Skip to content

[ROCm][CI] Reduce Flakiness For test_async_scheduling Using ROCM_ATTN With FP32 #30811

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tjtanaa merged 3 commits into from
Dec 18, 2025
Merged

[ROCm][CI] Reduce Flakiness For test_async_scheduling Using ROCM_ATTN With FP32 #30811

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
09edaab
update attn backend for async scheduling without spec decode test
micah-wil Dec 16, 2025
68b19df
update supported dtypes for ROCM_ATTN
micah-wil Dec 17, 2025
3d99245
Merge remote-tracking branch 'origin/main' into micah/async-schedulin…
micah-wil Dec 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 2 additions & 10 deletions tests/v1/e2e/test_async_scheduling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -148,7 +148,7 @@ def run_tests(
# Use TRITON_ATTN for spec decoding test for consistency
attention_config = {"backend": "TRITON_ATTN"}
else:
attention_config = {"backend": "ROCM_AITER_FA"}
attention_config = {"backend": "ROCM_ATTN"}
else:
attention_config = {"backend": "FLEX_ATTENTION"}

Expand Down Expand Up @@ -284,14 +284,6 @@ def run_test(
print(f"---- TESTING {test_str}: {test_config}")
print("-" * 80)

# On ROCm: use float16 for first test (ROCM_AITER_FA), but float32 for
# spec decoding test (TRITON_ATTN) for better precision.
# On others: always use float32.
if current_platform.is_rocm() and not is_testing_with_spec_decoding:
dtype = "float16"
else:
dtype = "float32"

with VllmRunner(
model,
max_model_len=512,
Expand All @@ -301,7 +293,7 @@ def run_test(
# enforce_eager=True,
async_scheduling=async_scheduling,
distributed_executor_backend=executor,
dtype=dtype,
dtype="float32",
speculative_config=spec_config,
disable_log_stats=False,
attention_config=attention_config,
Expand Down
6 changes: 5 additions & 1 deletion vllm/v1/attention/backends/rocm_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -152,7 +152,11 @@ def build(

class RocmAttentionBackend(AttentionBackend):
accept_output_buffer: bool = True
supported_dtypes: ClassVar[list[torch.dtype]] = [torch.float16, torch.bfloat16]
supported_dtypes: ClassVar[list[torch.dtype]] = [
torch.float16,
torch.bfloat16,
torch.float32,
]

@classmethod
def get_supported_head_sizes(cls) -> list[int]:
Expand Down
Loading