把在NPU上训练的pt模型保存为CPU # Npu模型路径 model = torch.load(opt.pth_path, map_location='cpu') # 保存模型 torch.save(model, opt.output_pth_path)