[CI] [Test] skip test_wna16_cuda_high_bit_skips_humming on non-CUDA platforms (#55660)
Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com>
Chaojun Zhang committed
Sep 7, 2026 at 16:57 UTC
392db567b260cb4d1e5f91547b65d527038fe641
1 file changed
+4
tests/quantization/test_auto_round.py
+4
@@ -726,6 +726,10 @@ def test_wna16_cuda_low_bit_moe_routes_to_humming(monkeypatch, bits) -> None:
726
assert captured["layer_config"] is layer_config
727
728
729
+@pytest.mark.skipif(
730
+ not current_platform.is_cuda(),
731
+ reason="This test only exercises the CUDA Marlin path.",
732
+)
733
@pytest.mark.parametrize("bits", [4, 8])
734
def test_wna16_cuda_high_bit_skips_humming(monkeypatch, bits) -> None:
735
"""4/8-bit int stays on the Marlin/GPTQ/AWQ path even on CUDA so a single