How to use different weights in forward and backward steps from update step in model.parameters

Hi everyone!
I want to use the same weights in the forward and backward steps in the training, but for the update step, i want to use different weights as old_weights in:

new_weights = old_weights - learning_rate * gradients

in fact, old_weights are different from the weights that are used in the forward and backward pass. How can i do this for model.parameters() in pytorch?

Thanks in advance