RuntimeError: No CUDA implementation of 'syev'

autograd
1

I’m using symeig function inside a customized augograd function. The network works fine before I use net.cuda(), however, once it is enabled, I got an error " RuntimeError: No CUDA implementation of ‘syev’. Install MAGMA and rebuild cutorch (http://icl.cs.utk.edu/magma/) at d:\pytorch\pytorch\torch\lib\thc\generic/THCTensorMathMagma.cu:200" Do I have to install MAGMA to fix the issue?
The forward function in the custom autograd function is as following:
def forward(self, input):
w, v = torch.symeig(input, eigenvectors=self.eigenvectors, upper=self.upper)
self.save_for_backward(input, w, v)
return w, v

2

Yes. I’m not sure if it works on windows, though (it looks like you’re on windows)