Hi, while training the model, I constantly get this warning:
UserWarning: User provided device_type of ‘cuda’, but CUDA is unavailable. Disabling warnings.warn(‘User provided device_type of ‘cuda’, but CUDA is not available. Disabling’)
which seems to indicate that I am running something on CUDA but my whole code has been running on the CPU. To prove that, I checked the .is_cuda property of all my tensors and is False.
dgl==0.9.0
torch==1.10.1
line loss_output.backward() generates this warning
Does your data loader have the device argument as cuda rather than cpu?
No, I don’t use device argument. Here is how I create DataLoader:
# Define training EdgeDataLoader
train_dataloader = dgl.dataloading.DataLoader(
graph, # The graph
train_eid_dict, # The edges to iterate over
sampler, # The neighbor sampler
batch_size=batch_size, # Batch size
shuffle=True, # Whether to shuffle the nodes for every epoch
drop_last=False, # Whether to drop the last incomplete batch
num_workers=sampling_workers, # Number of sampling processes
)
I also opened an issue on torch because I thought they were also relevant so maybe it helps, if not sorry: https://github.com/pytorch/pytorch/issues/83733
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.