RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_addmm)

I know this question has been asnwered a lot of times but for some reason all the tips I got didn’t work. Here’s the code, I am trying to use PGexplainer from pytorch geometric to explain graphs using a already trained model. Here’s the code:
from torch_geometric.nn import GCNConv
from torch_geometric.nn import global_mean_pool
import torch.nn.functional as F
import torch.nn as nn

class GCN(nn.Module):
def init(self, input_dim, hidden_channels):
super(GCN, self).init()
self.conv1 = GCNConv(input_dim, hidden_channels)
self.conv2 = GCNConv(hidden_channels, hidden_channels)
self.conv3 = GCNConv(hidden_channels, hidden_channels)
self.lin = nn.Linear(hidden_channels, 3)

def forward(self, x, edge_index, batch):
    x = self.conv1(x, edge_index)
    x = F.relu(x)
    x = self.conv2(x, edge_index)
    x = F.relu(x)
    x = self.conv3(x, edge_index)
    x = gnn.global_mean_pool(x, batch)
    x = self.lin(x)
    return x

model = GCN(input_dim = 3, hidden_channels = 256)
model.load_state_dict(torch.load(‘model1.pth’, map_location=device))

from torch_geometric.explain import Explainer, PGExplainer
explainer = Explainer(
algorithm=PGExplainer(epochs=10, lr=0.003),
# Include only the top 10 most important edges:
threshold_config=dict(threshold_type=‘topk’, value=10),

PGExplainer needs to be trained separately since it is a parametric

explainer i.e it uses a neural network to generate explanations:

for epoch in range(10):
for data in val_dataloader:
# Move the data to the chosen device
data =
x = data.x
x =

    edge_index = data.edge_index
    edge_index =

    target = data.y
    target =

    batch = data.batch
    batch =

    # Pass the data to the model's train method
    loss = explainer.algorithm.train(epoch, model, x, edge_index, target=target, batch=batch)

The error always occurs at the ‘loss =…’

Is the error message showing the layer which is raising the issue?
If not, could you try to isolate the layer by adding debug print statements or by stepping through the code with a debugger?
I don’t see an obvious issue in your code, but am also not deeply familiar with PyG.

Got the answer from the pytorch geometric GitHub, the problem was from algorithm=PGExplainer. Should have been