Python 3.x 当PyTorch中的特征向量出现一些损失时，如何制作背景词？

我的损失似乎是

e1, v1 = torch.eig(A)
e2, v2 = torch.eig(B)
sim = torch.matmul(v1, v2.permute(0, 2, 1))
loss_sim = torch.sum(sim)
loss_sim.backward()

其中

A

和

B

是具有形状

序列长度*序列长度的中间张量

我的代码遇到错误：

文件“/usr/local/Anaconda3/lib/python3.8/site packages/torch/tensor.py”，第185行，在向后torch.autograd.backward（self、gradient、retain\u graph、create\u graph）文件“/usr/local/Anaconda3/lib/python3.8/site packages/torch/autograd/\u init\u\u.py”中，第125行，在向后变量中。\执行\引擎运行\向后（RuntimeError:eig_backward:backward计算目前不支持复数特征值。在/opt/conda/conda bld/pytorch_1595629395347/work/torch/csc/autograd/generated/Functions处从eig_backward引发的异常。cpp:1877（最新调用优先）：frame#0:c10:：Error:：Error（c10:：SourceLocation，std:：string）+0x4d）（0x7f74932e077d in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/libc10.so）frame#1:torch:：autograd:：generated:：EigBackward:：apply（std:：vector&&）+0xb5d（0x7f74cc4b894d in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/libtorch\cpu.so）frame#2:+0x30d1017（0x7F74CB0C017 in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/libtorch\u cpu.so）帧#3:torch:：autograd:：Engine:：evaluate_函数（std:：shared\u ptr&，torch:：autograd:：Node*，torch:：autograd:：InputBuffer&，std:：shared\u ptr const&+0x1400（0x7F74CB07860 in/usr/local/Anaconda3/lib/python3.8/site packages/torch\libtorch\cpu.so）帧#4:torch:：autograd:：Engine:：thread#main（std:：sharedŠptr const&）+0x451（0x7f74ccb08401in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/libtorchŠcpu.so）帧#5:torch:：autograd:：Engine:：threadŠinit（int，std:：sharedŠŠŠŠŠŠ（0x7F74CB00579 in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/libtorch_cpu.so）帧#6:torch:：autograd:：PythonEngine:：thread_init（int，std:：shared_ptr const&，bool）+0x4a（0x7f74d0e2a1ba in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/libtorch_python.so）帧#7:+0xc819d（0x7f74d393719d in/usr/local/Anaconda3/lib/python3.8/site packages/torch/lib/../../../../..//libstdc++.so.6）frame#8:+0x76ba（0x7f74ed28c6ba in/lib/x86-linux-gnu/libpthread.so.0）frame#9:clone+0x6d（0x7f74ecfc241d in/lib/x86-linux-gnu/libc.so.6）

如何使用torch.eig（）完成backword