【解决】RuntimeError:Trying to backward throughthe graph a second time

pytorch 进行训练时报错：
RuntimeError:Trying to backward throughthe graph a second time (or directly access saved tensors after they have already been freed). Saved intermediate values of the graph are freed when you call.backward() or autograd.grad(). Specify retain graph=True if ou need to backward throug h the graph a secon time or if you need to access saved tensors after calling backward.

出现这个问题主要是梯度发生了变化，可以检查一下在loss.backwalk()执行前是否对loss进行了其它操作。
经过检查，我对loss进行了累加，如下：

loss += loss_func(outputs, labels).to(DEVICE) # loss_func是损失函数optimizer.zero_grad()  # gradient to zero
loss.backward()  # backward
optimizer.step()  # renew weight

因为我想计算训练误差，通过把各个批量的误差累加起来求平均，来表示本轮epoch的训练误差。
为了解决以上问题，不要对loss进行任何操作，只需添加一个新变量来保存即可，如下：

loss = loss_func(outputs, labels).to(DEVICE) # loss_func是损失函数
loss_totla += loss  # loss_total用来保存批量误差optimizer.zero_grad()  # gradient to zero
loss.backward()  # backward
optimizer.step()  # renew weight

以上操作，便可解决错误！

【解决】RuntimeError:Trying to backward throughthe graph a second time相关推荐

RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors a
问题描述: pytorch 更新网络参数报错: 报错行: loss1.backward(retain_graph=True) 错误信息: RuntimeError: Trying to backwar ...
RuntimeError: Trying to backward through the graph a second time
今天在测试conditional VAE的时候又遇到了这个问题,后来经过个人的分析发现了features embeddings时候比较坑的一个点,在此记录如下. 我通过: celltype = emb ...
如何解决“RuntimeError: CUDA Out of memory”问题
点击上方"AI公园",关注公众号,选择加"星标"或"置顶" 作者:Nitin Kishore 编译:ronghuaiyang 导读一些可以 ...
成功解决RuntimeError: Selected KDE bandwidth is 0. Cannot estiamte density
成功解决RuntimeError: Selected KDE bandwidth is 0. Cannot estiamte density 目录解决问题解决思路解决方法 1.重新定义bw参数 ...
成功解决RuntimeError: filter weights array has incorrect shape.
成功解决RuntimeError: filter weights array has incorrect shape. 目录解决问题解决思路解决方法解决问题 File "F:\Pro ...
成功解决RuntimeError: Decompression ‘SNAPPY‘ not available. Options: [‘GZIP‘, ‘UNCOMPRESSED‘]
成功解决RuntimeError: Decompression 'SNAPPY' not available. Options: ['GZIP', 'UNCOMPRESSED'] 目录解决问题解 ...
成功解决RuntimeError: Java is not installed, or the Java executable is not on system path
成功解决RuntimeError("Java is not installed, or the Java executable is not on system path") Ru ...
已解决RuntimeError: CUDA error: device-side assert triggered异常的正确解决方法，亲测有效！！！
已解决RuntimeError: CUDA error: device-side assert triggered异常的正确解决方法,亲测有效!!! 文章目录报错问题解决方法福利报错问题粉丝 ...
解决RuntimeError: cuda runtime error (30) : unknown error at /pytorch/aten/src/THC/THCGeneral.cpp:70N
解决两个问题: (1)RuntimeError: cuda runtime error (30) : unknown error at /pytorch/aten/src/THC/THCGeneral ...
解决RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED 踩过的坑（配置环境:linux+anaconda3+pytorch ...）
这个问题花了我好长好长时间,大概四五天~ 解决了之后好事很开心的,所以想记录一下,第一次写博客,有什么不对的地方希望大家可以指出来,大家一起学习~ 之前在linux系统(Ubantu)上稀里糊涂配置好 ...

【解决】RuntimeError:Trying to backward throughthe graph a second time

【解决】RuntimeError:Trying to backward throughthe graph a second time相关推荐

最新文章

热门文章