RuntimeError: expected device cuda:0 but got device cpu

[liuqi8827]

Thanks for your work.
I'm facing two problems when I run your project.

1.I installed the project successfully according to your README.md.

2.In order to use GPU, I changed the code in /SCR/crowd_nav/train.py def main():
parser.add_argument('--gpu', default=False, action='store_true')
to
parser.add_argument('--gpu', default=True, action='store_true')
However, I got some errors in the terminator
RuntimeError: Expected object of backend CUDA but got backend CPU ...
3. Then I changed the code in /SCR/crowd_nav/empowerment/source.py class Source(nn.Module):

    self.fc = nn.Linear(nb_states, hidden1)
    self.mu_head = nn.Linear(hidden1, nb_actions)
    self.sigma_head = nn.Linear(hidden1, nb_actions)

to

     self.fc = nn.Linear(nb_states, hidden1).cuda()
     self.mu_head = nn.Linear(hidden1, nb_actions).cuda()
     self.sigma_head = nn.Linear(hidden1, nb_actions).cuda()

I did the same change in /SCR/crowd_nav/empowerment/transition.py and /SCR/crowd_nav/empowerment/planning.py
4. Then I run python train.py --policy scr
However, I got the error in the terminator:
RuntimeError: expected device cuda:0 but got device cpu

5. What's the version of your torch?
   My environment is
   Ubuntu = 16.04
   python = 3.6.10
   torch version = 0.6.0
   pyTorch = 1.5.0
   GPU GTX 2070s
   NVIDIA-SMI 430.64
   CUDA Version: 10.1

6. I got an error when I use cpu:
   RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
   So I changed all the a += b to a = a + b in the code of your project.
   For example
   epoch_loss += loss to epoch_loss = epoch_loss + loss
   But it dosen't work, I got the same error.
   Maybe the version of torch is the reason of this error.

Thanks a lot!

[liuqi8827]

Hi,

I git clone your project and didn't modify any code.
However, when I run python train.py --policy scr
I got the error in the terminator:

Can you give me some suggestions to solve this problem?
Thanks a lot!

[tessavdheiden]

Hi!

The second issue (backward.step()) is resolved now. Let me know if it works!

[bektaskemal]

Hi,
I also have problem using Gpu. Any update on that?

[liuqi8827]

Hi!

The second issue (backward.step()) is resolved now. Let me know if it works!

@tessavdheiden Yes, it worked successfully.
Thanks a lot!