Flaky Test Failures

[theo-barfoot]

It seems typical when running the CI for one of the version combinations to fail due to the flakiness of some of the tests for which random tensors are used.

I don't have a complete list but, some examples of failure cases are:

1. test_sparse_batch_mv[batch_mv_test_data3]
2. test_bicgstab.py::test_bicgstab_2d_rhs
3. Maybe linear_cg?

It is worth revisiting:

• The use of @pytest.mark.flaky(reruns=5)
• The seeding of RNGs
• The tolerance and testing criteria used for these tests