there is no gradient

[xuridongshengpxd]

Hi author,
good job.when i ran the train.py,i meet this error.

File "D:\mycodeapp\anaconda\envs\py7\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "D:\mycode\deep-stabilization-master\dvs\model.py", line 27, in forward
self.hx, self.cx = self.LSTM(x, (self.hx, self.cx))
File "D:\mycodeapp\anaconda\envs\py7\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "D:\mycodeapp\anaconda\envs\py7\lib\site-packages\torch\nn\modules\rnn.py", line 974, in forward
self.bias_ih, self.bias_hh,
(function print_stack)
Traceback (most recent call last):
File "D:/mycode/deep-stabilization-master/dvs/train.py", line 276, in
train(args = args)
File "D:/mycode/deep-stabilization-master/dvs/train.py", line 254, in train
train_loss = run_epoch(model, train_loader, cf, count, lr, optimizer=optimizer, clip_norm=clip_norm, is_training=True, USE_CUDA=USE_CUDA)
File "D:/mycode/deep-stabilization-master/dvs/train.py", line 147, in run_epoch
loss.backward(retain_graph=True)
File "D:\mycodeapp\anaconda\envs\py7\lib\site-packages\torch\tensor.py", line 221, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "D:\mycodeapp\anaconda\envs\py7\lib\site-packages\torch\autograd_init.py", line 132, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [2048]] is at version 2; expected version 1 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

[WarlockWendell]

Hi! did you find a solution to your problem？ I also had the same problem.

[zhmeishi]

It seems the problem of the Pytorch version. Please try Pytorch 1.0.0 or replace LayerLSTM class with nn.LSTM.

[WarlockWendell]

It seems the problem of the Pytorch version. Please try Pytorch 1.0.0 or replace LayerLSTM class with nn.LSTM.

Problem solved! Thanks for your reply!

[etoilestar]

hello， could you tell me how to replace LayerLSTM with nn.LSTM

[HouBiaoLiu]

modify like this: @etoilestar
class LayerLSTM(nn.Module):
def init(self, input_size, hidden_size, bias):
super(LayerLSTM, self).init()
self.LSTM = nn.LSTMCell(input_size, hidden_size, bias)
self.hidden_size = hidden_size

def init_hidden(self, batch_size):
    self.hx = torch.zeros((batch_size, self.hidden_size)).cuda()
    self.cx = torch.zeros((batch_size, self.hidden_size)).cuda()

def forward(self, x):
    x,_ = self.LSTM(x, (self.hx, self.cx))
    return x