>>> while True:
... a = Variable(torch.FloatTensor(32,16).cuda())
... s1,s2 = int(a.size()[0]),int(a.size()[1])
... r = Variable(torch.FloatTensor(1).cuda())
... c = r.expand((s1,s2))
I tried and I have the same bug when I direclty use torch.size().
I have no idea what causes this thing…