diff --git a/ppdet/modeling/backbones/vision_transformer.py b/ppdet/modeling/backbones/vision_transformer.py
index 26e6ff2a062c85c309a91123ee41b6d6469723ca..e566e6a3b9e4fb5e1d526c85cc306bca4c9ed4ee 100644
--- a/ppdet/modeling/backbones/vision_transformer.py
+++ b/ppdet/modeling/backbones/vision_transformer.py
@@ -352,7 +352,8 @@ class VisionTransformer(nn.Layer):
         self.final_norm = final_norm
 
         if use_checkpoint:
-            print('please set: FLAGS_allocator_strategy=naive_best_fit')
+            paddle.seed(0)
+
         self.patch_embed = PatchEmbed(
             img_size=img_size,
             patch_size=patch_size,