Skip to content

Attempt to fix #2220: Enable Flash Attention in KV-cache path #2248

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
woaiwang wants to merge 1 commit into
base: main
Choose a base branch
from
+6 −0
Open

Attempt to fix #2220: Enable Flash Attention in KV-cache path #2248

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
6 changes: 6 additions & 0 deletions litgpt/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -343,6 +343,12 @@ def scaled_dot_product_attention(
scores = torch.nn.functional.softmax(scores, dim=-1, dtype=torch.float).to(dtype=q.dtype)
y = scores @ v
else:
# 修复 #2220: 在单步解码 (decoding) 阶段，q 的序列长度为 1，
# 此时不需要因果掩码，丢弃 mask 可以让 PyTorch 启用 Flash Attention 加速。
if mask is not None and q.size(2) == 1:
mask = None

# 注意这里！把 F. 换成了完整的 torch.nn.functional.
y = torch.nn.functional.scaled_dot_product_attention(
q, k, v, attn_mask=mask, dropout_p=0.0, scale=scale, is_causal=mask is None
)
Expand Down
Loading