Applying a Gradient Accumulation Algorithm
Linux GPU Model Optimization Intermediate Expert
Overview
This tutorial describes the gradient accumulation training methods to solve the problem that some large-scale networks cannot train large batch_size due to insufficient memory.
In a traditional training method, after a loss and a gradient are computed each time, a parameter is directly updated by using the obtained gradient.
Compared to the traditional training method, mini-batch is introduced to the gradient accumulation. The loss and gradient are computed for each mini-batch data, but the model parameters are not updated immediately. Instead, the obtained gradients are accumulated first, and then after a specified number (N) of mini-batches, the accumulated gradient is used to update the network parameters. Before the next training, the accumulated gradients are cleared and re-accumulated. The ultimate objective is to achieve the same effect as training with N x Mini-batch data.
This tutorial describes how to implement gradient accumulation training in standalone mode and parallel mode, respectively.
Standalone Mode
In standalone mode, the training process consists of three parts: forward and backward training, parameter update, and accumulated gradient clearance. MNIST is used as an example dataset. To customize a simple model to implement gradient accumulation, perform the following steps:
Download the main training sample code: https://gitee.com/mindspore/docs/tree/r1.3/docs/sample_code/gradient_accumulation
Importing Library Files
The following are the required public modules and MindSpore modules and library files.
import argparse
import os
from collections.abc import Iterable
import mindspore.nn as nn
from mindspore import ParameterTuple
from mindspore import context, DatasetHelper, save_checkpoint
from mindspore.nn import Cell
import mindspore.ops as ops
from model_zoo.official.cv.lenet.src.dataset import create_dataset
from model_zoo.official.cv.lenet.src.lenet import LeNet5
Loading the Dataset
Use the MnistDataset API provided by dataset of MindSpore to load the MNIST dataset. The code is imported from dataset.py in the lenet directory of model_zoo.
Defining the Network
LeNet is used as an example network. You can also use other networks, such as ResNet-50 and BERT. The code is imported from lenet.py in the lenet directory of model_zoo.
Defining the Training Process
The training process consists of three parts: forward and backward training, parameter update, and accumulated gradient clearance.
TrainForwardBackwardcalculates the loss and gradient, and uses grad_sum to implement gradient accumulation.TrainOptimupdates parameters.TrainClearclears the gradient accumulation variable grad_sum.
_sum_op = ops.MultitypeFuncGraph("grad_sum_op")
_clear_op = ops.MultitypeFuncGraph("clear_op")
@_sum_op.register("Tensor", "Tensor")
def _cumulative_grad(grad_sum, grad):
"""Apply grad sum to cumulative gradient."""
add = ops.AssignAdd()
return add(grad_sum, grad)
@_clear_op.register("Tensor", "Tensor")
def _clear_grad_sum(grad_sum, zero):
"""Apply zero to clear grad_sum."""
success = True
success = ops.depend(success, ops.assign(grad_sum, zero))
return success
class TrainForwardBackward(Cell):
def __init__(self, network, optimizer, grad_sum, sens=1.0):
super(TrainForwardBackward, self).__init__(auto_prefix=False)
self.network = network
self.network.set_grad()
self.network.add_flags(defer_inline=True)
self.weights = ParameterTuple(network.trainable_params())
self.optimizer = optimizer
self.grad_sum = grad_sum
self.grad = ops.GradOperation(get_by_list=True, sens_param=True)
self.sens = sens
self.hyper_map = ops.HyperMap()
def construct(self, *inputs):
weights = self.weights
loss = self.network(*inputs)
sens = ops.Fill()(ops.DType()(loss), ops.Shape()(loss), self.sens)
grads = self.grad(self.network, weights)(*inputs, sens)
return ops.depend(loss, self.hyper_map(ops.partial(_sum_op), self.grad_sum, grads))
class TrainOptim(Cell):
def __init__(self, optimizer, grad_sum):
super(TrainOptim, self).__init__(auto_prefix=False)
self.optimizer = optimizer
self.grad_sum = grad_sum
def construct(self):
return self.optimizer(self.grad_sum)
class TrainClear(Cell):
def __init__(self, grad_sum, zeros):
super(TrainClear, self).__init__(auto_prefix=False)
self.grad_sum = grad_sum
self.zeros = zeros
self.hyper_map = ops.HyperMap()
def construct(self):
success = self.hyper_map(ops.partial(_clear_op), self.grad_sum, self.zeros)
return success
Defining the Training Model
Each mini-batch computes the loss and gradient through forward and backward training, and uses mini_steps to control the accumulated times before each parameter update. After the number of accumulation times is reached, the parameter is updated and the accumulated gradient variable is cleared.
class GradientAccumulation:
def __init__(self, network, loss_fn, optimizer):
self._network = network
self._loss_fn = loss_fn
self._optimizer = optimizer
params = self._optimizer.parameters
self._grad_sum = params.clone(prefix="grad_sum", init='zeros')
self._zeros = params.clone(prefix="zeros", init='zeros')
self._train_forward_backward = self._build_train_forward_backward_network()
self._train_optim = self._build_train_optim()
self._train_clear = self._build_train_clear()
@staticmethod
def _transform_callbacks(callbacks):
"""Transform callback to a list."""
if callbacks is None:
return []
if isinstance(callbacks, Iterable):
return list(callbacks)
return [callbacks]
def _build_train_forward_backward_network(self):
"""Build forward and backward network"""
network = self._network
network = nn.WithLossCell(network, self._loss_fn)
loss_scale = 1.0
network = TrainForwardBackward(network, self._optimizer, self._grad_sum, loss_scale).set_train()
return network
def _build_train_optim(self):
"""Build optimizer network"""
network = TrainOptim(self._optimizer, self._grad_sum).set_train()
return network
def _build_train_clear(self):
"""Build clear network"""
network = TrainClear(self._grad_sum, self._zeros).set_train()
return network
def train_process(self, epoch, train_dataset, mini_steps=None):
"""
Training process. The data would be passed to network directly.
"""
dataset_helper = DatasetHelper(train_dataset, dataset_sink_mode=False, epoch_num=epoch)
for i in range(epoch):
step = 0
for k, next_element in enumerate(dataset_helper):
loss = self._train_forward_backward(*next_element)
if (k + 1) % mini_steps == 0:
step += 1
print("epoch:", i + 1, "step:", step, "loss is ", loss)
self._train_optim()
self._train_clear()
train_dataset.reset()
save_checkpoint(self._train_forward_backward, "gradient_accumulation.ckpt", )
Training and Saving the Model
Call the network, optimizer, and loss function, and then customize the train_process API of GradientAccumulation to train the model.
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='MindSpore Grad Cumulative Example')
parser.add_argument('--device_target', type=str, default="GPU", choices=['GPU'],
help='device where the code will be implemented (default: GPU)')
parser.add_argument('--data_path', type=str, default="./Data",
help='path where the dataset is saved')
args = parser.parse_args()
context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
ds_train = create_dataset(os.path.join(args.data_path, "train"), 32)
net = LeNet5(10)
net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
net_opt = nn.Momentum(net.trainable_params(), 0.01, 0.9)
model = GradientAccumulation(net, net_loss, net_opt)
print("============== Starting Training ==============")
model.train_process(10, ds_train, mini_steps=4)
Experiment Result
After 10 epochs, the accuracy on the test set is about 96.31%.
Start training.
Run the training code and view the running result.
python train.py --data_path=./MNIST_Data
The output is as follows. You can see that the loss value decreases with the training.
epoch: 1 step: 27 loss is 0.3660637 epoch: 1 step: 28 loss is 0.25238192 ... epoch: 3 step: 2 loss is 0.12296932 epoch: 3 step: 3 loss is 0.15799297 ... epoch: 10 step: 448 loss is 0.06443884 epoch: 10 step: 449 loss is 0.0067842817
Check the saved checkpoint files.
The checkpoint file
gradient_accumulation.ckpt, that is, the model file, is saved during training.
Validate the model.
Use the saved checkpoint file to load the validation dataset through eval.py in the lenet directory of model_zoo.
python eval.py --data_path=./MNIST_Data --ckpt_path=./gradient_accumulation.ckpt --device_target=GPU
The output is as follows. The accuracy of the validation dataset is about 96.31%, which is the same as the result when the value of batch_size is 32.
============== Starting Testing ==============
============== {'Accuracy': 0.9631730769230769} ==============
Parallel Mode
If gradient accumulation is used in SEMI_AUTO_PARALLEL and AUTO_PARALLEL modes, the accumulation steps and update steps are delivered as two graphs and executed alternately. In an accumulation step graph, only the forward and backward operations and gradient accumulation are performed. In an update step graph, the forward and backward operations and parameter updates are performed. The example in Parallel Distributed Training is used to describe the procedure.
Download the main training sample code: https://gitee.com/mindspore/docs/tree/r1.3/docs/sample_code/distributed_training
Defining the Parallel Training Process
Generally, after the forward network is defined, TrainOneStepCell is used to associate the forward and backward networks with the optimizer. However, two different situations, accumulation and update, exist during gradient accumulation. We need to make some modifications based on the original class definition. The sample code is as follows:
import numpy as np
import mindspore.common.dtype as mstype
from mindspore import ops, context, Tensor, Parameter
from mindspore.nn import TrainOneStepCell
from mindspore.common.initializer import initializer
zeroslike = ops.ZerosLike()
reset_accu_grads = ops.MultitypeFuncGraph("reset_accu_grads")
@reset_accu_grads.register("Tensor")
def _reset_accu_grads(accu_grad):
succ = True
return ops.depend(succ, ops.assign(accu_grad, zeroslike(accu_grad)))
cast = ops.Cast()
update_accu_grads = ops.MultitypeFuncGraph("update_accu_grads")
@update_accu_grads.register("Tensor", "Tensor")
def _update_accu_grads(accu_grad, grad):
succ = True
return ops.depend(succ, ops.assign_add(accu_grad, cast(grad, mstype.float32)))
class TrainAccuStepsCell(TrainOneStepCell):
def __init__(self, network, optimizer, sens=1.0):
super(TrainAccuStepsCell, self).__init__(network, optimizer, sens)
self.accumulation = False
self.accumulation_steps = context.get_auto_parallel_context("grad_accumulation_step")
self.accu_grads = self.weights.clone(prefix="accu_grads", init='zeros')
self.hyper_map = ops.HyperMap()
def construct(self, *inputs):
"""Defines the computation performed."""
weights = self.weights
loss = self.network(*inputs)
sens = ops.Fill()(ops.DType()(loss), ops.Shape()(loss), self.sens)
grads = self.grad(self.network, weights)(*inputs, sens)
if self.accumulation and self.accumulation_steps > 1:
accu_succ = self.hyper_map(update_accu_grads, self.accu_grads, grads)
loss = ops.depend(loss, accu_succ)
if self.accumulation:
succ = False
else:
grads = self.grad_reducer(grads)
accu_grads = ops.depend(self.accu_grads, grads)
accu_succ = self.hyper_map(reset_accu_grads, accu_grads)
loss = ops.depend(loss, accu_succ)
succ = self.optimizer(grads)
return ops.depend(loss, succ)
On the basis of TrainOneStepCell, definitions of the accumulation flag accumulation and the accumulation gradient parameter accu_grads are added to distinguish the training process and save the accumulation gradient value, respectively. In an accumulation step graph, if accumulation is set to True, only the forward and backward operations are performed and gradients are accumulated to the parameter accu_grads. In an update step graph, if accumulation is set to False, the forward and backward operations and parameter updates are performed.
The gradient accumulation in parallel mode needs to be implemented based on the internal graph optimization of the framework. Therefore,
accumulationandaccu_gradsdefined on the network are specific characters and cannot be modified.
In the dynamic loss scale scenario, in addition to the gradient, the overflow flag status also needs to be accumulated. The code can be modified based on TrainOneStepWithLossScaleCell. The implementation code is as follows:
import numpy as np
import mindspore.common.dtype as mstype
from mindspore import ops, context, Tensor, Parameter
from mindspore.nn import TrainOneStepWithLossScaleCell
from mindspore.nn.wrap.loss_scale import _grad_scale
from mindspore.common.initializer import initializer
zeroslike = ops.ZerosLike()
reset_accu_grads = ops.MultitypeFuncGraph("reset_accu_grads")
@reset_accu_grads.register("Tensor")
def _reset_accu_grads(accu_grad):
succ = True
return ops.depend(succ, ops.assign(accu_grad, zeroslike(accu_grad)))
cast = ops.Cast()
update_accu_grads = ops.MultitypeFuncGraph("update_accu_grads")
@update_accu_grads.register("Tensor", "Tensor")
def _update_accu_grads(accu_grad, grad):
succ = True
return ops.depend(succ, ops.assign_add(accu_grad, cast(grad, mstype.float32)))
class TrainAccuStepsWithLossScaleCell(TrainOneStepWithLossScaleCell):
def __init__(self, network, optimizer, scale_sense):
super(TrainAccuStepsWithLossScaleCell, self).__init__(network, optimizer, scale_sense)
self.accumulation = False
self.accumulation_steps = context.get_auto_parallel_context("grad_accumulation_step")
self.one = Tensor(np.array([1]).astype(np.int32))
self.zero = Tensor(np.array([0]).astype(np.int32))
self.accu_grads = self.weights.clone(prefix="accu_grads", init='zeros')
self.accu_overflow = Parameter(initializer(0, [1], mstype.int32))
self.accu_loss = Parameter(initializer(0, [1], mstype.float32))
self.cast = ops.Cast()
self.logical_or = ops.LogicalOr()
self.not_equal = ops.NotEqual()
self.select = ops.Select()
self.reshape = ops.Reshape()
def construct(self, *inputs):
"""Defines the computation performed."""
weights = self.weights
loss = self.network(*inputs)
scaling_sens = self.scale_sense
status, scaling_sens = self.start_overflow_check(loss, scaling_sens)
scaling_sens_filled = ops.ones_like(loss) * ops.cast(scaling_sens, ops.dtype(loss))
grads = self.grad(self.network, weights)(*inputs, scaling_sens_filled)
# accumulate gradients
if self.accumulation and self.accumulation_steps > 1:
accu_succ = self.hyper_map(update_accu_grads, self.accu_grads, grads)
loss = ops.depend(loss, accu_succ)
overflow = self.get_overflow_status(status, grads)
overflow = self.logical_or(self.not_equal(self.accu_overflow, self.zero), overflow)
accu_overflow = self.select(overflow, self.one, self.zero)
if self.accumulation:
succ = False
self.accu_overflow = accu_overflow
else:
self.accu_overflow = self.zero
# apply grad reducer on grads
grads = self.grad_reducer(grads)
grads = self.hyper_map(ops.partial(_grad_scale, scaling_sens), grads)
accu_overflow = self.allreduce(accu_overflow)
overflow = self.less_equal(self.base, accu_overflow)
accu_grads = ops.depend(self.accu_grads, grads)
accu_succ = self.hyper_map(reset_accu_grads, accu_grads)
overflow = ops.depend(overflow, accu_succ)
overflow = self.reshape(overflow, (()))
overflow = self.process_loss_scale(overflow)
if overflow:
succ = False
else:
succ = self.optimizer(grads)
ret = (loss, overflow, scaling_sens)
return ops.depend(ret, succ)
accu_overflow is a parameter used to store the accumulation overflow flag status.
Defining the Parallel Training Model
The network encapsulated by cell_wrapper contains the forward and backward operations and optimizer implementation. You need to connect the dataset to the network and execute the two graphs alternately. The preceding functions are implemented based on the Model API in the framework.
import math
from mindspore.train.callback import RunContext
from mindspore import context
from mindspore.context import ParallelMode
from mindspore import Model, connect_network_with_dataset
from mindspore.common.dtype import pytype_to_dtype
from mindspore._c_expression import init_exec_dataset
from mindspore.train.train_thor.dataset_helper import DatasetHelper
def _convert_type(types):
"""
Convert from numpy type to tensor type.
Args:
types (list): Numpy type list of element in dataset.
Returns:
list, list of element in dataset.
"""
ms_types = []
for np_type in types:
ms_type = pytype_to_dtype(np_type)
ms_types.append(ms_type)
return ms_types
def _get_types_and_shapes(dataset):
"""Get dataset types and shapes."""
dataset_types = _convert_type(dataset.output_types())
dataset_shapes = dataset.output_shapes()
return dataset_types, dataset_shapes
def _exec_datagraph(exec_dataset, dataset_size, phase='dataset'):
"""Initialize and execute the dataset graph."""
batch_size = exec_dataset.get_batch_size()
input_indexs = exec_dataset.input_indexs
# transform data format
dataset_types, dataset_shapes = _get_types_and_shapes(exec_dataset)
init_exec_dataset(exec_dataset.__transfer_dataset__.queue_name,
dataset_size,
batch_size,
dataset_types,
dataset_shapes,
input_indexs,
phase=phase,
need_run=False)
class Model_ACCU(Model):
def __init__(self, network, loss_fn=None, optimizer=None, metrics=None, eval_network=None,
eval_indexes=None, amp_level="O0", **kwargs):
super(Model_ACCU, self).__init__(network, loss_fn, optimizer, metrics, eval_network,
eval_indexes, amp_level, **kwargs)
self._frequency = context.get_auto_parallel_context("grad_accumulation_step")
self._train_network = self._build_train_network()
def _exec_preprocess(self, network, is_train, phase, dataset, dataset_sink_mode, sink_size=-1,
epoch_num=1, iter_first_order=1):
"""Initializes dataset."""
if dataset_sink_mode and not is_train:
dataset.__loop_size__ = 1
dataset_helper = DatasetHelper(dataset, dataset_sink_mode, sink_size, epoch_num, iter_first_order)
if dataset_sink_mode and context.get_context("device_target") != "GPU":
network = connect_network_with_dataset(network, dataset_helper)
network.set_train(is_train)
network.phase = phase
if self._parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
network.set_auto_parallel()
return dataset_helper, network
def _train_dataset_sink_process(self, epoch, train_dataset, list_callback=None, cb_params=None, sink_size=-1):
"""
Training process. The data would be passed to network through dataset channel.
Args:
epoch (int): Total number of iterations on the data.
train_dataset (Dataset): A training dataset iterator. If there is no
loss_fn, a tuple with multiple data (data1, data2, data3, ...) should be
returned and passed to the network. Otherwise, a tuple (data, label) should
be returned. The data and label would be passed to the network and loss
function respectively.
list_callback (Callback): Executor of callback list. Default: None.
cb_params (_InternalCallbackParam): Callback parameters. Default: None.
sink_size (int): Control the amount of data in each sink. Default: -1.
"""
if sink_size == -1:
epoch_num = epoch
else:
epoch_num = math.ceil(epoch * sink_size / train_dataset.get_dataset_size())
iter_first_order = 1
iter_second_order = self._frequency - 1
train_dataset.__loop_size__ = iter_second_order
dataset_helper, train_network = self._exec_preprocess(self._train_network,
is_train=True,
phase='train',
dataset=train_dataset,
dataset_sink_mode=True,
sink_size=sink_size,
epoch_num=epoch_num,
iter_first_order=iter_first_order)
self._train_network = train_network
cb_params.train_network = self._train_network
cb_params.cur_step_num = 0
run_context = RunContext(cb_params)
list_callback.begin(run_context)
# used to stop training for early stop, such as stopAtTIme or stopATStep
should_stop = False
switch_branch_one = True
index_first_order = 0
train_network_init_flag = True
has_do_dataset_init = False
for i in range(epoch):
cb_params.cur_epoch_num = i + 1
list_callback.epoch_begin(run_context)
# for data sink dataset_helper only iter once, other wise iter epoch_size times.
for inputs in dataset_helper:
list_callback.step_begin(run_context)
if switch_branch_one:
cb_params.cur_step_num += iter_second_order
if train_network_init_flag:
self._train_network.add_flags_recursive(accumulation=True)
self._train_network.phase = 'train0'
else:
cb_params.cur_step_num += iter_first_order
if train_network_init_flag:
self._train_network.add_flags_recursive(accumulation=False)
train_network_init_flag = False
self._train_network.phase = 'train1'
if not has_do_dataset_init:
_exec_datagraph(train_dataset, iter_first_order, phase='train1_dataset')
has_do_dataset_init = True
switch_branch_one = not switch_branch_one
outputs = self._train_network(*inputs)
cb_params.net_outputs = outputs
list_callback.step_end(run_context)
list_callback.epoch_end(run_context)
should_stop = should_stop or run_context.get_stop_requested()
if should_stop:
break
dataset_helper.stop_send()
list_callback.end(run_context)
In the sample code, the subclass Model_ACCU rewrites the _exec_preprocess dataset encapsulation and the _train_dataset_sink_process training offload method of the base class, and delivers the data subgraph and training graph of the accumulation step graph (accumulation=True) and the update step graph (accumulation=False), respectively. The training process is alternately executed. The number of offloaded steps of the accumulation step graph is the value of grad_accumulation_step minus 1 and that of the update step graph is 1.
Training the Model
After the preceding definition is complete, you can use the training API to complete model training. Configure the grad_accumulation_step parameter in context.set_auto_parallel_context to enable gradient accumulation. Then, use the modified cell_wrapper to encapsulate the network structure and transfer it to the Model_ACCU to initialize the model.
context.set_auto_parallel_context(parallel_mode=ParallelMode.AUTO_PARALLEL, gradients_mean=True, grad_accumulation_step=6)
loss_cb = LossMonitor()
data_path = os.getenv('DATA_PATH')
batch_size = 32
dataset = create_dataset(data_path, batch_size=batch_size)
num_classes = 10
net = resnet50(batch_size, num_classes)
loss = SoftmaxCrossEntropyExpand(sparse=True)
opt = Momentum(filter(lambda x: x.requires_grad, net.get_parameters()), 0.01, 0.9)
net_with_loss = nn.WithLossCell(net, loss)
net_with_loss = VirtualDatasetCell(net_with_loss)
wrap_net = TrainAccuStepsCell(net_with_loss, opt)
model = Model_ACCU(wrap_net)
model.train(epoch_size, dataset, callbacks=[loss_cb], dataset_sink_mode=True)
The following information can be found in the logs:
epoch: 1 step: 234, loss is 1.7588712
epoch: 2 step: 234, loss is 1.7275971
epoch: 3 step: 234, loss is 1.5423206
epoch: 4 step: 234, loss is 1.2762429
epoch: 5 step: 234, loss is 1.0915408