diff --git a/ggml/src/ggml-cuda/unary.cu b/ggml/src/ggml-cuda/unary.cu
index 77ef81545..bb048ba4b 100644
--- a/ggml/src/ggml-cuda/unary.cu
+++ b/ggml/src/ggml-cuda/unary.cu
@@ -213,8 +213,8 @@ static __global__ void unary_gated_op_kernel(const T * x, T * dst, const int k,
 
 template <float (*op)(float), typename T>
 static void unary_gated_cuda(const T * x, T * dst, const int k, const int n, const int o, cudaStream_t stream) {
-    const int num_blocks = (k + CUDA_NEG_BLOCK_SIZE - 1) / CUDA_NEG_BLOCK_SIZE;
-    unary_gated_op_kernel<op><<<num_blocks, CUDA_NEG_BLOCK_SIZE, 0, stream>>>(x, dst, k, n, o);
+    const int num_blocks = (k + CUDA_GLU_BLOCK_SIZE - 1) / CUDA_GLU_BLOCK_SIZE;
+    unary_gated_op_kernel<op><<<num_blocks, CUDA_GLU_BLOCK_SIZE, 0, stream>>>(x, dst, k, n, o);
 }
 
 template <float (*op)(float)>
diff --git a/ggml/src/ggml-cuda/unary.cuh b/ggml/src/ggml-cuda/unary.cuh
index d4533d24e..9094f1d0b 100644
--- a/ggml/src/ggml-cuda/unary.cuh
+++ b/ggml/src/ggml-cuda/unary.cuh
@@ -15,6 +15,7 @@
 #define CUDA_SQRT_BLOCK_SIZE 256
 #define CUDA_SIN_BLOCK_SIZE 256
 #define CUDA_COS_BLOCK_SIZE 256
+#define CUDA_GLU_BLOCK_SIZE 256
 
 void ggml_cuda_op_abs(ggml_backend_cuda_context & ctx, ggml_tensor * dst);