happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

CUDA: fix alignment check for FA (#19023)

This commit is contained in:
Johannes Gäßler 2026-01-22 20:39:25 +01:00 committed by GitHub
parent e34d6d03b2
commit e2baf02162
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cuda/fattn.cu
View File

@ -46,7 +46,7 @@ static void ggml_cuda_flash_attn_ext_mma_f16_switch_ncols2(ggml_backend_cuda_con
// are put into the template specialization without GQA optimizations.
bool use_gqa_opt = mask && max_bias == 0.0f && K->ne[1] % FATTN_KQ_STRIDE == 0;
for (const ggml_tensor * t : {Q, K, V, mask}) {
if (t == nullptr) {
if (t == nullptr || ggml_is_quantized(t->type)) {
continue;
}
for (size_t i = 1; i < GGML_MAX_DIMS; ++i) {
@ -236,7 +236,7 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
// The kernel versions without this optimization are also used for ALiBi, if there is no mask, or if the KV cache is not padded,
bool gqa_opt_applies = gqa_ratio % 2 == 0 && mask && max_bias == 0.0f && K->ne[1] % FATTN_KQ_STRIDE == 0;
for (const ggml_tensor * t : {Q, K, V, mask}) {
if (t == nullptr) {
if (t == nullptr || ggml_is_quantized(t->type)) {
continue;
}
for (size_t i = 1; i < GGML_MAX_DIMS; ++i) {