Skip to content

Commit 885ad07

Browse files
committed
ggml : disable FA for DS head sizes
1 parent f7e8397 commit 885ad07

File tree

2 files changed

+5
-0
lines changed

2 files changed

+5
-0
lines changed

ggml/src/ggml-cuda/ggml-cuda.cu

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3236,6 +3236,10 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
32363236
if (op->src[0]->ne[0] == 192) {
32373237
return false;
32383238
}
3239+
if (op->src[0]->ne[0] == 576) {
3240+
// DeepSeek MLA
3241+
return false;
3242+
}
32393243
if (op->src[0]->ne[3] != 1) {
32403244
return false;
32413245
}

ggml/src/ggml-vulkan/ggml-vulkan.cpp

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -9261,6 +9261,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
92619261
case 112:
92629262
case 128:
92639263
case 256:
9264+
case 575: // DeepSeek MLA
92649265
break;
92659266
default:
92669267
return false;

0 commit comments

Comments
 (0)