Skip to content
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

Support Seq length up to 8K #756

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Support Seq length up to 8K #756

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
00b1eba
seq len supported up to 8K
Aug 30, 2023
4a2c57a
fix typo
Aug 30, 2023
6ab4140
Merge branch 'main' of github.com:NVIDIA/FasterTransformer into main
Nov 8, 2023
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
17 changes: 13 additions & 4 deletions src/fastertransformer/kernels/unfused_attention_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -660,7 +660,10 @@ void invokeMaskedSoftmax(MaskedSoftmaxParam<T, T_IN>& param, cudaStream_t stream
bool is_half2 = sizeof(T) == 2 && sizeof(T_IN) == 2 && param.k_length % 2 == 0;
dim3 block((param.k_length / (is_half2 ? 2 : 1) + 31) / 32 * 32);

if (block.x > 2048 && block.x <= 4096) {
if (block.x > 4096 && block.x <= 8192) {
LAUNCH_MAKSED_SOFTMAX(8);
}
else if (block.x > 2048 && block.x <= 4096) {
LAUNCH_MAKSED_SOFTMAX(4)
}
else if (block.x > 1024) {
Expand Down Expand Up @@ -697,8 +700,11 @@ void invokeMaskedSoftmax(MaskedSoftmaxParam<__nv_bfloat16, float>& param, cudaSt

bool is_half2 = sizeof(T) == 2 && sizeof(T_IN) == 2 && param.k_length % 2 == 0;
dim3 block((param.k_length / (is_half2 ? 2 : 1) + 31) / 32 * 32);

if (block.x > 2048 && block.x <= 4096) {

if (block.x > 4096 && block.x <= 8192) {
LAUNCH_MAKSED_SOFTMAX_(__nv_bfloat16, 8);
}
else if (block.x > 2048 && block.x <= 4096) {
LAUNCH_MAKSED_SOFTMAX_(__nv_bfloat16, 4);
}
else if (block.x > 1024) {
Expand Down Expand Up @@ -730,7 +736,10 @@ void invokeMaskedSoftmax(MaskedSoftmaxParam<__nv_bfloat16, __nv_bfloat16>& param
bool is_half2 = sizeof(T) == 2 && sizeof(T_IN) == 2 && param.k_length % 2 == 0;
dim3 block((param.k_length / (is_half2 ? 2 : 1) + 31) / 32 * 32);

if (block.x > 2048 && block.x <= 4096) {
if (block.x > 4096 && block.x <= 8192) {
LAUNCH_MAKSED_SOFTMAX_(__nv_bfloat16, 8);
}
else if (block.x > 2048 && block.x <= 4096) {
LAUNCH_MAKSED_SOFTMAX_(__nv_bfloat16, 4);
}
else if (block.x > 1024) {
Expand Down