one of the variables needed for gradient :.....with torch.autograd.set_detect_anomaly(True).
一个很神奇的错误:RuntimeError:oneofthevariablesneededforgradientcomputationhasbeenmodifiedbyaninplaceoperation:[torch.cuda.FloatTensor[16,8,1024]],whichisoutput0ofLeakyReluBackward1,isatversion2;expectedversi