Closed euhruska closed 5 years ago
I don't get this error, which tensorflow version are you using? This was an issue appearing sometimes on tf 1.7-1.8
tf version 1.10.0
For us this is working with this version of tensorflow. We are unable to reproduce this. Are you working on a GPU? (Perhaps also the cuda version plays a role...) Can you write us your Cuda version and which Cudnn you are using if it applies? Thx Andreas
It might also be that it failed to compute the SVD for the first batch so there could be a problem with your data. The documentation of heevd (which is a eigenvalue/eigenvector solver) says:
if INFO = i and JOBZ = 'N', then the algorithm failed
to converge; i off-diagonal elements of an intermediate
tridiagonal form did not converge to zero;
if INFO = i and JOBZ = 'V', then the algorithm failed
to compute an eigenvalue while working on the submatrix
lying in rows and columns INFO/(N+1) through
mod(INFO,N+1).
with
vamp._loss_VAMP_sym
in lineI get an fatal error: