[mcj]RuntimeError: Trying to backward through the graph a second time, but the buffers have….-马春杰杰

pytorch 进行反向传播有时会报错：

RuntimeError                              Traceback (most recent call last)
&lt;ipython-input-52-049eeb1d9b54&gt; in &lt;module&gt;
----&gt; 1 v_out.backward()
      2 print(variable.grad)

~/.conda/envs/pytorch/lib/python3.6/site-packages/torch/tensor.py in backward(self, gradient, retain_graph, create_graph)
    100                 products. Defaults to ``False``.
    101         """
--&gt; 102         torch.autograd.backward(self, gradient, retain_graph, create_graph)
    103 
    104     def register_hook(self, hook):

~/.conda/envs/pytorch/lib/python3.6/site-packages/torch/autograd/__init__.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables)
     88     Variable._execution_engine.run_backward(
     89         tensors, grad_tensors, retain_graph, create_graph,
---&gt; 90         allow_unreachable=True)  # allow_unreachable flag
     91 
     92 

RuntimeError: Trying to backward through the graph a second time, but the buffers have already been freed. Specify retain_graph=True when calling backward the first time.

RuntimeError Traceback (most recent call last)

<ipython-input-52-049eeb1d9b54> in <module>

----> 1 v_out.backward()

2 print(variable.grad)

~/.conda/envs/pytorch/lib/python3.6/site-packages/torch/tensor.py in backward(self, gradient, retain_graph, create_graph)

100 products. Defaults to ``False``.

101 """

--> 102 torch.autograd.backward(self, gradient, retain_graph, create_graph)

103

104 def register_hook(self, hook):

~/.conda/envs/pytorch/lib/python3.6/site-packages/torch/autograd/__init__.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables)

88 Variable._execution_engine.run_backward(

89 tensors, grad_tensors, retain_graph, create_graph,

---> 90 allow_unreachable=True) # allow_unreachable flag

RuntimeError: Trying to backward through the graph a second time, but the buffers have already been freed. Specify retain_graph=True when calling backward the first time.

这是因为第一次反向传播之后，这个计算图的内存就会被释放掉，这样的话再次进行反向传播就不行了，解决方法就是添加retain_graph=True这个参数。

添加方法：

温馨提示: 此处内容需要评论本文后刷新本页才能查看！

本文最后更新于2019年5月26日，已超过 1 年没有更新，如果文章内容或图片资源失效，请留言反馈，我们会及时处理，谢谢！

赞一个 (357)收藏 (0)