diff --git a/tools/program.py b/tools/program.py
index 9d574a370b..a541472da1 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -17,6 +17,7 @@
 from __future__ import print_function
 
 import os
+import gc
 import sys
 import platform
 import yaml
@@ -208,6 +209,7 @@ def train(
     profiler_options = config["profiler_options"]
     print_mem_info = config["Global"].get("print_mem_info", True)
     uniform_output_enabled = config["Global"].get("uniform_output_enabled", False)
+    export_during_train = config["Global"].get("export_during_train", False)
 
     global_step = 0
     if "global_step" in pre_best_model_dict:
@@ -489,12 +491,14 @@ def train(
                     best_model_dict.update(cur_metric)
                     best_model_dict["best_epoch"] = epoch
                     prefix = "best_accuracy"
-                    if uniform_output_enabled:
+                    if export_during_train:
                         export(
                             config,
                             model,
                             os.path.join(save_model_dir, prefix, "inference"),
                         )
+                        gc.collect()
+                    if uniform_output_enabled:
                         model_info = {"epoch": epoch, "metric": best_model_dict}
                     else:
                         model_info = None
@@ -540,8 +544,10 @@ def train(
             reader_start = time.time()
         if dist.get_rank() == 0:
             prefix = "latest"
-            if uniform_output_enabled:
+            if export_during_train:
                 export(config, model, os.path.join(save_model_dir, prefix, "inference"))
+                gc.collect()
+            if uniform_output_enabled:
                 model_info = {"epoch": epoch, "metric": best_model_dict}
             else:
                 model_info = None
@@ -568,8 +574,10 @@ def train(
 
         if dist.get_rank() == 0 and epoch > 0 and epoch % save_epoch_step == 0:
             prefix = "iter_epoch_{}".format(epoch)
-            if uniform_output_enabled:
+            if export_during_train:
                 export(config, model, os.path.join(save_model_dir, prefix, "inference"))
+                gc.collect()
+            if uniform_output_enabled:
                 model_info = {"epoch": epoch, "metric": best_model_dict}
             else:
                 model_info = None