[Dygraph] Fix bugs in dp_pp_comm_overlap for HybridParallel (#53384)

python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py

@@ -75,6 +75,9 @@ class PipelineParallel(MetaParallelBase):
         ].dp_comm_overlap
         self._dp_comm_buffers = []
 
+        if self._dp_comm_overlap:
+            assert self.use_data_parallel and self.num_stages > 1
+
         p2p.initialize_p2p_groups(
             hcg, self._using_cache, self._enable_partial_send_recv
         )

python/paddle/distributed/fleet/optimizer.py

@@ -62,9 +62,16 @@ def _dygraph_distributed_optimizer(optimizer, strategy=None):
 
     if fleet_env.worker_num() > 1:
         if not fleet_env._user_defined_strategy.heter_ccl_mode:
-            return HybridParallelOptimizer(
+            hp_optim = HybridParallelOptimizer(
                 optimizer, fleet_env._hcg, fleet_env._user_defined_strategy
             )
+
+            if fleet_env._user_defined_strategy.hybrid_configs[
+                "pp_configs"
+            ].dp_comm_overlap:
+                hp_optim._dp_enable = False
+
+            return hp_optim
         else:
             return HeterParallelOptimizer(
                 optimizer, fleet_env._user_defined_strategy