问题报错
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [544, 768]], which is output 0 of RuleBackward, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).
问题排查
使用torch.autograd.set_detect_anomaly(True)
在训练之前
torch.autograd.set_detect_anomaly(True)
在反向传播时
with autograd.detect_anomaly():
loss.backward()
问题发现
是x=F.relu(x)出错，具体原因未知。