Skip to content

issue/1148: PagedAttentionPrefill 添加 KV cache 连续性 guard #1149

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
JoeZhang-0x000 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

issue/1148: PagedAttentionPrefill 添加 KV cache 连续性 guard #1149

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
8 changes: 8 additions & 0 deletions python/infinicore/ops/paged_attention_prefill.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,12 @@
from infinicore.tensor import Tensor


def _ensure_head_dim_contiguous(tensor: Tensor) -> Tensor:
if tensor.ndim > 0 and tensor.stride(tensor.ndim - 1) != 1:
return tensor.contiguous()
return tensor


def paged_attention_prefill(
q: Tensor,
k_cache: Tensor,
Expand All @@ -14,6 +20,8 @@ def paged_attention_prefill(
*,
out: Tensor | None = None,
):
k_cache = _ensure_head_dim_contiguous(k_cache)
v_cache = _ensure_head_dim_contiguous(v_cache)
alibi_ptr = alibi_slopes._underlying if alibi_slopes is not None else None

if out is None:
Expand Down