[ROCm][CI] Restore Torch defaults and type DSV4 scratch buffers (#52566)

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

Andreas Karatzas committed Aug 17, 2026 at 13:32 UTC c1e438728c55281fda46c8baed755f4148c63660
2 files changed +22 -4
tests/kernels/attention/test_mha_attn.py
+10 -2
@@ -27,9 +27,17 @@ from vllm.v1.attention.selector import _cached_get_attn_backend
27
28
29 @pytest.fixture(autouse=True)
30 -def clear_cache():
31 - """Clear lru cache to ensure each test case runs without caching."""
30 +def reset_test_state():
31 + """Clear cached selectors and restore process-wide torch defaults."""
32 + default_device = torch.get_default_device()
33 + default_dtype = torch.get_default_dtype()
34 _cached_get_attn_backend.cache_clear()
35 + try:
36 + yield
37 + finally:
38 + torch.set_default_device(default_device)
39 + torch.set_default_dtype(default_dtype)
40 + _cached_get_attn_backend.cache_clear()
41
42
43 devices = ["cpu"]
tests/kernels/attention/test_rocm_triton_attn_dsv4.py
+12 -2
@@ -730,9 +730,19 @@ def test_sparse_attn_decode_split_k_kernel(
730 @torch.inference_mode()
731 def test_sparse_attn_decode_gfx950_adaptive_reduce_ignores_stale_scratch() -> None:
732 device = torch.device("cuda")
733 - part_m = torch.full((1, 8, 1), torch.finfo(torch.float32).min, device=device)
733 + part_m = torch.full(
734 + (1, 8, 1),
735 + torch.finfo(torch.float32).min,
736 + dtype=torch.float32,
737 + device=device,
738 + )
739 part_l = torch.zeros_like(part_m)
735 - part_acc = torch.full((1, 8, 1, HEAD_DIM), float("nan"), device=device)
740 + part_acc = torch.full(
741 + (1, 8, 1, HEAD_DIM),
742 + float("nan"),
743 + dtype=torch.float32,
744 + device=device,
745 + )
746 part_m[:, :2] = 0
747 part_l[:, :2] = 1
748 part_acc[:, 0] = 1