how to use cuda acceleration about cooper

[juan43ramirez]

Yes, Cooper can benefit from Cuda acceleration.

As long as your tensors and model are on Cuda, calculations will be carried out on GPUs.
Cooper lets Pytorch handle the details.

Are you having specific issues using Cooper on Cuda? A trace or code snippet would be helpful.

[Timbrer]

Thanks for your reply.

I do not have errors, but I also do not find GPU is used when I run the code.

Maybe beacause my problem is too simple.

[gallego-posada]

You can place your tensors on the GPU before or during the call to closure. For example, you can do something like this:

model = model.cuda()

for batch_id, (input_, target_) in enumerate(train_loader):
    
    # Move desired tensors to the righ device.
    input_, target_ = input_.cuda(), target_.cuda()

    constrained_optimizer.zero_grad()
    lagrangian = formulation.composite_objective(cmp.closure, input_, target_, model ...)
    formulation.custom_backward(lagrangian)
    constrained_optimizer.step()

Cooper will automatically place the multipliers on the same device in which the constraint defects of the CMPState are located.