fix dev 2.6 (#60080)

wentaoyu · web-flow · commit 210442ec30e5 · 2023-12-19T00:25:09.000+08:00
diff --git a/python/paddle/distributed/fleet/meta_optimizers/dygraph_optimizer/hybrid_parallel_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/dygraph_optimizer/hybrid_parallel_optimizer.py
@@ -12,7 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import distutils.util
 import os
 
 import paddle
@@ -47,9 +46,6 @@ def __init__(self, clip, hcg):
         self._clip = clip
         self._hcg = hcg
         self.not_sharding_stage1 = True
-        self._force_align_vpp_grad_sum_order = distutils.util.strtobool(
-            os.getenv('FLAGS_force_align_vpp_grad_sum_order', '0')
-        )
 
     def _global_norm(self, global_norm_var_dist, global_norm_var_not_dist):
         # sharding first
@@ -103,10 +99,6 @@ def _global_norm(self, global_norm_var_dist, global_norm_var_not_dist):
 
     @no_grad()
     def _dygraph_clip(self, params_grads):
-        if self._force_align_vpp_grad_sum_order:
-            chunk_num = self._get_vpp_chunk_num(params_grads)
-            if chunk_num > 0:
-                return self._vpp_dygraph_clip(params_grads, chunk_num)
         sum_square_dist_fp16 = []
         sum_square_dist_bf16 = []
         sum_square_dist_fp32 = []
