Intel MKL Error when using torch.symeig

Hi,
I have a symmetric matrix X which is constructed using the outputs of network for each batch. I then find the eigenvalues and eigenvectors of X using

e, V = torch.symeig(X, True)

While this works initially, I get the error below after some time. I think this may be due to the outputs of the network. Pls, is there any approach I can use to avoid this error?

Intel MKL ERROR: Parameter 4 was incorrect on entry to SLASCL.

Intel MKL ERROR: Parameter 4 was incorrect on entry to SLASCL.
info: 99
, submat: 0
info: 901

symeig_cuda: the algorithm failed to converge; 901 off-diagonal elements of an intermediate tridiagonal form did not converge to zero.