Skip to content

feat: add flash attention3 and sageattention impl #35

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
DamonsJ wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat: add flash attention3 and sageattention impl #35

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
51 changes: 51 additions & 0 deletions flash_talk/wan/modules/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,18 @@
except ModuleNotFoundError:
FLASH_ATTN_2_AVAILABLE = False

try:
import xformers.flash_attn_3._C
XFORMERS_ATTN_3_AVAILABLE = True
except ModuleNotFoundError:
XFORMERS_ATTN_3_AVAILABLE = False

try:
import sageattention
SAGE_ATTN_AVAILABLE = True
except ModuleNotFoundError:
SAGE_ATTN_AVAILABLE = False

__all__ = [
'flash_attention',
'attention',
Expand Down Expand Up @@ -112,6 +124,45 @@ def half(x):
x = x.unflatten(0, (b, lq))
except:
x = x[0].unflatten(0, (b, lq))
elif (version is None or version == 3) and XFORMERS_ATTN_3_AVAILABLE:
x, _,_,_ = torch.ops.flash_attn_3.fwd(
q,
k,
v,
cu_seqlens_q=torch.cat([q_lens.new_zeros([1]), q_lens]).cumsum(
0, dtype=torch.int32).to(q.device, non_blocking=True),
cu_seqlens_k=torch.cat([k_lens.new_zeros([1]), k_lens]).cumsum(
0, dtype=torch.int32).to(q.device, non_blocking=True),
seqused_q=None,
seqused_k=None,
max_seqlen_q=lq,
max_seqlen_k=lk,
softmax_scale=softmax_scale,
is_causal=causal,
)
try:
x = x.unflatten(0, (b, lq))
except:
x = x[0].unflatten(0, (b, lq))
elif (version is None or version == 2) and SAGE_ATTN_AVAILABLE:
from sageattention import sageattn_varlen
x = sageattn_varlen(
q,
k,
v,
cu_seqlens_q=torch.cat([q_lens.new_zeros([1]), q_lens]).cumsum(
0, dtype=torch.int32).to(q.device, non_blocking=True),
cu_seqlens_k=torch.cat([k_lens.new_zeros([1]), k_lens]).cumsum(
0, dtype=torch.int32).to(q.device, non_blocking=True),
max_seqlen_q=lq,
max_seqlen_k=lk,
is_causal=causal,
softmax_scale=softmax_scale,
)
try:
x = x.unflatten(0, (b, lq))
except:
x = x[0].unflatten(0, (b, lq))
else:
assert FLASH_ATTN_2_AVAILABLE
x = flash_attn.flash_attn_varlen_func(
Expand Down