Revert pull request 34212 (#34558)

* revert commit id 34212

paddle/fluid/framework/distributed_strategy.proto

@@ -183,7 +183,7 @@ message DistributedStrategy {
   optional bool use_hierarchical_allreduce = 15 [ default = false ];
   optional int32 hierarchical_allreduce_inter_nranks = 16 [ default = 1 ];
   optional bool sync_batch_norm = 17 [ default = false ];
-  optional bool fuse_all_reduce_ops = 18 [ default = false ];
+  optional bool fuse_all_reduce_ops = 18 [ default = true ];
   optional int32 fuse_grad_size_in_MB = 19 [ default = 32 ];
   optional float fuse_grad_size_in_TFLOPS = 20 [ default = 50 ];
   optional bool cudnn_exhaustive_search = 21 [ default = false ];

python/paddle/fluid/tests/unittests/test_fleet_sharding_meta_optimizer.py

@@ -366,6 +366,8 @@ class TestFleetShardingHybridOptimizer(TestFleetMetaOptimizer):
             "gradient_merge_acc_step": 1,
             "mp_degree": 1
         }
+
+        strategy.fuse_all_reduce_ops = False
         self.optimizer(avg_cost, strategy, train_prog, startup_prog)
         startup_prog_ops = startup_prog.global_block().ops
         main_prog_ops = train_prog.global_block().ops