From: Johannes Gäßler <redacted>
Date: Wed, 1 Apr 2026 19:28:19 +0000 (+0200)
Subject: CUDA: fix FA kernel selection logic (llama/21271)
X-Git-Tag: v0.9.11~4
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=878642b85d1d6f2ba2d9f98e3298fab4d504500f;p=pkg%2Fggml%2Fsources%2Fggml

CUDA: fix FA kernel selection logic (llama/21271)
---

diff --git a/src/ggml-cuda/fattn.cu b/src/ggml-cuda/fattn.cu
index a21c5361..addf9320 100644
--- a/src/ggml-cuda/fattn.cu
+++ b/src/ggml-cuda/fattn.cu
@@ -340,7 +340,14 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
         case 128:
         case 112:
         case 256:
+            if (V->ne[0] != K->ne[0]) {
+                return BEST_FATTN_KERNEL_NONE;
+            }
+            break;
         case 512:
+            if (V->ne[0] != K->ne[0]) {
+                return BEST_FATTN_KERNEL_NONE;
+            }
             if (!gqa_opt_applies) {
                 return BEST_FATTN_KERNEL_NONE;
             }