Skip to content

Fix attn bug in qwen3-8b benchmark test #721

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

Open
wants to merge 1 commit into
base: 355_wip
Choose a base branch
from
Open

Fix attn bug in qwen3-8b benchmark test #721

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
1 change: 0 additions & 1 deletion vllm/attention/layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,11 +30,11 @@
USE_XFORMERS_OPS = None

if current_platform.is_rocm() and envs.VLLM_ROCM_USE_AITER:
VLLM_ROCM_USE_AITER_TRITON_FUSED_ROPE_ZEROS_KV_CACHE = envs.VLLM_ROCM_USE_AITER_TRITON_FUSED_ROPE_ZEROS_KV_CACHE

Check failure on line 33 in vllm/attention/layer.py

View workflow job for this annotation

GitHub Actions / pre-commit

Ruff (E501) 

vllm/attention/layer.py:33:81: E501 Line too long (116 > 80)
else:
VLLM_ROCM_USE_AITER_TRITON_FUSED_ROPE_ZEROS_KV_CACHE = False

logger.info(f"[Aiter] {VLLM_ROCM_USE_AITER_TRITON_FUSED_ROPE_ZEROS_KV_CACHE=}")

Check failure on line 37 in vllm/attention/layer.py

View workflow job for this annotation

GitHub Actions / pre-commit

Ruff (G004) 

vllm/attention/layer.py:37:13: G004 Logging statement uses f-string

def check_xformers_availability():
global USE_XFORMERS_OPS
Expand Down Expand Up @@ -536,13 +536,12 @@
key,
value,
kv_cache,
attn_metadata,
output=output,

Check failure on line 540 in vllm/attention/layer.py

View workflow job for this annotation

GitHub Actions / pre-commit

Ruff (SIM101) 

vllm/attention/layer.py:539:13: SIM101 Multiple `isinstance` calls for expression, merge into a single call
output_scale=output_scale,
positions=positions)
else:
assert positions is None, f"positions must be None {positions=}"
self.impl.forward(self,

Check failure on line 544 in vllm/attention/layer.py

View workflow job for this annotation

GitHub Actions / pre-commit

Ruff (E501) 

vllm/attention/layer.py:544:81: E501 Line too long (142 > 80)
query,
key,
value,
Expand Down
Loading