Routetest_output_shape_matches_inputpytest.mark.parametrize("num_patches,world_size", [
(16, 4),
(16, 2),
(9, 3),
])
deepspeed/sequence/test_autosp.py:None
Routetest_zero2_tp_gradient_reducer_uses_shared_param_family_strategypytest.mark.parametrize(("mark_router", "expected_grad"), ((True, 1.0), (False, 1.0)))
tests/unit/v1/moe/test_autoep_autotp_grad_parity.py:None