Skip to content

[rocm]use aiter triton kernel as triton mha fallback path #809

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

Draft
zhuyuhua-v wants to merge 2 commits into
base: dev/perf
Choose a base branch
from
Draft

[rocm]use aiter triton kernel as triton mha fallback path #809

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
31ea5c1
use aiter triton kernel as triton mha fallback path
zhuyuhua-v Nov 14, 2025
58a25ff
update ROCM condition in flash attention
zhuyuhua-v Nov 20, 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
15 changes: 15 additions & 0 deletions vllm/v1/attention/backends/mla/triton_mla.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,6 +128,21 @@ def _flash_attn_varlen_diff_headdims(
return self._flash_attn_varlen_diff_headdims_rocm(
q, k, v, softmax_scale=softmax_scale, **kwargs
)
elif current_platform.is_rocm():
from aiter.ops.triton.mha import flash_attn_varlen_func
result = flash_attn_varlen_func(
q=q,
k=k,
v=v,
return_lse=return_softmax_lse,
softmax_scale=softmax_scale,
**kwargs,
)
if type(result) is tuple and return_softmax_lse:
output, lse = result
lse = lse.T.contiguous()
return (output, lse)
return result
else:
return super()._flash_attn_varlen_diff_headdims(
q,
Expand Down
Loading