RuntimeError('Scale factor has NaN entries')

[RaphaelWW]

Hi,

thank you for providing all the code and the hyperparameter specifications for the experiments. Unfortunately, I'm having issues reproducing your results for the Glow architecture with MNIST and FashionMNIST. When I run the provided command, the model starts training but stops after 15 epochs with RuntimeError("Scale factor has NaN entries'). Before the error, the loss increases up to 2.8e+12.

It would be great if you could find why it fails or tell me what I am doing wrong :slightly_smiling_face:

Traceback (most recent call last):
  File "/nfs/homedirs/wildr/flows_ood/train_unsup.py", line 361, in <module>
    train(epoch, net, trainloader, device, optimizer, loss_fn, args.max_grad_norm, writer,
  File "/nfs/homedirs/wildr/flows_ood/train_unsup.py", line 68, in train
    z = net(x)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/parallel/data_parallel.py", line 153, in forward
    return self.module(*inputs[0], **kwargs[0])
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/nfs/homedirs/wildr/flows_ood/flow_ssl/glow/glow.py", line 24, in forward
    return self.body(x)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/container.py", line 100, in forward
    input = module(input)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/container.py", line 100, in forward
    input = module(input)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/nfs/homedirs/wildr/flows_ood/flow_ssl/invertible/parts.py", line 121, in forward
    return self.module1(x), self.module2(z)
  File "/nfs/homedirs/wildr/anaconda3/envs/ood_flows/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/nfs/homedirs/wildr/flows_ood/flow_ssl/realnvp/coupling_layer.py", line 327, in forward
    raise RuntimeError('Scale factor has NaN entries')
RuntimeError: Scale factor has NaN entries

[AissamDjahnine]

Any updates about this error ?