Attempted to set the storage of a tensor on device "cpu" to a storage on different device "cuda:0".

Question

问题遇到的现象和发生背景

pytorch 1.12.0
载入pt的时候，原本保存的是在cpu上面，现在我想载入cuda:0上，使用map_location报错

问题相关代码

net.load_state_dict(torch.load(model_weight_path, map_location=torch.device('cuda:0')))

运行结果及报错内容

RuntimeError: Attempted to set the storage of a tensor on device "cpu" to a storage on different device "cuda:0". This is no longer allowed; the devices must match.

我的解答思路和尝试过的方法

github上有人问出类似问题，但国内这边没有结果，pytorch官网也没给出load的特殊变化，无法定位根本问题