consistency

/

pipeline

like 1

Follow

consistency 1

Hello guys,
I am using a pre-trained consistency model for some tasks. I want to optimize the output of the model(the generated image ) with respect to some loss.
The generated image(output of consistency model) has no requires_grad attribute and is just a regular tensor. I am an able to modify the code so that the generated image is with requires_grad attribute. In this case I can backpropagate the loss to compute the gradient with respect to the model input.