Value cannot be converted to type float without overflow: inf

1

Hello all, I am trying to use the learning rate adaptions, and I got the error:

value cannot be converted to type float without overflow: inf

with code:

optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.0)
scheduler = ReduceLROnPlateau(optimizer, 'min')
for epoch in range(10):
    for i, batch in enumerate(dataloader):
    	output = model(batch)
    	loss = criteria(output)
    	scheduler.step(loss)

I wonder why does this happen and how to solve it??
Also, in the case that no lr scheduler but the momentum = 0.9, the run time error occurred.

1 Like
2

Just a guess: are there any inf values appearing in your output or loss?

3

I think there are no inf value,
when the code is:
optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.0)

for epoch in range(10):
    for i, batch in enumerate(dataloader):
    	output = model(batch)
    	loss = criteria(output)
        print loss
    	optimizer .step()

There are no inf values.