torch.linalg.ldl_factor bugs out when applied to a symmetric, but non-PD matrix. I am not sure what the exact algorithm is, but I was expecting the “modified Cholesky” used in Newton optimization steps to correct the Hessian to be P.D. What does this error mean?
Could you post a minimal and executable code snippet to reproduce the issue as well as the output of python -m torch.utils.collect_env
, please?