search

salesforce / pytorch-qrnn

PyTorch implementation of the Quasi-Recurrent Neural Network - up to 16 times faster than NVIDIA's cuDNN LSTM
BSD 3-Clause "New" or "Revised" License
1.26k stars 193 forks source link

Legacy autograd Runtime error #29

Open vaib-saxena opened 4 years ago

vaib-saxena commented 4 years ago

Hi all,

I am currently doing a small project on image captioning. I came across QRNN and thought of replacing LSTM with QRNN. Everything was working fine with LSTM with longer training times but as soon as I replace LSTM with QRNN, I am getting this error.

Legacy autograd function with non-static forward method is deprecated. Please use new-style autograd function with static forward method. (Example: https://pytorch.org/docs/stable/autograd.html#torch.autograd.Function)

Even on running the sample code provided on this repo I am getting the same above error. import torch from torchqrnn import QRNN

seq_len, batch_size, hidden_size = 7, 20, 256 size = (seq_len, batch_size, hidden_size) X = torch.autograd.Variable(torch.rand(size), requires_grad=True).cuda()

qrnn = QRNN(hidden_size, hidden_size, num_layers=2, dropout=0.4) qrnn.cuda() output, hidden = qrnn(X)

print(output.size(), hidden.size())

RUntime error:Legacy autograd function with non-static forward method is deprecated. Please use new-style autograd function with static forward method. (Example: https://pytorch.org/docs/stable/autograd.html#torch.autograd.Function)

Please tell me how to get rid of this error. Thanks

korakot commented 4 years ago

I face the same problem.

Here's the old vs the new method.

https://discuss.pytorch.org/t/custom-autograd-function-must-it-be-static/14980

gegallego commented 4 years ago

Hello, I had the same error. It seems that PyTorch removed the backward compatibility, for the old non-static methods of torch.autograd.Function, in version 1.5.0. I downgraded PyTorch to 1.4.0, and it's working.