Got an error when training social NCE model

[vietanhle-hri]

Hello,

I tried to train a social NCE with GPU option using python imitate.py --contrast_weight=2.0 --contrast_sampling='event' --gpu, and I got an error below

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mask in method wrapper__masked_select)

Does anyone help me to fix this issue? Thanks in advance.

[YuejiangLIU]

Hi @vietanhle-hri, sorry for the slow response.

I did not encounter this error on my end, but it's possible that there are some hidden device settings that I overlooked.

Do you know which line/variable that results in the device error?

Maybe you could solve the error by simply adding .to(device) as follows.

social-nce/crowd_nav/imitate.py

Line 81 in 6163060

policy_net = policy.get_model().to(device)