CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: fix bad asserts for partial offload (#13337)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-05-06 13:58:51 +02:00
committed by GitHub
parent 764b85627b
commit 2356fb1d53
6 changed files with 21 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/fattn-common.cuh
View File

@@ -719,6 +719,7 @@ void launch_fattn(
size_t nb23 = V->nb[3];
if (need_f16_K && K->type != GGML_TYPE_F16) {
GGML_ASSERT(ggml_is_contiguously_allocated(K));
K_f16.alloc(ggml_nelements(K));
to_fp16_cuda_t to_fp16 = ggml_get_to_fp16_cuda(K->type);
to_fp16(K_data, K_f16.ptr, ggml_nelements(K), main_stream);
@@ -733,6 +734,7 @@ void launch_fattn(
}
if (need_f16_V && V->type != GGML_TYPE_F16) {
GGML_ASSERT(ggml_is_contiguously_allocated(V));
V_f16.alloc(ggml_nelements(V));
to_fp16_cuda_t to_fp16 = ggml_get_to_fp16_cuda(V->type);
to_fp16(V_data, V_f16.ptr, ggml_nelements(V), main_stream);