When running gs_small or gst_small examples the numpy version will always return a non-positive definite hessian. This comes from the fact that the last row of f_ii_loc is incorrect. I will attach two output files generated from gs_small, the gpu and cpu version. When doing a diff between the files the problem becomes clearly visible. i dont know if its computed incorrectly or overwritten incorrectly. the issue is for sure in the function _evaluate_hessian_f()
output_gpu.txt
output_cpu.txt
When running gs_small or gst_small examples the numpy version will always return a non-positive definite hessian. This comes from the fact that the last row of f_ii_loc is incorrect. I will attach two output files generated from gs_small, the gpu and cpu version. When doing a diff between the files the problem becomes clearly visible. i dont know if its computed incorrectly or overwritten incorrectly. the issue is for sure in the function _evaluate_hessian_f()
output_gpu.txt
output_cpu.txt