Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

CUDA: tune GLM 4.7 Flash FA kernel selection logic (DGX Spark) (#19142)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-01-28 09:15:11 +02:00
committed by GitHub
parent b931f81b5a
commit 2eee6c866c
2 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/fattn.cu
+8
View File
@@ -147,6 +147,14 @@ static void ggml_cuda_flash_attn_ext_mma_f16(ggml_backend_cuda_context & ctx, gg
GGML_ASSERT(Q->ne[2] % K->ne[2] == 0);
const int gqa_ratio = Q->ne[2] / K->ne[2];
if (gqa_ratio == 20) { // GLM 4.7 Flash
if (cc >= GGML_CUDA_CC_DGX_SPARK) {
if (Q->ne[1] <= 8) {
ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<576, 512, 16>(ctx, dst);
break;
}
ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<576, 512, 4>(ctx, dst);
break;
}
if (cc >= GGML_CUDA_CC_BLACKWELL) {
if (Q->ne[1] <= 4 && K->ne[1] >= 65536) {
ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<576, 512, 16>(ctx, dst);