diff --git a/.buildkite/test-amd.yaml b/.buildkite/test-amd.yaml index 3af9195428..3e631fe5d2 100644 --- a/.buildkite/test-amd.yaml +++ b/.buildkite/test-amd.yaml @@ -1576,10 +1576,14 @@ steps: - vllm/third_party/flash_linear_attention/ops/kda.py - vllm/third_party/flash_linear_attention/ops/chunk_delta_h.py - vllm/third_party/flash_linear_attention/ops/l2norm.py - - tests/kernels/test_kda.py + - vllm/models/kimi_k3/nvidia/kda.py + - vllm/models/kimi_k3/nvidia/kda_metadata.py + - vllm/models/kimi_k3/nvidia/ops/third_party/kda/ + - tests/models/kimi_k3/test_kda.py + - tests/models/kimi_k3/test_kda_metadata.py - vllm/platforms/rocm.py commands: - - pytest -v -s kernels/test_kda.py + - pytest -v -s models/kimi_k3/test_kda.py models/kimi_k3/test_kda_metadata.py - label: Kernels Mamba Test # TBD timeout_in_minutes: 180 diff --git a/vllm/models/kimi_k3/nvidia/kda.py b/vllm/models/kimi_k3/nvidia/kda.py index d4fec1e008..3bc3ab5de0 100644 --- a/vllm/models/kimi_k3/nvidia/kda.py +++ b/vllm/models/kimi_k3/nvidia/kda.py @@ -164,6 +164,8 @@ def is_flashkda_supported( dtype: torch.dtype, lower_bound: float | None, ) -> bool: + if not current_platform.is_cuda(): + return False capability = current_platform.get_device_capability() return ( capability is not None