Skip to content

[Misc][Bugfix] FA3 support to ViT MHA layer #12435

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 2 commits into from
Jan 26, 2025
Merged

[Misc][Bugfix] FA3 support to ViT MHA layer #12435

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0d5228d
update flash attn API
ywang96 Jan 26, 2025
159f0f2
flatten qkv
Isotr0py Jan 26, 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
25 changes: 22 additions & 3 deletions vllm/attention/layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -251,9 +251,28 @@ def forward(
_Backend.FLASH_ATTN,
_Backend.FLASH_ATTN_VLLM_V1,
}:
from vllm.vllm_flash_attn import flash_attn_func

out = flash_attn_func(query, key, value, softmax_scale=self.scale)
from vllm.vllm_flash_attn import flash_attn_varlen_func

cu_seqlens_q = torch.arange(0, (bsz + 1) * q_len,
step=q_len,
dtype=torch.int32,
device=query.device)
cu_seqlens_k = torch.arange(0, (bsz + 1) * kv_len,
step=kv_len,
dtype=torch.int32,
device=key.device)

out = flash_attn_varlen_func(
query.flatten(0, 1),
key.flatten(0, 1),
value.flatten(0, 1),
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=q_len,
max_seqlen_k=kv_len,
softmax_scale=self.scale,
)
out = out.reshape(bsz, q_len, -1)
elif self.attn_backend == _Backend.XFORMERS:
from xformers import ops as xops

Expand Down
Loading