Fix

Author: co63oc

python/paddle/distributed/auto_parallel/static/process_group.py

@@ -249,9 +249,7 @@ def instantiate(self):
                 alltoall_tmp = paddle.empty(
                     shape=[self.nranks, self.nranks], dtype="int32"
                 )
-                paddle._legacy_C_ops.alltoall(
-                    alltoall_tmp, 'use_calc_stream', True, 'ring_id', ring_id
-                )
+                paddle._C_ops.all_to_all(alltoall_tmp, ring_id)
                 paddle.device.cuda.synchronize()
 
         if self.nranks > 1: