From 392db567b260cb4d1e5f91547b65d527038fe641 Mon Sep 17 00:00:00 2001 From: Chaojun Zhang Date: Mon, 7 Sep 2026 16:57:15 +0800 Subject: [PATCH] [CI] [Test] skip test_wna16_cuda_high_bit_skips_humming on non-CUDA platforms (#55660) Signed-off-by: Chaojun Zhang --- tests/quantization/test_auto_round.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/tests/quantization/test_auto_round.py b/tests/quantization/test_auto_round.py index 5250f73b82..bfc532126a 100644 --- a/tests/quantization/test_auto_round.py +++ b/tests/quantization/test_auto_round.py @@ -726,6 +726,10 @@ def test_wna16_cuda_low_bit_moe_routes_to_humming(monkeypatch, bits) -> None: assert captured["layer_config"] is layer_config +@pytest.mark.skipif( + not current_platform.is_cuda(), + reason="This test only exercises the CUDA Marlin path.", +) @pytest.mark.parametrize("bits", [4, 8]) def test_wna16_cuda_high_bit_skips_humming(monkeypatch, bits) -> None: """4/8-bit int stays on the Marlin/GPTQ/AWQ path even on CUDA so a single