[ROCm][Spec Decode] Fix probabilistic draft probs test attention backend (#45706)

Signed-off-by: Stefan Koncarevic <stefan.koncarevic@amd.com>
This commit is contained in:
stefankoncarevic
2026-06-18 11:59:37 -05:00
committed by GitHub
parent 25faa1f4cc
commit e2352c2974
2 changed files with 12 additions and 2 deletions
+6 -2
View File
@@ -1002,7 +1002,11 @@ def test_propose(method, attn_backend, num_speculative_tokens, monkeypatch):
assert torch.equal(result, expected_tokens)
def test_propose_stores_probabilistic_draft_probs(monkeypatch):
@pytest.mark.parametrize(
"attn_backend",
["ROCM_ATTN", "TRITON_ATTN"] if current_platform.is_rocm() else ["FLASH_ATTN"],
)
def test_propose_stores_probabilistic_draft_probs(attn_backend, monkeypatch):
device = torch.device(DEVICE_TYPE)
batch_size = 2
seq_lens = [5, 3]
@@ -1053,7 +1057,7 @@ def test_propose_stores_probabilistic_draft_probs(monkeypatch):
)
attn_metadata_builder_cls, _ = try_get_attention_backend(
AttentionBackendEnum.FLASH_ATTN
AttentionBackendEnum[attn_backend]
)
attn_metadata_builder = attn_metadata_builder_cls(
kv_cache_spec=create_standard_kv_cache_spec(proposer.vllm_config),