Upload SDXL v1 Lora 12GB Optimize Test.json
Browse files
SDXL v1 Lora 12GB Optimize Test.json
ADDED
@@ -0,0 +1,152 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"training_method": "LORA",
|
3 |
+
"model_type": "STABLE_DIFFUSION_XL_10_BASE",
|
4 |
+
"debug_mode": false,
|
5 |
+
"debug_dir": "debug",
|
6 |
+
"workspace_dir": "workspace/run",
|
7 |
+
"cache_dir": "workspace-cache/run",
|
8 |
+
"tensorboard": true,
|
9 |
+
"tensorboard_expose": false,
|
10 |
+
"continue_last_backup": false,
|
11 |
+
"weight_dtype": "FLOAT_16",
|
12 |
+
"output_dtype": "FLOAT_32",
|
13 |
+
"output_model_format": "SAFETENSORS",
|
14 |
+
"gradient_checkpointing": true,
|
15 |
+
"circular_mask_generation": false,
|
16 |
+
"random_rotate_and_crop": false,
|
17 |
+
"aspect_ratio_bucketing": true,
|
18 |
+
"latent_caching": true,
|
19 |
+
"clear_cache_before_training": true,
|
20 |
+
"learning_rate_scheduler": "CONSTANT",
|
21 |
+
"learning_rate": 0.0003,
|
22 |
+
"learning_rate_warmup_steps": 200,
|
23 |
+
"learning_rate_cycles": 1,
|
24 |
+
"epochs": 20,
|
25 |
+
"batch_size": 4,
|
26 |
+
"gradient_accumulation_steps": 1,
|
27 |
+
"ema": "OFF",
|
28 |
+
"ema_decay": 0.999,
|
29 |
+
"ema_update_step_interval": 5,
|
30 |
+
"train_device": "cuda",
|
31 |
+
"temp_device": "cpu",
|
32 |
+
"train_dtype": "FLOAT_16",
|
33 |
+
"fallback_train_dtype": "BFLOAT_16",
|
34 |
+
"only_cache": false,
|
35 |
+
"resolution": "896",
|
36 |
+
"attention_mechanism": "XFORMERS",
|
37 |
+
"align_prop": false,
|
38 |
+
"align_prop_probability": 0.1,
|
39 |
+
"align_prop_loss": "AESTHETIC",
|
40 |
+
"align_prop_weight": 0.01,
|
41 |
+
"align_prop_steps": 20,
|
42 |
+
"align_prop_truncate_steps": 0.5,
|
43 |
+
"align_prop_cfg_scale": 7.0,
|
44 |
+
"mse_strength": 1.0,
|
45 |
+
"mae_strength": 0.0,
|
46 |
+
"vb_loss_strength": 1.0,
|
47 |
+
"loss_scaler": "NONE",
|
48 |
+
"learning_rate_scaler": "NONE",
|
49 |
+
"train_unet": true,
|
50 |
+
"train_unet_epochs": 100000,
|
51 |
+
"unet_learning_rate": null,
|
52 |
+
"offset_noise_weight": 0.0,
|
53 |
+
"perturbation_noise_weight": 0.0,
|
54 |
+
"rescale_noise_scheduler_to_zero_terminal_snr": false,
|
55 |
+
"force_v_prediction": false,
|
56 |
+
"force_epsilon_prediction": false,
|
57 |
+
"min_noising_strength": 0.0,
|
58 |
+
"max_noising_strength": 1.0,
|
59 |
+
"noising_weight": 0.0,
|
60 |
+
"noising_bias": 0.5,
|
61 |
+
"unet_weight_dtype": "NONE",
|
62 |
+
"train_prior": true,
|
63 |
+
"train_prior_epochs": 10000,
|
64 |
+
"prior_learning_rate": null,
|
65 |
+
"prior_weight_dtype": "NONE",
|
66 |
+
"train_text_encoder": false,
|
67 |
+
"train_text_encoder_epochs": 30,
|
68 |
+
"text_encoder_learning_rate": null,
|
69 |
+
"text_encoder_layer_skip": 0,
|
70 |
+
"text_encoder_weight_dtype": "NONE",
|
71 |
+
"train_text_encoder_2": false,
|
72 |
+
"train_text_encoder_2_epochs": 30,
|
73 |
+
"text_encoder_2_learning_rate": null,
|
74 |
+
"text_encoder_2_layer_skip": 0,
|
75 |
+
"text_encoder_2_weight_dtype": "NONE",
|
76 |
+
"vae_model_name": "",
|
77 |
+
"vae_weight_dtype": "FLOAT_32",
|
78 |
+
"effnet_encoder_model_name": "",
|
79 |
+
"effnet_encoder_weight_dtype": "NONE",
|
80 |
+
"decoder_model_name": "",
|
81 |
+
"decoder_weight_dtype": "NONE",
|
82 |
+
"decoder_text_encoder_weight_dtype": "NONE",
|
83 |
+
"decoder_vqgan_weight_dtype": "NONE",
|
84 |
+
"masked_training": false,
|
85 |
+
"unmasked_probability": 0.1,
|
86 |
+
"unmasked_weight": 0.1,
|
87 |
+
"normalize_masked_area_loss": false,
|
88 |
+
"embedding_model_names": [],
|
89 |
+
"token_count": 1,
|
90 |
+
"initial_embedding_text": "*",
|
91 |
+
"embedding_weight_dtype": "FLOAT_32",
|
92 |
+
"lora_model_name": "",
|
93 |
+
"lora_rank": 16,
|
94 |
+
"lora_alpha": 1.0,
|
95 |
+
"lora_weight_dtype": "FLOAT_32",
|
96 |
+
"optimizer": "ADAMW",
|
97 |
+
"optimizer_adam_w_mode": false,
|
98 |
+
"optimizer_alpha": null,
|
99 |
+
"optimizer_amsgrad": false,
|
100 |
+
"optimizer_beta1": 0.9,
|
101 |
+
"optimizer_beta2": 0.999,
|
102 |
+
"optimizer_beta3": null,
|
103 |
+
"optimizer_bias_correction": false,
|
104 |
+
"optimizer_block_wise": false,
|
105 |
+
"optimizer_capturable": false,
|
106 |
+
"optimizer_centered": false,
|
107 |
+
"optimizer_clip_threshold": null,
|
108 |
+
"optimizer_d0": null,
|
109 |
+
"optimizer_d_coef": null,
|
110 |
+
"optimizer_dampening": null,
|
111 |
+
"optimizer_decay_rate": null,
|
112 |
+
"optimizer_decouple": false,
|
113 |
+
"optimizer_differentiable": false,
|
114 |
+
"optimizer_eps": 1e-08,
|
115 |
+
"optimizer_eps2": null,
|
116 |
+
"optimizer_foreach": false,
|
117 |
+
"optimizer_fsdp_in_use": false,
|
118 |
+
"optimizer_fused": true,
|
119 |
+
"optimizer_growth_rate": null,
|
120 |
+
"optimizer_initial_accumulator_value": null,
|
121 |
+
"optimizer_is_paged": false,
|
122 |
+
"optimizer_log_every": null,
|
123 |
+
"optimizer_lr_decay": null,
|
124 |
+
"optimizer_max_unorm": null,
|
125 |
+
"optimizer_maximize": false,
|
126 |
+
"optimizer_min_8bit_size": null,
|
127 |
+
"optimizer_momentum": null,
|
128 |
+
"optimizer_nesterov": false,
|
129 |
+
"optimizer_no_prox": false,
|
130 |
+
"optimizer_optim_bits": null,
|
131 |
+
"optimizer_percentile_clipping": null,
|
132 |
+
"optimizer_relative_step": false,
|
133 |
+
"optimizer_safeguard_warmup": false,
|
134 |
+
"optimizer_scale_parameter": false,
|
135 |
+
"optimizer_use_bias_correction": false,
|
136 |
+
"optimizer_use_triton": false,
|
137 |
+
"optimizer_warmup_init": false,
|
138 |
+
"optimizer_weight_decay": 0.01,
|
139 |
+
"sample_definition_file_name": "training_samples/samples.json",
|
140 |
+
"sample_after": 10,
|
141 |
+
"sample_after_unit": "NEVER",
|
142 |
+
"sample_image_format": "JPG",
|
143 |
+
"samples_to_tensorboard": true,
|
144 |
+
"non_ema_sampling": true,
|
145 |
+
"backup_after": 10,
|
146 |
+
"backup_after_unit": "MINUTE",
|
147 |
+
"rolling_backup": false,
|
148 |
+
"rolling_backup_count": 3,
|
149 |
+
"backup_before_save": true,
|
150 |
+
"save_after": 2,
|
151 |
+
"save_after_unit": "EPOCH"
|
152 |
+
}
|