torch.autograd.function.PReLUBackward's attribution about pytorch2caffe HOT 2 CLOSED

iknow that:
pytorch/torch/nn/_functions/thnn/activation.py:

@staticmethod
def backward(ctx, grad_output):
    **input, weight = ctx.saved_variables**
    # alternatively, we could recalculate _backend
    return PReLUBackward.apply(input, weight, grad_output, ctx._backend)

using "output.grad_fn.saved_variables[1]" ok!

from pytorch2caffe.

i read a website which tells me that i can seatch it from baskward function(def backward)

from pytorch2caffe.