i’m trying to train my model in PyTorch but I have the following error:
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
<ipython-input-47-4c3349f66d9b> in <module>
26
27 # Backpropagate loss
---> 28 loss.backward()
29
30 # Steps
~/anaconda3/lib/python3.8/site-packages/torch/tensor.py in backward(self, gradient, retain_graph, create_graph)
219 retain_graph=retain_graph,
220 create_graph=create_graph)
--> 221 torch.autograd.backward(self, gradient, retain_graph, create_graph)
222
223 def register_hook(self, hook):
~/anaconda3/lib/python3.8/site-packages/torch/autograd/__init__.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables)
128 retain_graph = create_graph
129
--> 130 Variable._execution_engine.run_backward(
131 tensors, grad_tensors_, retain_graph, create_graph,
132 allow_unreachable=True) # allow_unreachable flag
RuntimeError: Found dtype Half but expected Float
My code:
criterion_KL = torch.nn.KLDivLoss()
logits_per_image, logits_per_text = model(images.to(device), texts.to(device)) # both NxN matrix
optim.zero_grad()
labels = labels_matrix(tax) # returns a NxN matrix, len(tax) = N
loss_i = criterion_KL(logits_per_image, labels)
loss_j = criterion_KL(logits_per_text, labels)
loss = (loss_i + loss_j)/2
print(loss_i, loss_j, loss)
# tensor(-0.6089, device='cuda:0', grad_fn=<KlDivBackward>) tensor(-0.6089, device='cuda:0', grad_fn=<KlDivBackward>) tensor(-0.6089, device='cuda:0', grad_fn=<DivBackward0>)
print(loss_i.dtype, loss_j.dtype, loss.dtype)
# torch.float32 torch.float32 torch.float32
I have no idea why this error is happening. Tried to cast the labels tensor to float, but didn’t have results.
Any help?