Skip to content

[Misc] Add num_splits input arg to flash_attn_varlen_func #72

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
merged 1 commit into from
Jul 1, 2025
Merged

[Misc] Add num_splits input arg to flash_attn_varlen_func #72

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
5 changes: 4 additions & 1 deletion vllm_flash_attn/flash_attn_interface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,6 +142,7 @@ def flash_attn_varlen_func(
q_descale=None,
k_descale=None,
v_descale=None,
num_splits: int = 0,
# Version selector
fa_version: int = DEFAULT_FA_VERSION,
):
Expand Down Expand Up @@ -224,6 +225,8 @@ def flash_attn_varlen_func(
"FA2 does not support scheduler_metadata, q_descale, "
"k_descale, v_descale"
)
if num_splits > 1:
raise NotImplementedError("FA2 does not support num_splits > 1")
out, softmax_lse = torch.ops._vllm_fa2_C.varlen_fwd(
q, k, v,
out,
Expand Down Expand Up @@ -270,7 +273,7 @@ def flash_attn_varlen_func(
softcap,
True, # rotary_interleaved
scheduler_metadata,
0, # num_splits
num_splits,
None, # pack_gqa
0, # sm_margin
)
Expand Down