Skip to content

Bugfix: append_kv out of bound when using GQA #16

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
yzh119 merged 1 commit into from
Nov 24, 2023
Merged

Bugfix: append_kv out of bound when using GQA #16

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
4 changes: 2 additions & 2 deletions include/flashinfer/page.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -497,7 +497,7 @@ cudaError_t AppendPagedKVCacheDecode(paged_kv_t<page_storage, DType, IdType> pag
SWITCH_HEAD_DIM(head_dim, HEAD_DIM, {
constexpr uint32_t vec_size = std::max(16 / sizeof(DType), HEAD_DIM / 32);
constexpr uint32_t bdx = HEAD_DIM / vec_size;
constexpr uint32_t bdy = 128 / bdx;
constexpr uint32_t bdy = 1;
assert(num_heads % bdy == 0);
dim3 nblks(batch_size * num_heads / bdy);
dim3 nthrs(bdx, bdy);
Expand Down Expand Up @@ -531,7 +531,7 @@ cudaError_t AppendPagedKVCachePrefill(paged_kv_t<page_storage, DType, IdType> pa
SWITCH_HEAD_DIM(head_dim, HEAD_DIM, {
constexpr uint32_t vec_size = std::max(16 / sizeof(DType), HEAD_DIM / 32);
constexpr uint32_t bdx = HEAD_DIM / vec_size;
constexpr uint32_t bdy = 128 / bdx;
constexpr uint32_t bdy = 1;
assert(num_heads % bdy == 0);
dim3 nblks(batch_size * num_heads / bdy);
dim3 nthrs(bdx, bdy);
Expand Down