解决Diffuers训练LoRA，推理时LoRA权重不起作用的问题_diffusers 训练的lora效果不好

原因是最新版的Diffusers更新之后，利用新训练的LoRA权重做推理时，加载模型的API更新了，需要将原来的

model_path = "预训练的LoRA权重" 
pipe = StableDiffusionPipeline.from_pretrained("stable diffusion底模路径",
                                               torch_dtype=torch.float16, safety_checker = None, requires_safety_checker = False)
pipe.unet.load_attn_procs(model_path)
1
2
3
4

替换为：

pipeline = StableDiffusionPipeline.from_pretrained("stable diffusion底模路径",
                                               torch_dtype=torch.float16, safety_checker = None, requires_safety_checker = False).to("cuda")
pipeline.load_lora_weights("预训练的LoRA权重", weight_name="pytorch_lora_weights.safetensors")
1
2
3

或

pipeline = AutoPipelineForText2Image.from_pretrained("stable diffusion底模路径",
                                                     torch_dtype=torch.float16, safety_checker = None, requires_safety_checker = False).to("cuda")
pipeline.load_lora_weights("预训练的LoRA权重", weight_name="pytorch_lora_weights.safetensors")
1
2
3