From f2f4cea6ccaad20becb6f02e253ae673f8a249ae Mon Sep 17 00:00:00 2001 From: rasmith Date: Thu, 4 Dec 2025 03:30:22 -0600 Subject: [PATCH] [CI/Build][AMD] Skip test on test_hybrid_attention_mamba_tensor_shapes on ROCm, requires FLASHINFER (#29995) Signed-off-by: Randall Smith Co-authored-by: Randall Smith --- tests/v1/worker/test_gpu_model_runner.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/tests/v1/worker/test_gpu_model_runner.py b/tests/v1/worker/test_gpu_model_runner.py index 0439bef1226e3..459abcfdd53cf 100644 --- a/tests/v1/worker/test_gpu_model_runner.py +++ b/tests/v1/worker/test_gpu_model_runner.py @@ -761,6 +761,10 @@ def test_init_kv_cache_with_kv_sharing_valid(): assert kv_cache_config_after_init.kv_cache_groups[0].layer_names[1] == layer_1 +@pytest.mark.skipif( + current_platform.is_rocm(), + reason="Attention backend FLASHINFER is not supported on ROCm.", +) def test_hybrid_attention_mamba_tensor_shapes(monkeypatch): """ The GPU model runner creates different views into the