accelerate训练SD-LoRA，解决ValueError: Attempting to unscale FP16 gradients.问题-CSDN博客

本文链接：https://blog.csdn.net/qq_38327353/article/details/139115441

训练指令需从原来的：

accelerate launch –mixed_precision=“fp16” train_text_to_image_lora.py
–pretrained_model_name_or_path=“F:/XYX/Documents/AITools/ControlNetImplement/ControlNetTest/models–runwayml–stable-diffusion-v1-5”
–dataset_name=“dataset-300mul15” --caption_column=“text” --resolution=512 --train_batch_size=1 --num_train_epochs=100 --checkpointing_steps=5000 --learning_rate=1e-04 --lr_scheduler=“constant” --lr_warmup_steps=0 --seed=42 --output_dir=“output/sd1d5-lora” --report_to=“wandb”

变换一下顺序：

accelerate launch train_text_to_image_lora.py –mixed_precision=“fp16”
–pretrained_model_name_or_path=“F:/XYX/Documents/AITools/ControlNetImplement/ControlNetTest/models–runwayml–stable-diffusion-v1-5”
–dataset_name=“dataset-300mul15” --caption_column=“text” --resolution=512 --train_batch_size=1 --num_train_epochs=100 --checkpointing_steps=5000 --learning_rate=1e-04 --lr_scheduler=“constant” --lr_warmup_steps=0 --seed=42 --output_dir=“output/sd1d5-lora” --report_to=“wandb”