Jingwei Zhang

So my solution is that: instead of just use the data from usage and drop the variable, I just implement a ```fake_cumprod``` using the existing torch ops, this way the...

@ypxie Hey, thanks for the quick reply:) What I'm confused is Line ```98``` in ```utils.py```: ``` output = Variable(inputs.data.new(*shape_).fill_(1.0), requires_grad = True) ``` Here the ```inputs``` is detached from the...