diff --git a/python/paddle/distributed/fleet/base/topology.py b/python/paddle/distributed/fleet/base/topology.py
index 8f38ba447fcb3d59a2c609dacff7c921f01935fd..470a4d83aac3fedb2135d44567fe31688894b093 100644
--- a/python/paddle/distributed/fleet/base/topology.py
+++ b/python/paddle/distributed/fleet/base/topology.py
@@ -147,7 +147,7 @@ class HybridCommunicateGroup(object):
         debug_str = "HybridParallelInfo: rank_id: %d, dp_degree: %d, " \
                     "mp_degree: %d, pp_degree: %d" % (self.global_rank, self._dp_degree,
                     self._mp_degree,self._pp_degree)
-        debug_str += "dp_group: %s, mp_group: %s, pp_group: %s, check/clip group: %s" % (
+        debug_str += ", dp_group: %s, mp_group: %s, pp_group: %s, check/clip group: %s" % (
             self._dp_group, self._mp_group, self._pp_group, self._check_group)
         logger.info(debug_str)
 
diff --git a/python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py b/python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py
index 280f1a06e1bfb58a4bf44f4e4dbd3c79c0beede8..79e5bc2ffeda06d62b24aec2e10ae3ad071d856a 100644
--- a/python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py
+++ b/python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py
@@ -136,9 +136,9 @@ class PipelineParallel(MetaParallelBase):
             self._recv_activations(cache_id)
 
         if isinstance(self.caches['inputs'][cache_id], tuple):
-            inputs = tuple(t.clone() for t in self.caches['inputs'][cache_id])
+            inputs = tuple(t for t in self.caches['inputs'][cache_id])
         else:
-            inputs = self.caches['inputs'][cache_id].clone()
+            inputs = self.caches['inputs'][cache_id]
 
         self._clear_grads(inputs)
         outputs = self._layers.forward(inputs)