diff --git a/tests/kernels/attention/test_use_trtllm_attention.py b/tests/kernels/attention/test_use_trtllm_attention.py index fba18fe46e..89ff86b47b 100644 --- a/tests/kernels/attention/test_use_trtllm_attention.py +++ b/tests/kernels/attention/test_use_trtllm_attention.py @@ -6,12 +6,19 @@ from unittest.mock import patch import pytest import torch +from vllm.platforms import current_platform from vllm.utils.flashinfer import ( can_use_trtllm_attention, supports_trtllm_attention, use_trtllm_attention, ) +if not current_platform.is_cuda(): + pytest.skip( + "TRTLLM attention is only supported on CUDA platforms.", + allow_module_level=True, + ) + MODEL_CONFIGS = { "Llama-3-70B": dict(num_qo_heads=64, num_kv_heads=8), "Llama-3-8B": dict(num_qo_heads=32, num_kv_heads=8),