diff --git a/ggml-opencl.cpp b/ggml-opencl.cpp
index 73088560b..52ba3aaac 100644
--- a/ggml-opencl.cpp
+++ b/ggml-opencl.cpp
@@ -983,11 +983,11 @@ static void ggml_cl_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor *
     cl_kernel* dmmv = ggml_get_dequantize_mul_mat_vec_cl(type);
     GGML_ASSERT(to_fp32_cl != nullptr);
 
+    size_t ev_idx = 0;
+    std::vector<cl_event> events;
+
     for (int64_t i03 = 0; i03 < ne03; i03++) {
         for (int64_t i02 = 0; i02 < ne02; i02++) {
-            size_t ev_idx = 0;
-            std::vector<cl_event> events;
-
             // copy src0 to device if necessary
             if (src0->backend == GGML_BACKEND_CPU) {
                 events.emplace_back();
@@ -1050,6 +1050,9 @@ static void ggml_cl_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor *
             for (auto *event : events) {
                 clReleaseEvent(event);
             }
+
+            ev_idx = 0;
+            events.clear();
         }
     }
 
diff --git a/ggml-opencl.h b/ggml-opencl.h
index 797b92c12..c850bb8ad 100644
--- a/ggml-opencl.h
+++ b/ggml-opencl.h
@@ -17,7 +17,7 @@ void * ggml_cl_host_malloc(size_t size);
 void   ggml_cl_host_free(void * ptr);
 
 void ggml_cl_transform_tensor(struct ggml_tensor * tensor);
-void ggml_cl_load_data(const char * fname, struct ggml_tensor * tensor, const size_t offset);
+void ggml_cl_load_data(const char * fname, struct ggml_tensor * tensor, size_t offset);
 
 #ifdef  __cplusplus
 }