diff --git a/tests/v1/spec_decode/test_eagle.py b/tests/v1/spec_decode/test_eagle.py index 5d587fa3ec..188e84abca 100644 --- a/tests/v1/spec_decode/test_eagle.py +++ b/tests/v1/spec_decode/test_eagle.py @@ -755,12 +755,6 @@ def test_load_model( use_distinct_lm_head, monkeypatch, ): - if attn_backend == "TRITON_ATTN" and not current_platform.is_rocm(): - pytest.skip( - "TRITON_ATTN does not support " - "multi-token eagle spec decode on current platform" - ) - if attn_backend == "ROCM_AITER_FA" and current_platform.is_rocm(): monkeypatch.setenv("VLLM_ROCM_USE_AITER", "1") diff --git a/tests/v1/spec_decode/test_eagle_step_kernel.py b/tests/v1/spec_decode/test_eagle_step_kernel.py index 275a157d1b..83a29bc046 100644 --- a/tests/v1/spec_decode/test_eagle_step_kernel.py +++ b/tests/v1/spec_decode/test_eagle_step_kernel.py @@ -15,8 +15,8 @@ DEVICE_TYPE = current_platform.device_type # Skip if no CUDA - Triton kernel requires GPU pytest.importorskip("triton") -if not torch.cuda.is_available(): - pytest.skip("CUDA required for EAGLE kernel tests", allow_module_level=True) +if not current_platform.is_cuda_alike() and not current_platform.is_xpu(): + pytest.skip("CUDA/XPU required for EAGLE kernel tests", allow_module_level=True) def _reference_eagle_step_slot_mapping( diff --git a/tests/v1/spec_decode/test_max_len.py b/tests/v1/spec_decode/test_max_len.py index 42991f9f1a..1e1c674519 100644 --- a/tests/v1/spec_decode/test_max_len.py +++ b/tests/v1/spec_decode/test_max_len.py @@ -38,12 +38,6 @@ def test_ngram_max_len(num_speculative_tokens: int): def test_eagle_max_len( monkeypatch: pytest.MonkeyPatch, num_speculative_tokens: int, attn_backend: str ): - if attn_backend == "TRITON_ATTN" and not current_platform.is_rocm(): - pytest.skip( - "TRITON_ATTN does not support " - "multi-token eagle spec decode on current platform" - ) - if attn_backend == "ROCM_AITER_FA" and current_platform.is_rocm(): monkeypatch.setenv("VLLM_ROCM_USE_AITER", "1")