diff --git a/ge/graph/manager/graph_caching_allocator.cc b/ge/graph/manager/graph_caching_allocator.cc
index 82bfbda9..d08fd50d 100644
--- a/ge/graph/manager/graph_caching_allocator.cc
+++ b/ge/graph/manager/graph_caching_allocator.cc
@@ -163,6 +163,7 @@ Status CachingAllocator::Free(uint8_t *ptr, uint32_t device_id) {
   }
 
   std::lock_guard<std::recursive_mutex> lock(mutex_);
+  GELOGI("=============block ptr is %p",ptr);
   auto it = allocated_blocks_.find(ptr);
   if (it == allocated_blocks_.end()) {
     REPORT_INNER_ERROR("E19999", "Param ptr not allocated before, device_id:%u, check invalid", device_id);
diff --git a/ge/hybrid/common/tensor_value.h b/ge/hybrid/common/tensor_value.h
index 348e4e6d..3e0c9d5c 100644
--- a/ge/hybrid/common/tensor_value.h
+++ b/ge/hybrid/common/tensor_value.h
@@ -41,8 +41,10 @@ class TensorBuffer {
   ~TensorBuffer();
 
   void* Release() {
+    GELOGI("===========buffer is %p",buffer_);
     auto ret = buffer_;
     buffer_ = nullptr;
+    GELOGI("============ret is %p", ret);
     return ret;
   }
 
diff --git a/ge/hybrid/executor/hybrid_model_async_executor.cc b/ge/hybrid/executor/hybrid_model_async_executor.cc
index f05d7c13..cec5695e 100644
--- a/ge/hybrid/executor/hybrid_model_async_executor.cc
+++ b/ge/hybrid/executor/hybrid_model_async_executor.cc
@@ -489,6 +489,7 @@ void HybridModelAsyncExecutor::BuildDeviceTensor(TensorValue &output_tensor, GeT
   GELOGD("Mem type is %d", static_cast<uint32_t>(mem_type));
   auto deleter = [=](uint8_t *device_data) {
     if (device_data != nullptr) {
+      GELOGI("Free device addr int build device tensor is %p",device_data);
       if (mem_type == RDMA_HBM) {
         MemManager::Instance().RdmaPoolInstance(RT_MEMORY_HBM).Free(device_data, device_id_);
       } else if (mem_type == HOST_DDR) {