LLM-Inference/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-08-12 22:29:39 +04:00
Code Issues Packages Projects Releases Wiki Activity

Fix #2201 (#2255)

Browse Source
This commit is contained in:
Kawrakow
2026-08-05 07:48:20 +03:00
committed by GitHub
parent 6633d19df8
commit 7257cabc49
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/getrows.cu
+2 -2
View File
@@ -213,13 +213,13 @@ void ggml_cuda_op_get_rows(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
auto type_size = row_size % 4 == 0 ? 4 : 2;
dim3 grid(dst->ne[1], dst->ne[2], dst->ne[3]);
if (type_size == 2) {
k_get_rows_dim1<<<grid, k_block_size, 0, ctx.stream()>>>(dst->ne[0],
k_get_rows_dim1<<<grid, k_block_size, 0, ctx.stream()>>>(row_size/type_size,
src0->nb[1]/type_size, src0->nb[2]/type_size, src0->nb[3]/type_size,
src1->nb[1]/type_size, src1->nb[2]/type_size,
dst->nb[1]/type_size, dst->nb[2]/type_size, dst->nb[3]/type_size,
(const uint16_t *)src0->data, (const int *)src1->data, (uint16_t *)dst->data);
} else {
k_get_rows_dim1<<<grid, k_block_size, 0, ctx.stream()>>>(dst->ne[0],
k_get_rows_dim1<<<grid, k_block_size, 0, ctx.stream()>>>(row_size/type_size,
src0->nb[1]/type_size, src0->nb[2]/type_size, src0->nb[3]/type_size,
src1->nb[1]/type_size, src1->nb[2]/type_size,
dst->nb[1]/type_size, dst->nb[2]/type_size, dst->nb[3]/type_size,