 174 |
+
"weights": {
|
| 175 |
+
"l1": 0.0,
|
| 176 |
+
"l2": 0.0
|
| 177 |
+
}
|
| 178 |
+
},
|
| 179 |
+
|
| 180 |
+
"bottleneck": {
|
| 181 |
+
"weights": {
|
| 182 |
+
"softnorm": 1e-4
|
| 183 |
+
}
|
| 184 |
+
}
|
| 185 |
+
},
|
| 186 |
+
"eval_loss_configs": {
|
| 187 |
+
"stft": {"log_eps": 1},
|
| 188 |
+
"mel": {"log_eps": 1},
|
| 189 |
+
"sisdr": {}
|
| 190 |
+
},
|
| 191 |
+
"demo": {
|
| 192 |
+
"demo_every": 1000
|
| 193 |
+
}
|
| 194 |
}
|
| 195 |
}
|