Skip to content

[Dont Merge] Simple Append Attention impl for TPA & MFA benchmark #10419

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
l1cacheDell wants to merge 5 commits into
base: develop
Choose a base branch
from
Open

[Dont Merge] Simple Append Attention impl for TPA & MFA benchmark #10419

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
e4088cc
simple append attn kernel
l1cacheDell Mar 28, 2025
909d825
stage
l1cacheDell Apr 2, 2025
aacf82a
modify dispatch
l1cacheDell Apr 16, 2025
5b3ee47
new
l1cacheDell Apr 16, 2025
1409654
upload
l1cacheDell Apr 17, 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
1 change: 0 additions & 1 deletion csrc/gpu/append_attn/append_attention_c16_impl.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1429,7 +1429,6 @@ void MultiQueryAppendAttention(
static_cast<float *>(tmp_d->ptr()),
reinterpret_cast<OUT_NV_TYPE *>(out->data<OutT>()),
speculate_max_draft_token_num);

// merge
constexpr int vec_size = num_elems_per_128b<NV_TYPE>();
if (is_decoder) {
Expand Down
1 change: 0 additions & 1 deletion csrc/gpu/append_attn/get_block_shape_and_split_kv_block.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,7 +219,6 @@ std::vector<paddle::Tensor> GetBlockShapeAndSplitKVBlock(
);
auto max_len_kv_cpu =
max_len_kv.copy_to(paddle::CPUPlace(), false);

// decoder
int max_dec_len_this_time_data = max_dec_len_this_time.data<int>()[0];
if (max_dec_len_this_time_data > 0) {
Expand Down
13 changes: 13 additions & 0 deletions csrc/gpu/append_attn/utils.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -284,6 +284,16 @@ __forceinline__ __host__ __device__ void vec_cast<nv_bfloat16, float>(
__VA_ARGS__ \
break; \
} \
case 256: { \
constexpr size_t HEAD_DIM = 256; \
__VA_ARGS__ \
break; \
} \
case 512: { \
constexpr size_t HEAD_DIM = 512; \
__VA_ARGS__ \
break; \
} \
default: { \
PD_THROW("not support the head_dim: ", head_dim); \
} \
Expand Down Expand Up @@ -377,6 +387,9 @@ __forceinline__ __host__ __device__ void vec_cast<nv_bfloat16, float>(
} else if (group_size == 8) { \
constexpr size_t GROUP_SIZE = 8; \
__VA_ARGS__ \
} else if (group_size == 64) { \
constexpr size_t GROUP_SIZE = 64; \
__VA_ARGS__ \
} else if (group_size == 16) { \
constexpr size_t GROUP_SIZE = 16; \
__VA_ARGS__ \
Expand Down
Loading
Loading