# Copyright 2021 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""adasum"""
import copy
import hashlib
import math
from mindspore.nn.cell import Cell
from mindspore.communication.management import create_group
from mindspore.ops import composite as C
from mindspore.ops import functional as F
from mindspore.ops import operations as P
from mindspore.ops.operations._inner_ops import Send, Receive
__all__ = ["AdaSum"]
MAX_NUM_HASH = 2 ** 31
_update_parameters = C.MultitypeFuncGraph("update_parameters")
@_update_parameters.register("Tensor", "Tensor", "Tensor", "Tensor")
def _update_parameters_after_broadcast(delta_weight, update_delta_weight, parameter, old_parameter):
shape = F.shape(delta_weight)
update_delta_weight = P.Reshape()(update_delta_weight, shape)
new_parameter = old_parameter - update_delta_weight
return P.Assign()(parameter, new_parameter)
def _send_before_receive(send_part, send, recv):
send_ok = send(send_part)
return recv(send_ok)
def _receive_before_send(send_part, send, recv):
receive_ok = recv(send_part)
send_part = F.depend(send_part, receive_ok)
return F.depend(receive_ok, send(send_part))
def _send_recv_res(left_send, recv_part, local_part, allreduce, parameter_divisibility, allreduce_node_num):
"""send result and receive result."""
if parameter_divisibility:
recv_part = P.Squeeze()(recv_part)
local_part = F.depend(local_part, recv_part)
eps = 1e-12
value_0 = P.ReduceSum()(local_part * recv_part) + eps
if left_send:
value_1 = P.ReduceSum()(local_part * local_part) + eps
value_2 = P.ReduceSum()(recv_part * recv_part) + eps
else:
value_1 = P.ReduceSum()(recv_part * recv_part) + eps
value_2 = P.ReduceSum()(local_part * local_part) + eps
value_0 = allreduce(value_0)
value_1 = F.depend(allreduce(value_1), value_0)
value_2 = F.depend(allreduce(value_2), value_1)
if left_send:
res = (1 - (value_0 / (2 * value_1))) * local_part + (1 - (value_0 / (2 * value_2))) * recv_part
else:
res = (1 - (value_0 / (2 * value_1))) * recv_part + (1 - (value_0 / (2 * value_2))) * local_part
else:
res = allreduce(local_part)
res /= allreduce_node_num
return res
_adasum_opt_forward = C.MultitypeFuncGraph("adasum_opt_forward")
@_adasum_opt_forward.register("Bool", "Function", "Bool", "Int64", "Function", "Function", "Tensor")
def _adasum_opt_forward_process(left_send, allreduce, parameter_divisibility, allreduce_node_num, send, recv, delta_w):
"""adasum optimizer process."""
if parameter_divisibility:
delta_w = P.Squeeze()(delta_w)
ori_len = F.shape(delta_w)[0]
divide_len = ori_len / 2
left_part = delta_w[:divide_len]
right_part = delta_w[divide_len:]
else:
left_part = delta_w
right_part = delta_w
if left_send:
if parameter_divisibility:
recv_part = _send_before_receive(left_part, send, recv)
else:
recv_part = right_part
update_delta_w = _send_recv_res(left_send, recv_part, right_part, allreduce, parameter_divisibility,
allreduce_node_num)
else:
if parameter_divisibility:
recv_part = _receive_before_send(right_part, send, recv)
else:
recv_part = left_part
update_delta_w = _send_recv_res(left_send, recv_part, left_part, allreduce, parameter_divisibility,
allreduce_node_num)
return update_delta_w
_adasum_opt_rollback = C.MultitypeFuncGraph("adasum_opt_rollback")
@_adasum_opt_rollback.register("Bool", "Bool", "Tensor", "Function", "Function")
def _adasum_opt_rollback_process(left_send, parameter_divisibility, delta_w, send, recv):
"""adasum optimizer rollback process."""
if parameter_divisibility:
if left_send:
recv_part = _send_before_receive(delta_w, send, recv)
else:
recv_part = _receive_before_send(delta_w, send, recv)
recv_part = P.Squeeze()(recv_part)
recv_part = P.Reshape()(recv_part, (-1,))
delta_w = P.Reshape()(delta_w, (-1,))
if left_send:
res = P.Concat()((recv_part, delta_w))
else:
res = P.Concat()((delta_w, recv_part))
else:
res = delta_w
return res
[文档]class AdaSum(Cell):
r"""
The Adaptive Summation, or AdaSum, is a novel algorithm for improving distributed data
parallel training of Deep Learning models.
Args:
rank (int): Rank number.
device_number (int): Device number.
group_number (int): Group number.
parameter_tuple (Tuple(Parameter)): Tuple of parameters.
Inputs:
- **delta_weights** (Tuple(Tensor)) - Tuple of gradients.
- **parameters** (Tuple(Parameter)) - Tuple of current parameters.
- **old_parameters** (Tuple(Parameter)) - Tuple of last parameters.
Outputs:
- **adasum_parameters** (Tuple(Tensor)) - Tuple of parameters after adasum process.
"""
def __init__(self, rank, device_number, group_number, parameter_tuple):
super(AdaSum, self).__init__()
self.rank = rank
self.device_number = device_number
self.group_number = group_number
self.parameter_tuple = parameter_tuple
self._generate_communication_op()
self.hyper_map = C.HyperMap()
def _generate_communication_op(self):
"""generate communication op."""
self.calc_times = int(math.log(self.group_number, 2))
self.send_node = []
self.send_list_forward = []
self.recv_list_forward = []
self.send_list_rollback = []
self.recv_list_rollback = []
self.allreduce_list = []
self.broadcast_list = []
self.parameter_divisibility_list = []
self.allreduce_node_num_list = []
last_delta_weights = []
group_start_rank = (self.rank // self.device_number) * self.device_number
for step in range(self.calc_times):
current_group = self.device_number * (2 ** step)
sr_target = self.rank
if (sr_target // current_group) % 2 == 0:
dest_target = sr_target + current_group
self.send_node.append(True)
else:
dest_target = sr_target - current_group
self.send_node.append(False)
neighbor_ids = []
group_name_last = 0
for index in range(2 ** (step + 1)):
node_rank = self.rank // self.device_number
double_d = 2 ** (step + 1)
neighbor_id = (node_rank // double_d * double_d + index) * self.device_number + \
self.rank % self.device_number
neighbor_ids.append(neighbor_id)
group_name_last += neighbor_id
group_name = "adasum_" + str(step) + "_" + str(group_name_last)
create_group(group_name, neighbor_ids)
send_left = []
send_right = []
recv_left = []
recv_right = []
allreduce_node_num = ()
left_delta_weights, right_delta_weights, delta_weights_divisibility = \
self._get_delta_weights_info(last_delta_weights)
self.parameter_divisibility_list.append(delta_weights_divisibility)
weights_index = 0
fusion_id = (step + 1) * 3
for shape, dtype in left_delta_weights:
send_tag = self._hash(step, sr_target, weights_index)
send = Send(sr_tag=send_tag, dest_rank=dest_target, group="hccl_world_group")
send.add_prim_attr("fusion", fusion_id)
recv_tag = self._hash(step, dest_target, weights_index)
recv = Receive(sr_tag=recv_tag, src_rank=dest_target, shape=shape, dtype=dtype,
group="hccl_world_group")
recv.add_prim_attr("fusion", fusion_id)
send_left.append(send)
recv_left.append(recv)
weights_index += 1
for shape, dtype in right_delta_weights:
send_tag = self._hash(step, sr_target, weights_index)
send = Send(sr_tag=send_tag, dest_rank=dest_target, group="hccl_world_group")
send.add_prim_attr("fusion", fusion_id + 1)
recv_tag = self._hash(step, dest_target, weights_index)
recv = Receive(sr_tag=recv_tag, src_rank=dest_target, shape=shape, dtype=dtype,
group="hccl_world_group")
recv.add_prim_attr("fusion", fusion_id + 1)
send_right.append(send)
recv_right.append(recv)
weights_index += 1
if self.send_node and self.send_node[-1]:
self.send_list_forward.append(send_left)
self.send_list_rollback.append(send_right)
self.recv_list_forward.append(recv_right)
self.recv_list_rollback.append(recv_left)
last_delta_weights = right_delta_weights
else:
self.send_list_forward.append(send_right)
self.send_list_rollback.append(send_left)
self.recv_list_forward.append(recv_left)
self.recv_list_rollback.append(recv_right)
last_delta_weights = left_delta_weights
server_all_reduce = P.AllReduce("sum", group_name)
server_all_reduce.add_prim_attr("fusion", fusion_id + 2)
self.allreduce_list.append(server_all_reduce)
for param_divisibility in delta_weights_divisibility:
if param_divisibility:
allreduce_node_num += (0,)
else:
allreduce_node_num += (2 ** (step + 1),)
self.allreduce_node_num_list.append(allreduce_node_num)
broadcast_group = [x for x in range(group_start_rank, group_start_rank + self.device_number)]
broadcast_group_name = "broadcast_group_" + str(group_start_rank)
create_group(broadcast_group_name, broadcast_group)
for b_rank in range(len(broadcast_group)):
self.broadcast_list.append(P.Broadcast(b_rank, group=broadcast_group_name))
self.sync_barrier = P.AllReduce("sum", group=broadcast_group_name)
def _get_delta_weights_info(self, last_delta_weights):
"""get delta weights info."""
half_delta_weights = []
if last_delta_weights:
half_delta_weights = last_delta_weights
else:
for parameter in self.parameter_tuple:
new_shape = [int(x) for x in parameter.shape]
half_delta_weights.append((new_shape, parameter.dtype))
left_delta_weights = []
right_delta_weights = []
delta_weights_divisibility = ()
for shape, dtype in half_delta_weights:
left_shape = copy.deepcopy(shape)
right_shape = copy.deepcopy(shape)
divisibility_flag = False
for i in range(len(shape)):
if shape[i] > 1:
left_shape[i] = int(shape[i] // 2)
right_shape[i] = shape[i] - int(shape[i] // 2)
divisibility_flag = True
break
left_delta_weights.append((left_shape, dtype))
right_delta_weights.append((right_shape, dtype))
delta_weights_divisibility += (divisibility_flag,)
return left_delta_weights, right_delta_weights, delta_weights_divisibility
def _hash(self, step, target, weights_index):
target = "tag" + str(step) + str(target) + str(weights_index)
target_hash = hashlib.sha1(target.encode()).hexdigest()
hash_res = int(int(target_hash, 16) % MAX_NUM_HASH)
return hash_res
def construct(self, delta_weights, parameters, old_parameters):
forward_weights = [delta_weights]
for i in range(self.calc_times):
process_weights = self.hyper_map(F.partial(_adasum_opt_forward, self.send_node[i], self.allreduce_list[i]),
self.parameter_divisibility_list[i], self.allreduce_node_num_list[i],
self.send_list_forward[i], self.recv_list_forward[i], forward_weights[-1])
forward_weights.append(process_weights)
for i in range(self.calc_times):
j = self.calc_times - i - 1
process_weights = self.hyper_map(F.partial(_adasum_opt_rollback, self.send_node[j]),
self.parameter_divisibility_list[j], forward_weights[j + 1],
self.send_list_rollback[j], self.recv_list_rollback[j])
forward_weights[j] = process_weights
adasum_parameters = self.hyper_map(F.partial(_update_parameters), delta_weights, forward_weights[0],
parameters, old_parameters)
return adasum_parameters