Open harrkee opened 6 days ago
Is it using OpenMP parallelism (via the math libraries)? If so, then the execution is nondeterministic and such things can happen, especially if the code hasn't been optimised for numerical stability.
Can you try with OMP_NUM_THREADS set to 0?
With a pytest
abs
tol of 1e-6,test_accas_H2
seems to fail occasionally (see this workflow where it fails, compared to a rerun of the workflow, where it succeeds).I've softened the tolerance to 2e-6 for now, but I'm not sure where this stochasticity is coming from. Could it really just be propagation of floating point err?