Upload sc-lora.json
Browse files- sc-lora.json +280 -0
sc-lora.json
ADDED
|
@@ -0,0 +1,280 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"__version": 2,
|
| 3 |
+
"training_method": "LORA",
|
| 4 |
+
"model_type": "STABLE_CASCADE_1",
|
| 5 |
+
"debug_mode": false,
|
| 6 |
+
"debug_dir": "debug",
|
| 7 |
+
"workspace_dir": "workspace/run",
|
| 8 |
+
"cache_dir": "workspace-cache/run",
|
| 9 |
+
"tensorboard": true,
|
| 10 |
+
"tensorboard_expose": false,
|
| 11 |
+
"continue_last_backup": false,
|
| 12 |
+
"include_train_config": "NONE",
|
| 13 |
+
"base_model_name": "stabilityai/stable-cascade-prior",
|
| 14 |
+
"weight_dtype": "BFLOAT_16",
|
| 15 |
+
"output_dtype": "BFLOAT_16",
|
| 16 |
+
"output_model_format": "SAFETENSORS",
|
| 17 |
+
"output_model_destination": "models/model",
|
| 18 |
+
"gradient_checkpointing": true,
|
| 19 |
+
"concept_file_name": "training_concepts/concepts.json",
|
| 20 |
+
"concepts": null,
|
| 21 |
+
"circular_mask_generation": false,
|
| 22 |
+
"random_rotate_and_crop": false,
|
| 23 |
+
"aspect_ratio_bucketing": true,
|
| 24 |
+
"latent_caching": true,
|
| 25 |
+
"clear_cache_before_training": true,
|
| 26 |
+
"learning_rate_scheduler": "CONSTANT",
|
| 27 |
+
"learning_rate": 3e-06,
|
| 28 |
+
"learning_rate_warmup_steps": 200,
|
| 29 |
+
"learning_rate_cycles": 1,
|
| 30 |
+
"epochs": 100,
|
| 31 |
+
"batch_size": 1,
|
| 32 |
+
"gradient_accumulation_steps": 1,
|
| 33 |
+
"ema": "OFF",
|
| 34 |
+
"ema_decay": 0.999,
|
| 35 |
+
"ema_update_step_interval": 5,
|
| 36 |
+
"train_device": "cuda",
|
| 37 |
+
"temp_device": "cpu",
|
| 38 |
+
"train_dtype": "FLOAT_16",
|
| 39 |
+
"fallback_train_dtype": "BFLOAT_16",
|
| 40 |
+
"only_cache": false,
|
| 41 |
+
"resolution": "1024",
|
| 42 |
+
"attention_mechanism": "XFORMERS",
|
| 43 |
+
"align_prop": false,
|
| 44 |
+
"align_prop_probability": 0.1,
|
| 45 |
+
"align_prop_loss": "AESTHETIC",
|
| 46 |
+
"align_prop_weight": 0.01,
|
| 47 |
+
"align_prop_steps": 20,
|
| 48 |
+
"align_prop_truncate_steps": 0.5,
|
| 49 |
+
"align_prop_cfg_scale": 7.0,
|
| 50 |
+
"mse_strength": 1.0,
|
| 51 |
+
"mae_strength": 0.0,
|
| 52 |
+
"vb_loss_strength": 1.0,
|
| 53 |
+
"min_snr_gamma": 0.0,
|
| 54 |
+
"dropout_probability": 0.0,
|
| 55 |
+
"loss_scaler": "NONE",
|
| 56 |
+
"learning_rate_scaler": "NONE",
|
| 57 |
+
"offset_noise_weight": 0.0,
|
| 58 |
+
"perturbation_noise_weight": 0.0,
|
| 59 |
+
"rescale_noise_scheduler_to_zero_terminal_snr": false,
|
| 60 |
+
"force_v_prediction": false,
|
| 61 |
+
"force_epsilon_prediction": false,
|
| 62 |
+
"min_noising_strength": 0.0,
|
| 63 |
+
"max_noising_strength": 1.0,
|
| 64 |
+
"noising_weight": 0.0,
|
| 65 |
+
"noising_bias": 0.5,
|
| 66 |
+
"unet": {
|
| 67 |
+
"__version": 0,
|
| 68 |
+
"model_name": "",
|
| 69 |
+
"train": true,
|
| 70 |
+
"stop_training_after": 0,
|
| 71 |
+
"stop_training_after_unit": "NEVER",
|
| 72 |
+
"learning_rate": null,
|
| 73 |
+
"weight_dtype": "NONE"
|
| 74 |
+
},
|
| 75 |
+
"prior": {
|
| 76 |
+
"__version": 0,
|
| 77 |
+
"model_name": "",
|
| 78 |
+
"train": true,
|
| 79 |
+
"stop_training_after": 0,
|
| 80 |
+
"stop_training_after_unit": "NEVER",
|
| 81 |
+
"learning_rate": null,
|
| 82 |
+
"weight_dtype": "NONE"
|
| 83 |
+
},
|
| 84 |
+
"text_encoder": {
|
| 85 |
+
"__version": 0,
|
| 86 |
+
"model_name": "",
|
| 87 |
+
"train": true,
|
| 88 |
+
"stop_training_after": 30,
|
| 89 |
+
"stop_training_after_unit": "EPOCH",
|
| 90 |
+
"learning_rate": null,
|
| 91 |
+
"weight_dtype": "NONE"
|
| 92 |
+
},
|
| 93 |
+
"text_encoder_layer_skip": 0,
|
| 94 |
+
"text_encoder_2": {
|
| 95 |
+
"__version": 0,
|
| 96 |
+
"model_name": "",
|
| 97 |
+
"train": true,
|
| 98 |
+
"stop_training_after": 30,
|
| 99 |
+
"stop_training_after_unit": "EPOCH",
|
| 100 |
+
"learning_rate": null,
|
| 101 |
+
"weight_dtype": "NONE"
|
| 102 |
+
},
|
| 103 |
+
"text_encoder_2_layer_skip": 0,
|
| 104 |
+
"vae": {
|
| 105 |
+
"__version": 0,
|
| 106 |
+
"model_name": "",
|
| 107 |
+
"train": true,
|
| 108 |
+
"stop_training_after": null,
|
| 109 |
+
"stop_training_after_unit": "NEVER",
|
| 110 |
+
"learning_rate": null,
|
| 111 |
+
"weight_dtype": "FLOAT_32"
|
| 112 |
+
},
|
| 113 |
+
"effnet_encoder": {
|
| 114 |
+
"__version": 0,
|
| 115 |
+
"model_name": "TODO: MANUALLY DOWNLOAD EFFNET WEIGHTS",
|
| 116 |
+
"train": true,
|
| 117 |
+
"stop_training_after": null,
|
| 118 |
+
"stop_training_after_unit": "NEVER",
|
| 119 |
+
"learning_rate": null,
|
| 120 |
+
"weight_dtype": "FLOAT_16"
|
| 121 |
+
},
|
| 122 |
+
"decoder": {
|
| 123 |
+
"__version": 0,
|
| 124 |
+
"model_name": "stabilityai/stable-cascade",
|
| 125 |
+
"train": true,
|
| 126 |
+
"stop_training_after": null,
|
| 127 |
+
"stop_training_after_unit": "NEVER",
|
| 128 |
+
"learning_rate": null,
|
| 129 |
+
"weight_dtype": "FLOAT_16"
|
| 130 |
+
},
|
| 131 |
+
"decoder_text_encoder": {
|
| 132 |
+
"__version": 0,
|
| 133 |
+
"model_name": "",
|
| 134 |
+
"train": true,
|
| 135 |
+
"stop_training_after": null,
|
| 136 |
+
"stop_training_after_unit": "NEVER",
|
| 137 |
+
"learning_rate": null,
|
| 138 |
+
"weight_dtype": "NONE"
|
| 139 |
+
},
|
| 140 |
+
"decoder_vqgan": {
|
| 141 |
+
"__version": 0,
|
| 142 |
+
"model_name": "",
|
| 143 |
+
"train": true,
|
| 144 |
+
"stop_training_after": null,
|
| 145 |
+
"stop_training_after_unit": "NEVER",
|
| 146 |
+
"learning_rate": null,
|
| 147 |
+
"weight_dtype": "FLOAT_16"
|
| 148 |
+
},
|
| 149 |
+
"masked_training": false,
|
| 150 |
+
"unmasked_probability": 0.1,
|
| 151 |
+
"unmasked_weight": 0.1,
|
| 152 |
+
"normalize_masked_area_loss": false,
|
| 153 |
+
"embeddings": [
|
| 154 |
+
{
|
| 155 |
+
"__version": 0,
|
| 156 |
+
"model_name": "",
|
| 157 |
+
"train": true,
|
| 158 |
+
"stop_training_after": null,
|
| 159 |
+
"stop_training_after_unit": "NEVER",
|
| 160 |
+
"token_count": 1,
|
| 161 |
+
"initial_embedding_text": "*",
|
| 162 |
+
"weight_dtype": "FLOAT_32"
|
| 163 |
+
}
|
| 164 |
+
],
|
| 165 |
+
"embedding_weight_dtype": "FLOAT_32",
|
| 166 |
+
"lora_model_name": "",
|
| 167 |
+
"lora_rank": 16,
|
| 168 |
+
"lora_alpha": 1.0,
|
| 169 |
+
"lora_weight_dtype": "FLOAT_32",
|
| 170 |
+
"optimizer": {
|
| 171 |
+
"__version": 0,
|
| 172 |
+
"optimizer": "ADAFACTOR",
|
| 173 |
+
"adam_w_mode": false,
|
| 174 |
+
"alpha": null,
|
| 175 |
+
"amsgrad": false,
|
| 176 |
+
"beta1": null,
|
| 177 |
+
"beta2": null,
|
| 178 |
+
"beta3": null,
|
| 179 |
+
"bias_correction": false,
|
| 180 |
+
"block_wise": false,
|
| 181 |
+
"capturable": false,
|
| 182 |
+
"centered": false,
|
| 183 |
+
"clip_threshold": 1.0,
|
| 184 |
+
"d0": null,
|
| 185 |
+
"d_coef": null,
|
| 186 |
+
"dampening": null,
|
| 187 |
+
"decay_rate": -0.8,
|
| 188 |
+
"decouple": false,
|
| 189 |
+
"differentiable": false,
|
| 190 |
+
"eps": 1e-30,
|
| 191 |
+
"eps2": 0.001,
|
| 192 |
+
"foreach": false,
|
| 193 |
+
"fsdp_in_use": false,
|
| 194 |
+
"fused": false,
|
| 195 |
+
"growth_rate": null,
|
| 196 |
+
"initial_accumulator_value": null,
|
| 197 |
+
"is_paged": false,
|
| 198 |
+
"log_every": null,
|
| 199 |
+
"lr_decay": null,
|
| 200 |
+
"max_unorm": null,
|
| 201 |
+
"maximize": false,
|
| 202 |
+
"min_8bit_size": null,
|
| 203 |
+
"momentum": null,
|
| 204 |
+
"nesterov": false,
|
| 205 |
+
"no_prox": false,
|
| 206 |
+
"optim_bits": null,
|
| 207 |
+
"percentile_clipping": null,
|
| 208 |
+
"relative_step": false,
|
| 209 |
+
"safeguard_warmup": false,
|
| 210 |
+
"scale_parameter": false,
|
| 211 |
+
"stochastic_rounding": true,
|
| 212 |
+
"use_bias_correction": false,
|
| 213 |
+
"use_triton": false,
|
| 214 |
+
"warmup_init": false,
|
| 215 |
+
"weight_decay": 0.0
|
| 216 |
+
},
|
| 217 |
+
"optimizer_defaults": {
|
| 218 |
+
"ADAFACTOR": {
|
| 219 |
+
"__version": 0,
|
| 220 |
+
"optimizer": "ADAFACTOR",
|
| 221 |
+
"adam_w_mode": false,
|
| 222 |
+
"alpha": null,
|
| 223 |
+
"amsgrad": false,
|
| 224 |
+
"beta1": null,
|
| 225 |
+
"beta2": null,
|
| 226 |
+
"beta3": null,
|
| 227 |
+
"bias_correction": false,
|
| 228 |
+
"block_wise": false,
|
| 229 |
+
"capturable": false,
|
| 230 |
+
"centered": false,
|
| 231 |
+
"clip_threshold": 1.0,
|
| 232 |
+
"d0": null,
|
| 233 |
+
"d_coef": null,
|
| 234 |
+
"dampening": null,
|
| 235 |
+
"decay_rate": -0.8,
|
| 236 |
+
"decouple": false,
|
| 237 |
+
"differentiable": false,
|
| 238 |
+
"eps": 1e-30,
|
| 239 |
+
"eps2": 0.001,
|
| 240 |
+
"foreach": false,
|
| 241 |
+
"fsdp_in_use": false,
|
| 242 |
+
"fused": false,
|
| 243 |
+
"growth_rate": null,
|
| 244 |
+
"initial_accumulator_value": null,
|
| 245 |
+
"is_paged": false,
|
| 246 |
+
"log_every": null,
|
| 247 |
+
"lr_decay": null,
|
| 248 |
+
"max_unorm": null,
|
| 249 |
+
"maximize": false,
|
| 250 |
+
"min_8bit_size": null,
|
| 251 |
+
"momentum": null,
|
| 252 |
+
"nesterov": false,
|
| 253 |
+
"no_prox": false,
|
| 254 |
+
"optim_bits": null,
|
| 255 |
+
"percentile_clipping": null,
|
| 256 |
+
"relative_step": false,
|
| 257 |
+
"safeguard_warmup": false,
|
| 258 |
+
"scale_parameter": false,
|
| 259 |
+
"stochastic_rounding": true,
|
| 260 |
+
"use_bias_correction": false,
|
| 261 |
+
"use_triton": false,
|
| 262 |
+
"warmup_init": false,
|
| 263 |
+
"weight_decay": 0.0
|
| 264 |
+
}
|
| 265 |
+
},
|
| 266 |
+
"sample_definition_file_name": "training_samples/samples.json",
|
| 267 |
+
"samples": null,
|
| 268 |
+
"sample_after": 10,
|
| 269 |
+
"sample_after_unit": "MINUTE",
|
| 270 |
+
"sample_image_format": "JPG",
|
| 271 |
+
"samples_to_tensorboard": true,
|
| 272 |
+
"non_ema_sampling": true,
|
| 273 |
+
"backup_after": 30,
|
| 274 |
+
"backup_after_unit": "MINUTE",
|
| 275 |
+
"rolling_backup": false,
|
| 276 |
+
"rolling_backup_count": 3,
|
| 277 |
+
"backup_before_save": true,
|
| 278 |
+
"save_after": 0,
|
| 279 |
+
"save_after_unit": "NEVER"
|
| 280 |
+
}
|