diff --git a/ge/common/helper/model_cache_helper.cc b/ge/common/helper/model_cache_helper.cc
index 9cd88ef1..0e6c6329 100755
--- a/ge/common/helper/model_cache_helper.cc
+++ b/ge/common/helper/model_cache_helper.cc
@@ -1679,6 +1679,13 @@ Status ModelCacheHelper::LoadOmModelFromCache(GeModelPtr &ge_model) const {
     GELOGW("LoadOmModelFromCache: Load model from file failed. ret = %u", ret);
     return ret;
   }
+  std::function<void()> callback = [&]() {
+    if (model_data.model_data != nullptr) {
+      delete[] reinterpret_cast<char *>(model_data.model_data);
+      model_data.model_data = nullptr;
+    }
+  };
+  GE_MAKE_GUARD(release, callback);
 
   ModelHelper model_helper;
   ret = model_helper.LoadModel(model_data);
diff --git a/ge/hybrid/node_executor/hccl/hccl_node_executor.cc b/ge/hybrid/node_executor/hccl/hccl_node_executor.cc
index 7672a1e3..0412e06d 100644
--- a/ge/hybrid/node_executor/hccl/hccl_node_executor.cc
+++ b/ge/hybrid/node_executor/hccl/hccl_node_executor.cc
@@ -445,6 +445,7 @@ Status AllToAllNodeTask::ExecuteAsync(TaskContext &context, std::function<void()
       return FAILED;
     }
     HcomAllToAllVParams params;
+    params.group = nullptr;
     GE_CHK_STATUS_RET(BuildAllToAllVparams(context, params));
     HcclResult hccl_ret = HcomExecEnqueueAllToAllV(params, callback);
     if (hccl_ret != HCCL_SUCCESS) {