nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Use fused gemv+add only for TG (#933) 

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow 2025-11-10 08:34:24 +02:00 committed by GitHub
parent 56ee303254
commit bf474e9bff
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda.cu
View File

@ -2067,7 +2067,7 @@ static int ggml_cuda_mul_mat_q(ggml_backend_cuda_context & ctx, const ggml_tenso
auto stream = ctx.stream();
auto fusion = ctx.fusion;
auto fusion = ctx.fusion && src1->ne[1] == 1;
auto ne10_padded = GGML_PAD(src1->ne[0], MATRIX_ROW_PADDING);
auto nb10_padded = ne10_padded*sizeof(block_q8_1)/QK8_1;