Error: Expected all tensors to be on the same device, While all are on same device

Question

我不知道为什么我会收到这个错误，而我的所有张量都在同一个设备“cuda”上。这是代码

def train(self):
    self.model.train()
    self.optimizer.zero_grad()  # Clear gradients.
    print("X is",self.XT.get_device())
    print("edge_index_ is", self.edge_index_.get_device())
    print("R_train_mask is", self.R_train_mask.get_device())

    print("datas.y is", self.datas.y.get_device())
    print("edge_index_ is", self.edge_index_.get_device())
    out = self.model(self.XT, self.edge_index_)  # Perform a single forward pass.
    loss = self.criterion(out[self.R_train_mask],
                          self.datas.y[self.R_train_mask])

这里是错误

X is 0
edge_index_ is 0
R_train_mask is 0
datas.y is 0
edge_index_ is 0
Traceback (most recent call last):
  File "/home/adnan/GNNPaperCodes/PyGCL/examples/GCAPaperAd.py", line 126, in <module>
    main()
  File "/home/adnan/GNNPaperCodes/PyGCL/examples/GCAPaperAd.py", line 91, in main
    topfeatures=tf.TopFeaturesFind()
  File "/home/adnan/GNNPaperCodes/PyGCL/examples/PatchFinder.py", line 64, in TopFeaturesFind
    loss = self.train()
  File "/home/adnan/GNNPaperCodes/PyGCL/examples/PatchFinder.py", line 154, in train
    out = self.model(self.XT, self.edge_index_)  # Perform a single forward pass.
  File "/home/adnan/.conda/envs/forGit/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/adnan/GNNPaperCodes/PyGCL/examples/PatchFinder.py", line 211, in forward
    x = self.conv1(x, edge_index_)
  File "/home/adnan/.conda/envs/forGit/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/adnan/.conda/envs/forGit/lib/python3.9/site-packages/torch_geometric/nn/conv/gcn_conv.py", line 181, in forward
    x = self.lin(x)
  File "/home/adnan/.conda/envs/forGit/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/adnan/.conda/envs/forGit/lib/python3.9/site-packages/torch_geometric/nn/dense/linear.py", line 103, in forward
    return F.linear(x, self.weight, self.bias)
  File "/home/adnan/.conda/envs/forGit/lib/python3.9/site-packages/torch/nn/functional.py", line 1848, in linear
    return torch._C._nn.linear(input, weight, bias)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_mm)

很明显，如果我将设备更改为 'cpu'

，一切正常

Answer 1

您忘记将 self.model 传输到设备。实际上，模型具有必须传输到正确设备的权重，因为它们与您的输入交互。

你可以用 self.model.to("cuda") 来完成。

Error: Expected all tensors to be on the same device, While all are on same device

Error: Expected all tensors to be on the same device, While all are on same device

pytorch

tensor