AttributeError: 'Leaky' object has no attribute 'reset'

[lelucermaireme]

• snntorch version: 0.8.0
• Python version: 3.7
• Operating System: Linux

Description

I used two GPUs to run my code and run into this error

But the code can be run with no error if I use one GPU.
The error comes from the two devices, The tensors are run on different devices.

[jeshraghian]

Share a minimal working example of your code.

…

On Tue, 16 July 2024, 6:11 am Tracy, ***@***.***> wrote: - snntorch version: 0.8.0 - Python version: 3.7 - Operating System: Linux Description I used two GPUs to run my code and run into this error image.png (view on web) <https://github.com/user-attachments/assets/f046aa0c-7dba-4d7d-be15-db8931f23244> But the code can be run with no error if I use one GPU. The error comes from the two devices, The tensors are run on different devices. — Reply to this email directly, view it on GitHub <#339>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJTFT4SPTAHUUISJJT6LZVTZMUEYRAVCNFSM6AAAAABK6ORZVSVHI2DSMVQWIX3LMV43ASLTON2WKOZSGQYTAOJYGIYDONA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[lelucermaireme]

Take the tutorial 5(https://snntorch.readthedocs.io/en/latest/tutorials/tutorial_5.html) as an example, I just add

os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
net = Net()
net = nn.DataParallel(net).cuda()

to make it work on multiple GPUs.

[lelucermaireme]

I think the mistake comes from the variable self.reset that defined in class Leaky. The variable self.reset is defined when an object of class Leaky is defined. So the variable self.reset is on a fixed GPU. But the inputs to the network are on different GPUs.