RuntimeError: cudnn RNN backward can only be called in training mode

[cook-kuk]

I found an error like that.

I think this is because the model mode is set to eval. So I change all models to train, but it doesn't work well.

self.predictor.train() self.encoder.train() self.generator.train()

rl_loss.backward()

Which model should I turn into a train to work?

C:\Users\seungho.kuk\anaconda3\envs\paccmann_rl\lib\site-packages\torch\nn\modules\container.py:100: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument. input = module(input) Traceback (most recent call last): File "C:/Users/seungho.kuk/Desktop/Python_project/paccmann_rl/code/paccmann_generator/examples/train_paccmann_rl.py", line 314, in main() File "C:/Users/seungho.kuk/Desktop/Python_project/paccmann_rl/code/paccmann_generator/examples/train_paccmann_rl.py", line 220, in main cell_line, epoch, params['batch_size'] File "C:\Users\seungho.kuk\anaconda3\envs\paccmann_rl\lib\site-packages\paccmann_generator\reinforce.py", line 442, in policy_gradient rl_loss.backward() File "C:\Users\seungho.kuk\anaconda3\envs\paccmann_rl\lib\site-packages\torch\tensor.py", line 195, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph) File "C:\Users\seungho.kuk\anaconda3\envs\paccmann_rl\lib\site-packages\torch\autograd__init__.py", line 99, in backward allow_unreachable=True) # allow_unreachable flag RuntimeError: cudnn RNN backward can only be called in training mode

[jannisborn]

The initial code for the RL optimization (i.e. the code you are running) was written for CPU, not GPU execution. I suggest you to try running the code on CPU.

[cook-kuk]

Does training with CPU take a long time?

[jannisborn]

If all components are pre-trained, RL optimization usually takes not more than 30min on a standard Laptop.

[cook-kuk]

Thanks

It works fine after changing the pytorch version to cpu.