Add fused_allreduce_gradients_with_group for PPFleetX (#47447)

* add fused_allreduce_gradients_with_group

* add scale

* fix ci

python/paddle/distributed/fleet/utils/hybrid_parallel_util.py

@@ -26,7 +26,7 @@ from .log_util import logger
 __all__ = []
 
 
-def _apply_collective_grads(parameters, comm_group):
+def _apply_collective_grads(parameters, comm_group, bucket_size, scale=None):
     grad_var_set = set()
     grad_vars = []
     sparse_grad_vars = []
@@ -41,16 +41,21 @@ def _apply_collective_grads(parameters, comm_group):
             assert g_var not in grad_var_set
             grad_var_set.add(g_var)
 
-    coalesced_grads_and_vars = build_groups(grad_vars, 128 * 1024 * 1024)
+    coalesced_grads_and_vars = build_groups(grad_vars, bucket_size)
 
     nranks = (
         paddle.distributed.get_world_size()
         if comm_group is None
         else comm_group.nranks
     )
+
+    scale = nranks if scale is None else 1.0 / scale
+    scale = None if scale == 1.0 else scale
+
     for coalesced_grad, _, _ in coalesced_grads_and_vars:
         # need to div nranks
-        div_factor = paddle.to_tensor(nranks, dtype=coalesced_grad.dtype)
+        if scale is not None:
+            div_factor = paddle.to_tensor(scale, dtype=coalesced_grad.dtype)
             paddle.fluid.framework._dygraph_tracer().trace_op(
                 type="elementwise_div",
                 inputs={'X': coalesced_grad, 'Y': div_factor},
@@ -62,7 +67,9 @@ def _apply_collective_grads(parameters, comm_group):
     _split_tensors(coalesced_grads_and_vars)
 
 
-def _apply_collective_grads_eager(parameters, comm_group):
+def _apply_collective_grads_eager(
+    parameters, comm_group, bucket_size, scale=None
+):
     grad_var_set = set()
     grad_vars = []
 
@@ -76,16 +83,21 @@ def _apply_collective_grads_eager(parameters, comm_group):
             assert g_var not in grad_var_set
             grad_var_set.add(g_var)
 
-    coalesced_grads_and_vars = build_groups(grad_vars, 128 * 1024 * 1024)
+    coalesced_grads_and_vars = build_groups(grad_vars, bucket_size)
 
     nranks = (
         paddle.distributed.get_world_size()
         if comm_group is None
         else comm_group.nranks
     )
+
+    scale = 1.0 / nranks if scale is None else scale
+    scale = None if scale == 1.0 else scale
+
     for coalesced_grad, _, _ in coalesced_grads_and_vars:
         # need to div nranks
-        coalesced_grad.scale_(1.0 / nranks)
+        if scale is not None:
+            coalesced_grad.scale_(scale)
         paddle.distributed.all_reduce(coalesced_grad, group=comm_group)
 
     _split_tensors(coalesced_grads_and_vars)
@@ -172,16 +184,22 @@ def broadcast_dp_parameters(model, hcg):
     )
 
 
-def fused_allreduce_gradients(parameter_list, hcg):
-    data_parallel_group = None if hcg is None else hcg.get_data_parallel_group()
-    logger.debug("dp start fuse allreduce gradients")
+def fused_allreduce_gradients_with_group(
+    parameter_list, group, bucket_size=128 * 1024 * 1024, scale=None
+):
     apply_func = (
         _apply_collective_grads_eager
         if in_dygraph_mode()
         else _apply_collective_grads
     )
     with framework.no_grad():
-        apply_func(parameter_list, data_parallel_group)
+        apply_func(parameter_list, group, bucket_size)
+
+
+def fused_allreduce_gradients(parameter_list, hcg):
+    data_parallel_group = None if hcg is None else hcg.get_data_parallel_group()
+    logger.debug("dp start fuse allreduce gradients")
+    fused_allreduce_gradients_with_group(parameter_list, data_parallel_group)
 
 
 def sharding_reduce_gradients(parameter_list, hcg):