rm detach (#34644)

Author: ForFishes

python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py

@@ -257,7 +257,7 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
             for d in tensor_send_prev:
                 paddle.distributed.wait(d, use_calc_stream=True)
                 send_partial(
-                    d.detach(),
+                    d,
                     dst=0,
                     nranks=mp_degree,
                     rank_id=mp_rank,
@@ -266,7 +266,7 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
         else:
             paddle.distributed.wait(tensor_send_prev, use_calc_stream=True)
             send_partial(
-                tensor_send_prev.detach(),
+                tensor_send_prev,
                 dst=0,
                 nranks=mp_degree,
                 rank_id=mp_rank,
@@ -277,28 +277,28 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
         if isinstance(tensor_recv_prev, tuple):
             for d in tensor_recv_prev:
                 recv_partial(
-                    d.detach(),
+                    d,
                     src=0,
                     nranks=mp_degree,
                     rank_id=mp_rank,
                     group=_hcg.recv_prev_group,
                     use_calc_stream=True)
                 allgather_partial(
-                    d.detach(),
+                    d,
                     nranks=mp_degree,
                     rank_id=mp_rank,
                     group=mp_group,
                     use_calc_stream=True)
         else:
             recv_partial(
-                tensor_recv_prev.detach(),
+                tensor_recv_prev,
                 src=0,
                 nranks=mp_degree,
                 rank_id=mp_rank,
                 group=_hcg.recv_prev_group,
                 use_calc_stream=True)
             allgather_partial(
-                tensor_recv_prev.detach(),
+                tensor_recv_prev,
                 nranks=mp_degree,
                 rank_id=mp_rank,
                 group=mp_group,
@@ -309,7 +309,7 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
             for d in tensor_send_next:
                 paddle.distributed.wait(d, use_calc_stream=True)
                 send_partial(
-                    d.detach(),
+                    d,
                     dst=1,
                     nranks=mp_degree,
                     rank_id=mp_rank,
@@ -318,7 +318,7 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
         else:
             paddle.distributed.wait(tensor_send_next, use_calc_stream=True)
             send_partial(
-                tensor_send_next.detach(),
+                tensor_send_next,
                 dst=1,
                 nranks=mp_degree,
                 rank_id=mp_rank,
@@ -329,30 +329,30 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
         if isinstance(tensor_recv_next, tuple):
             for d in tensor_recv_next:
                 recv_partial(
-                    d.detach(),
+                    d,
                     src=1,
                     nranks=mp_degree,
                     rank_id=mp_rank,
                     group=_hcg.recv_next_group,
                     use_calc_stream=True)
                 allgather_partial(
-                    d.detach(),
+                    d,
                     nranks=mp_degree,
                     rank_id=mp_rank,
                     group=mp_group,
                     use_calc_stream=True)
 
         else:
             recv_partial(
-                tensor_recv_next.detach(),
+                tensor_recv_next,
                 src=1,
                 nranks=mp_degree,
                 rank_id=mp_rank,
                 group=_hcg.recv_next_group,
                 use_calc_stream=True)
 
             allgather_partial(
-                tensor_recv_next.detach(),
+                tensor_recv_next,
                 nranks=mp_degree,
                 rank_id=mp_rank,
                 group=mp_group,