訓練 LoRA (Low-Rank Adaptation) 模型是一個有效的技術，可以在不大幅度改變原有模型的情況下，針對特定任務進行微調。這裡是使用 Stable Diffusion WebUI 訓練 LoRA 模型的詳細步驟

步驟 1：準備環境

首先，確保你的環境已經安裝並配置了 Stable Diffusion WebUI 和所需的依賴項。如果還未安裝，可以參考以下命令：

安裝依賴項

pip install torch torchvision transformers diffusers

步驟 2：下載並準備數據集

訓練 LoRA 模型需要準備合適的數據集。這些數據集應該包含你希望模型學習的新內容或風格。數據集可以是圖像對（圖像和其對應的描述文本）。

git clone https://github.com/derrian-distro/LoRA_Easy_Training_Scripts.git
#複製儲存庫
cd LoRA_Easy_Training_Scripts
git submodule init
git submodule update
cd sd_scripts
pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu116
pip install --upgrade -r requirements.txt
pip install -U xformers

新增訓練設定檔trainingconfig.json

vim trainingconfig.json

填入以下內容(雙斜線的註解記得刪除) LoRA的總訓練步數計算公式為： 訓練圖片數量 × 重複次數 ÷ train_batch_size × epoch

{
  //基於何種模型訓練
  "pretrained_model_name_or_path": "/home/user/桌面/heralora/anything-v4.5-pruned.ckpt",
  "v2": false,
  "v_parameterization": false,
  //紀錄檔輸出目錄
  "logging_dir": "/home/user/桌面/heralora/log_dir/",
  //訓練資料目錄
  "train_data_dir": "/home/user/桌面/heralora/image_dir/",
  //註冊目錄
  "reg_data_dir": "/home/user/桌面/heralora/reg_dir/",
  //輸出目錄
  "output_dir": "/home/user/桌面/heralora/output_dir",
  //訓練的圖片最大長寬
  "max_resolution": "512,512",
  //學習率
  "learning_rate": "1e-5",
  "lr_scheduler": "constant_with_warmup",
  "lr_warmup": "5",
  "train_batch_size": 3,
  //訓練時期
  "epoch": "4",
  "save_every_n_epochs": "",
  "mixed_precision": "fp16",
  "save_precision": "fp16",
  "seed": "",
  "num_cpu_threads_per_process": 32,
  "cache_latents": true,
  "caption_extension": ".txt",
  "enable_bucket": true,
  "gradient_checkpointing": false,
  "full_fp16": false,
  "no_token_padding": false,
  "stop_text_encoder_training": 0,
  "use_8bit_adam": true,
  "xformers": true,
  "save_model_as": "safetensors",
  "shuffle_caption": true,
  "save_state": false,
  "resume": "",
  "prior_loss_weight": 1.0,
  "text_encoder_lr": "1.5e-5",
  "unet_lr": "1.5e-4",
  "network_dim": 128,
  "lora_network_weights": "",
  "color_aug": false,
  "flip_aug": false,
  "clip_skip": 2,
  "mem_eff_attn": false,
  "output_name": "",
  "model_list": "",
  "max_token_length": "150",
  "max_train_epochs": "",
  "max_data_loader_n_workers": "",
  "network_alpha": 128,
  "training_comment": "",
  "keep_tokens": 2,
  "lr_scheduler_num_cycles": "",
  "lr_scheduler_power": "",
  "persistent_data_loader_workers": true,
  "bucket_no_upscale": true,
  "random_crop": false,
  "caption_dropout_every_n_epochs": 0.0,
  "caption_dropout_rate": 0
}