mindspore/tests/ut/python/parallel/test_strategy_checkpoint.py

282 lines
11 KiB
Python

# Copyright 2019 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.
import numpy as np
from mindspore import context
from mindspore.context import set_auto_parallel_context
import mindspore.nn as nn
from mindspore.ops import operations as P
from mindspore import Tensor
from tests.ut.python.ops.test_math_ops import VirtualLoss
import mindspore as ms
from mindspore.common.api import _executor
from mindspore.ops import composite as C
# model_parallel test
# export PARALLEL_CHECKPOINT_ON=on
# export PARALLEL_TRAIN_TIMES=4
def test_six_matmul():
class NetWithLoss(nn.Cell):
def __init__(self, network):
super(NetWithLoss, self).__init__()
self.loss = VirtualLoss()
self.network = network
def construct(self, x1, x2, x3, x4, x5, x6, x7):
predict = self.network(x1, x2, x3, x4, x5, x6, x7)
return self.loss(predict)
class GradWrap(nn.Cell):
def __init__(self, network):
super(GradWrap, self).__init__()
self.network = network
def construct(self, x1, x2, x3, x4, x5, x6, x7):
return C.grad_all(self.network)(x1, x2, x3, x4, x5, x6, x7)
class Net(nn.Cell):
def __init__(self, strategy1, strategy2, strategy3, strategy4, strategy5, strategy6):
super().__init__()
self.matmul1 = P.MatMul().set_strategy(strategy1)
self.matmul2 = P.MatMul().set_strategy(strategy2)
self.matmul3 = P.MatMul().set_strategy(strategy3)
self.matmul4 = P.MatMul().set_strategy(strategy4)
self.matmul5 = P.MatMul().set_strategy(strategy5)
self.matmul6 = P.MatMul().set_strategy(strategy6)
def construct(self, x1, x2, x3, x4, x5, x6, x7):
out = self.matmul1(x1, x2)
out = self.matmul2(out, x3)
out = self.matmul3(out, x4)
out = self.matmul4(out, x5)
out = self.matmul5(out, x6)
out = self.matmul6(out, x7)
return out
set_auto_parallel_context(device_num=512, global_rank=0)
strategy1 = ((8, 1), (1, 1))
strategy2 = ((1, 8), (8, 1))
strategy3 = ((2, 2), (2, 2))
strategy4 = ((4, 2), (2, 4))
strategy5 = ((2, 4), (4, 2))
strategy6 = ((4, 4), (4, 4))
net = GradWrap(NetWithLoss(Net(strategy1, strategy2, strategy3, strategy4, strategy5, strategy6)))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
x1 = Tensor(np.ones([128, 32]), dtype=ms.float32)
x2 = Tensor(np.ones([32, 64]), dtype=ms.float32)
x3 = Tensor(np.ones([64, 64]), dtype=ms.float32)
x4 = Tensor(np.ones([64, 128]), dtype=ms.float32)
x5 = Tensor(np.ones([128, 64]), dtype=ms.float32)
x6 = Tensor(np.ones([64, 32]), dtype=ms.float32)
x7 = Tensor(np.ones([32, 32]), dtype=ms.float32)
_executor.compile(net, x1, x2, x3, x4, x5, x6, x7)
# remove matmul2
def test_six_matmul_repeated1():
class NetWithLoss(nn.Cell):
def __init__(self, network):
super(NetWithLoss, self).__init__()
self.loss = VirtualLoss()
self.network = network
def construct(self, x1, x2, x4, x5, x6, x7):
predict = self.network(x1, x2, x4, x5, x6, x7)
return self.loss(predict)
class GradWrap(nn.Cell):
def __init__(self, network):
super(GradWrap, self).__init__()
self.network = network
def construct(self, x1, x2, x4, x5, x6, x7):
return C.grad_all(self.network)(x1, x2, x4, x5, x6, x7)
class Net(nn.Cell):
def __init__(self, strategy1, strategy3, strategy4, strategy5, strategy6):
super().__init__()
self.matmul1 = P.MatMul().set_strategy(strategy1)
self.matmul3 = P.MatMul().set_strategy(strategy3)
self.matmul4 = P.MatMul().set_strategy(strategy4)
self.matmul5 = P.MatMul().set_strategy(strategy5)
self.matmul6 = P.MatMul().set_strategy(strategy6)
def construct(self, x1, x2, x4, x5, x6, x7):
out = self.matmul1(x1, x2)
out = self.matmul3(out, x4)
out = self.matmul4(out, x5)
out = self.matmul5(out, x6)
out = self.matmul6(out, x7)
return out
set_auto_parallel_context(device_num=512, global_rank=0)
strategy1 = ((8, 1), (1, 1))
strategy3 = ((8, 1), (1, 1))
strategy4 = ((8, 1), (1, 1))
strategy5 = ((8, 1), (1, 1))
strategy6 = ((8, 1), (1, 1))
net = GradWrap(NetWithLoss(Net(strategy1, strategy3, strategy4, strategy5, strategy6)))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
x1 = Tensor(np.ones([128, 32]), dtype=ms.float32)
x2 = Tensor(np.ones([32, 64]), dtype=ms.float32)
x4 = Tensor(np.ones([64, 128]), dtype=ms.float32)
x5 = Tensor(np.ones([128, 64]), dtype=ms.float32)
x6 = Tensor(np.ones([64, 32]), dtype=ms.float32)
x7 = Tensor(np.ones([32, 32]), dtype=ms.float32)
_executor.compile(net, x1, x2, x4, x5, x6, x7)
# add matmul7
def test_six_matmul_repeated2():
class NetWithLoss(nn.Cell):
def __init__(self, network):
super(NetWithLoss, self).__init__()
self.loss = VirtualLoss()
self.network = network
def construct(self, x1, x2, x4, x5, x6, x7, x8):
predict = self.network(x1, x2, x4, x5, x6, x7, x8)
return self.loss(predict)
class GradWrap(nn.Cell):
def __init__(self, network):
super(GradWrap, self).__init__()
self.network = network
def construct(self, x1, x2, x4, x5, x6, x7, x8):
return C.grad_all(self.network)(x1, x2, x4, x5, x6, x7, x8)
class Net(nn.Cell):
def __init__(self, strategy1, strategy3, strategy4, strategy5, strategy6, strategy7):
super().__init__()
self.matmul1 = P.MatMul().set_strategy(strategy1)
self.matmul3 = P.MatMul().set_strategy(strategy3)
self.matmul4 = P.MatMul().set_strategy(strategy4)
self.matmul5 = P.MatMul().set_strategy(strategy5)
self.matmul6 = P.MatMul().set_strategy(strategy6)
self.matmul7 = P.MatMul().set_strategy(strategy7)
def construct(self, x1, x2, x4, x5, x6, x7, x8):
out = self.matmul1(x1, x2)
out = self.matmul3(out, x4)
out = self.matmul4(out, x5)
out = self.matmul5(out, x6)
out = self.matmul6(out, x7)
out = self.matmul7(out, x8)
return out
set_auto_parallel_context(device_num=512, global_rank=0)
strategy1 = ((8, 1), (1, 1))
strategy3 = ((8, 1), (1, 1))
strategy4 = ((8, 1), (1, 1))
strategy5 = ((8, 1), (1, 1))
strategy6 = ((8, 1), (1, 1))
strategy7 = ((8, 1), (1, 1))
net = GradWrap(NetWithLoss(Net(strategy1, strategy3, strategy4, strategy5, strategy6, strategy7)))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
x1 = Tensor(np.ones([128, 32]), dtype=ms.float32)
x2 = Tensor(np.ones([32, 64]), dtype=ms.float32)
x4 = Tensor(np.ones([64, 128]), dtype=ms.float32)
x5 = Tensor(np.ones([128, 64]), dtype=ms.float32)
x6 = Tensor(np.ones([64, 32]), dtype=ms.float32)
x7 = Tensor(np.ones([32, 32]), dtype=ms.float32)
x8 = Tensor(np.ones([32, 128]), dtype=ms.float32)
_executor.compile(net, x1, x2, x4, x5, x6, x7, x8)
# add scope2
def test_six_matmul_repeated3():
class NetWithLoss(nn.Cell):
def __init__(self, network1, network2):
super(NetWithLoss, self).__init__()
self.loss = VirtualLoss()
self.network = network1
self.network2 = network2
def construct(self, x1, x2, x4, x5, x6, x7, x8, x9, x10):
predict = self.network(x1, x2, x4, x5, x6, x7, x8)
predict = self.network2(predict, x9, x10)
return self.loss(predict)
class GradWrap(nn.Cell):
def __init__(self, network):
super(GradWrap, self).__init__()
self.network = network
def construct(self, x1, x2, x4, x5, x6, x7, x8, x9, x10):
return C.grad_all(self.network)(x1, x2, x4, x5, x6, x7, x8, x9, x10)
class Net(nn.Cell):
def __init__(self, strategy1, strategy3, strategy4, strategy5, strategy6, strategy7):
super().__init__()
self.matmul1 = P.MatMul().set_strategy(strategy1)
self.matmul3 = P.MatMul().set_strategy(strategy3)
self.matmul4 = P.MatMul().set_strategy(strategy4)
self.matmul5 = P.MatMul().set_strategy(strategy5)
self.matmul6 = P.MatMul().set_strategy(strategy6)
self.matmul7 = P.MatMul().set_strategy(strategy7)
def construct(self, x1, x2, x4, x5, x6, x7, x8):
out = self.matmul1(x1, x2)
out = self.matmul3(out, x4)
out = self.matmul4(out, x5)
out = self.matmul5(out, x6)
out = self.matmul6(out, x7)
out = self.matmul7(out, x8)
return out
class Net1(nn.Cell):
def __init__(self, strategy1, strategy2):
super().__init__()
self.matmul1 = P.MatMul().set_strategy(strategy1)
self.matmul2 = P.MatMul().set_strategy(strategy2)
def construct(self, x1, x2, x3):
out = self.matmul1(x1, x2)
out = self.matmul2(out, x3)
return out
set_auto_parallel_context(device_num=512, global_rank=0)
strategy1 = ((8, 1), (1, 1))
strategy3 = ((8, 1), (1, 1))
strategy4 = ((8, 1), (1, 1))
strategy5 = ((8, 1), (1, 1))
strategy6 = ((8, 1), (1, 1))
strategy7 = ((8, 1), (1, 1))
strategy8 = ((8, 1), (1, 1))
strategy9 = ((8, 1), (1, 1))
net1 = Net(strategy1, strategy3, strategy4, strategy5, strategy6, strategy7)
net2 = Net1(strategy8, strategy9)
net = GradWrap(NetWithLoss(net1, net2))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
x1 = Tensor(np.ones([128, 32]), dtype=ms.float32)
x2 = Tensor(np.ones([32, 64]), dtype=ms.float32)
x4 = Tensor(np.ones([64, 128]), dtype=ms.float32)
x5 = Tensor(np.ones([128, 64]), dtype=ms.float32)
x6 = Tensor(np.ones([64, 32]), dtype=ms.float32)
x7 = Tensor(np.ones([32, 32]), dtype=ms.float32)
x8 = Tensor(np.ones([32, 128]), dtype=ms.float32)
x9 = Tensor(np.ones([128, 64]), dtype=ms.float32)
x10 = Tensor(np.ones([64, 64]), dtype=ms.float32)
_executor.compile(net, x1, x2, x4, x5, x6, x7, x8, x9, x10)