CUDA: fix typo in FlashAttention code (llama/13926)

diff --git a/src/ggml-cuda/fattn-mma-f16.cuh b/src/ggml-cuda/fattn-mma-f16.cuh
index 7120053b6ee01efe73b729ee83f1e352c0ce12d8..925f39e890db927eee9c27cef3f50a575838e7af 100644 (file)
--- a/src/ggml-cuda/fattn-mma-f16.cuh
+++ b/src/ggml-cuda/fattn-mma-f16.cuh
@@ -1246,7 +1246,7 @@ static __global__ void flash_attn_ext_f16(
         NO_DEVICE_CODE;
         return;
     }
-#endif __CUDA_ARCH__ == GGML_CUDA_CC_TURING
+#endif // __CUDA_ARCH__ == GGML_CUDA_CC_TURING
 
     static_assert(!mla || DKQ >= DV, "MLA needs DKQ >= DV");