A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 2, 01:00 UTC
[XPU] skip unapplied UT in test_gpu_model_runner.py (#44289)
Signed-off-by: Yan Ma <yan.ma@intel.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Yan Ma committed
Jun 4, 2026 at 08:48 UTC
0414d7541033596894e72ca75cf29df593ad60fc
1 file changed
+4
-4
tests/v1/worker/test_gpu_model_runner.py
+4
-4
@@ -1072,8 +1072,8 @@ def test_init_kv_cache_with_kv_sharing_valid(default_vllm_config):
1072
1073
1074
@pytest.mark.skipif(
1075
- current_platform.is_rocm(),
1076
- reason="Attention backend FLASHINFER is not supported on ROCm.",
1075
+ not current_platform.is_cuda(),
1076
+ reason="Attention backend FLASHINFER is only supported on CUDA.",
1077
)
1078
def test_hybrid_attention_mamba_tensor_shapes():
1079
"""
@@ -1508,8 +1508,8 @@ def test_is_uniform_decode() -> None:
1508
1509
1510
@pytest.mark.skipif(
1511
- current_platform.is_rocm(),
1512
- reason="Attention backend FLASHINFER is not supported on ROCm.",
1511
+ not current_platform.is_cuda(),
1512
+ reason="Attention backend FLASHINFER is only supported on CUDA.",
1513
)
1514
def test_mamba_cache_raises_when_max_num_seqs_exceeds_blocks():
1515
"""Test that a ValueError is raised when max_num_seqs exceeds the