报错：RuntimeError: Attempting to deserialize object on CUDA device 2 but torch.cuda.device_count() is 1

报错：RuntimeError: Attempting to deserialize object on CUDA device 2 but torch.cuda.device_count() is 1
原因：在使用Pytorch加载模型时报错。加载的模型是用两个GPU训练的，而加载模型的电脑只有一个GPU，所以会出错。
解决：model = torch.load(model_path)
改为：model = torch.load(model_path, map_location='cuda:0')
如果是4块到2块：就把map_location改为：map_location={'cuda:1': 'cuda:0'}