# Copyright 2021-2022 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.
# ============================================================================
"""Boost Mode Cell Wrapper."""
from __future__ import absolute_import
import numpy as np
from mindspore.nn.wrap import TrainOneStepCell
import mindspore.context as context
from mindspore.context import ParallelMode
from mindspore.parallel._utils import _get_global_rank, _get_device_num, _get_gradients_mean
from mindspore.communication.management import get_group_size, create_group
from mindspore.nn.cell import Cell
from mindspore.nn import SequentialCell
from mindspore.common import Tensor
from mindspore.common.sparse_tensor import RowTensorInner
from mindspore.common.parameter import Parameter, ParameterTuple
from mindspore.nn.wrap.grad_reducer import DistributedGradReducer
from mindspore.ops.operations.math_ops import NPUGetFloatStatusV2, NPUClearFloatStatusV2
from mindspore.ops import functional as F
from mindspore.ops import composite as C
from mindspore.ops import operations as P
from mindspore.common import dtype as mstype
from mindspore.boost.boost import AutoBoost
from mindspore.boost.grad_freeze import FreezeOpt, freeze_cell
from mindspore.boost.adasum import AdaSum
from mindspore.boost.dim_reduce import DimReduce
from mindspore.boost.grad_accumulation import gradient_accumulation_op, gradient_clear_op
from mindspore.boost.base import _load_local_pca_mat
__all__ = ["BoostTrainOneStepCell", "BoostTrainOneStepWithLossScaleCell"]
_get_delta_weight = C.MultitypeFuncGraph("_get_delta_weight")
@_get_delta_weight.register("Tensor", "Tensor")
def _get_delta_weight_process(new_parameter, old_parameter):
delta_w = old_parameter - new_parameter
return delta_w
_save_weight = C.MultitypeFuncGraph("_save_weight")
@_save_weight.register("Tensor", "Tensor")
def _save_weight_process(new_parameter, old_parameter):
P.Assign()(new_parameter, old_parameter)
return new_parameter
_grad_scale = C.MultitypeFuncGraph("grad_scale")
reciprocal = P.Reciprocal()
@_grad_scale.register("Tensor", "Tensor")
def tensor_grad_scale(scale, grad):
"""grad scale function for tensor"""
return grad * F.cast(reciprocal(scale), F.dtype(grad))
@_grad_scale.register("Tensor", "RowTensor")
def tensor_grad_scale_row_tensor(scale, grad):
"""grad scale function for row tensor"""
return RowTensorInner(grad.indices,
grad.values * F.cast(reciprocal(scale), F.dtype(grad.values)),
grad.dense_shape)
_grad_overflow = C.MultitypeFuncGraph("_grad_overflow")
grad_overflow = P.FloatStatus()
@_grad_overflow.register("Tensor")
def _tensor_grad_overflow(grad):
return grad_overflow(grad)
@_grad_overflow.register("RowTensor")
def _tensor_grad_overflow_row_tensor(grad):
return grad_overflow(grad.values)
class _OutputToFloat16(Cell):
"Wrap cell for amp. Cast network output back to float16"
def __init__(self, op):
super(_OutputToFloat16, self).__init__(auto_prefix=False)
self._op = op
def construct(self, *inputs):
return F.cast(self._op(*inputs), mstype.float16)
[文档]class BoostTrainOneStepCell(TrainOneStepCell):
r"""
Boost Network training package class.
Wraps the network with an optimizer. The resulting Cell is trained with input '\*inputs'.
The backward graph will be created in the construct function to update the parameter, and different
parallel modes are available for training.
Args:
network (Cell): The training network. The network only supports single output.
optimizer (Union[Cell]): Optimizer for updating the weights.
sens (numbers.Number): The scaling number to be filled as the input of backpropagation.
Default: ``None`` , which is ``1.0`` .
Inputs:
- **\*inputs** (Tuple(Tensor)) - Tuple of input tensors with shape :math:`(N, \ldots)`.
Outputs:
Tensor, a tensor means the loss value, the shape of which is usually :math:`()`.
- loss(Tensor): A scalar Tensor.
- overflow(Tensor): A scalar Tensor which type is bool.
- loss scaling value(Tensor): A scalar Tensor.
Raises:
TypeError: If `sens` is not a number.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Examples:
>>> from mindspore import boost
>>> from mindspore import nn
>>> # Define the network structure of LeNet5. Refer to
>>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
>>> net = LeNet5()
>>> loss_fn = nn.SoftmaxCrossEntropyWithLogits()
>>> optim = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> #1) Using the WithLossCell existing provide
>>> loss_net = nn.WithLossCell(net, loss_fn)
>>> train_net = boost.BoostTrainOneStepCell(loss_net, optim)
>>>
>>> #2) Using user-defined WithLossCell
>>> class MyWithLossCell(nn.Cell):
... def __init__(self, backbone, loss_fn):
... super(MyWithLossCell, self).__init__(auto_prefix=False)
... self._backbone = backbone
... self._loss_fn = loss_fn
...
... def construct(self, x, y, label):
... out = self._backbone(x, y)
... return self._loss_fn(out, label)
...
... @property
... def backbone_network(self):
... return self._backbone
...
>>> loss_net = MyWithLossCell(net, loss_fn)
>>> train_net = boost.BoostTrainOneStepCell(loss_net, optim)
"""
def __init__(self, network, optimizer, sens=None):
super(BoostTrainOneStepCell, self).__init__(network, optimizer, sens)
self.hyper_map = C.HyperMap()
self.freeze = isinstance(optimizer, FreezeOpt)
if not self.freeze:
self.weights = self.optimizer.parameters
self.train_strategy = getattr(self.optimizer, 'train_strategy', None)
self.auto_boost = AutoBoost()
self.use_grad_accumulation = self.parallel_mode in (ParallelMode.DATA_PARALLEL, ParallelMode.STAND_ALONE)
self.use_grad_accumulation = self.use_grad_accumulation & \
self.auto_boost.boost_config.get("grad_accumulation", False)
self.max_accumulation_step = 1
if self.use_grad_accumulation:
self.max_accumulation_step = self.auto_boost.grad_accumulation_step
if self.max_accumulation_step <= 1:
self.max_accumulation_step = 1
self.use_grad_accumulation = False
self.accumulation_step = Parameter(Tensor(0, dtype=mstype.int32), name="accumulation_step")
if self.use_grad_accumulation:
self.grad_accumulation = self.weights.clone(prefix="grad_accumulation", init='zeros')
self.enable_dim_reduce = self.check_dim_reduce_enable()
if self.enable_dim_reduce:
self.__init_dim_reduce()
self.freeze_nets = None
self.step = Parameter(Tensor(0, dtype=mstype.int32))
if self.freeze:
if self.reducer_flag:
self.mean = _get_gradients_mean()
self.degree = _get_device_num()
else:
self.mean = None
self.degree = None
self.freeze_nets = freeze_cell(self.reducer_flag, self.network, self.optimizer, self.sens,
self.grad, self.use_grad_accumulation, self.mean, self.degree,
self.max_accumulation_step)
self.enable_adasum = self.check_adasum_enable()
self.sync_tensor = Parameter(Tensor(0, dtype=mstype.int32))
if self.enable_adasum:
self.__init_adasum()
def construct(self, *inputs):
if self.freeze:
loss = self.gradient_freeze_process(*inputs)
else:
if not self.sense_flag:
return self._no_sens_impl(*inputs)
loss = self.network(*inputs)
sens = F.fill(loss.dtype, loss.shape, self.sens)
grads = self.grad(self.network, self.weights)(*inputs, sens)
grads = self.grad_reducer(grads)
if self.use_grad_accumulation:
loss = self.gradient_accumulation_process(loss, grads, sens, *inputs)
else:
if self.enable_dim_reduce:
loss = F.depend(loss, self.dim_reduce(loss, grads, sens, self.weights, self.weights_clone, *inputs))
elif self.enable_adasum:
loss = F.depend(loss, self.adasum_process(loss, grads))
else:
loss = F.depend(loss, self.optimizer(grads))
return loss
[文档] def gradient_freeze_process(self, *inputs):
r"""
Gradient freeze algorithm process.
Args:
inputs (tuple(Tensor)): Tuple of input tensors with shape :math:`(N, \ldots)`.
Returns:
- **loss** (Tensor) - Network loss, tensor with shape :math:`()`.
"""
if self.train_strategy is None:
step = self.step
max_index = len(self.freeze_nets)
else:
step = self.train_strategy[self.step]
max_index = len(self.train_strategy)
loss = self.freeze_nets[step](*inputs)
if self.step + 1 >= max_index:
self.step = 0
else:
self.step += 1
return loss
[文档] def gradient_accumulation_process(self, loss, grads, sens, *inputs):
r"""
Gradient accumulation algorithm process.
Args:
loss (Tensor): Tensor with shape :math:`()`.
grads (tuple(Tensor)): Tuple of gradient tensors.
sens (Tensor): Tensor with shape :math:`()`.
inputs (tuple(Tensor)): Tuple of input tensors with shape :math:`(N, \ldots)`.
Returns:
- **loss** (Tensor) - Network loss, tensor with shape :math:`()`.
"""
loss = F.depend(loss, self.hyper_map(F.partial(gradient_accumulation_op, self.max_accumulation_step),
self.grad_accumulation, grads))
self.accumulation_step += 1
if self.accumulation_step >= self.max_accumulation_step:
if self.enable_dim_reduce:
loss = F.depend(loss, self.dim_reduce(loss, self.grad_accumulation, sens, self.weights,
self.weights_clone, *inputs))
elif self.enable_adasum:
loss = F.depend(loss, self.adasum_process(loss, self.grad_accumulation))
else:
loss = F.depend(loss, self.optimizer(self.grad_accumulation))
self.accumulation_step = 0
if self.accumulation_step == 0:
loss = F.depend(loss, self.hyper_map(F.partial(gradient_clear_op), self.grad_accumulation))
return loss
[文档] def adasum_process(self, loss, grads):
r"""
Adasum algorithm process.
Args:
loss (Tensor): Tensor with shape :math:`()`.
grads (tuple(Tensor)): Tuple of gradient tensors.
Returns:
- **loss** (Tensor) - Network loss, tensor with shape :math:`()`.
"""
loss = F.depend(loss, self.optimizer(grads))
rank_weights = self.weights[self.start[self.server_rank]: self.end[self.server_rank]]
grad_clone = F.depend(self.grad_clone, loss)
delta_w = self.hyper_map(F.partial(_get_delta_weight), rank_weights, grad_clone)
adasum_res = self.adasum(delta_w, rank_weights, grad_clone)
sync_tensor = F.depend(self.sync_tensor, adasum_res)
sync_flag = self.adasum.sync_barrier(sync_tensor)
for i in range(self.device_number):
weight_tuple = self.weights[self.start[i]: self.end[i]]
node_rank = F.depend(weight_tuple, sync_flag)
update_weights = self.adasum.broadcast_list[i](node_rank)
if i == self.server_rank:
self.hyper_map(F.partial(_save_weight), self.grad_clone, update_weights)
else:
self.hyper_map(F.partial(_save_weight), weight_tuple, update_weights)
return loss
[文档] def check_adasum_enable(self):
r"""
Check adasum enable.
Returns:
- **enable_adasum** (bool) - Check whether the Adasum algorithm is enabled.
"""
if not getattr(self.optimizer, "adasum", None) or not self.reducer_flag:
return False
_rank_size = get_group_size()
_device_number = 8
group_number = _rank_size // _device_number
is_enable = bool(group_number > 1 and group_number & (group_number - 1) == 0)
return is_enable
[文档] def check_dim_reduce_enable(self):
r"""
Check dim_reduce enable.
Returns:
- **enable_dim_reduce** (bool) - Check whether the dimensionality reduction second-order training
algorithm is enabled.
"""
if not getattr(self.optimizer, "dim_reduce", None):
return False
return True
def _no_sens_impl(self, *inputs):
"""construct implementation when the 'sens' parameter is passed in."""
loss = self.network(*inputs)
sens = F.fill(loss.dtype, loss.shape, self.sens)
grads = self.grad_no_sens(self.network, self.weights)(*inputs)
grads = self.grad_reducer(grads)
if self.use_grad_accumulation:
loss = self.gradient_accumulation_process(loss, grads, sens, *inputs)
else:
if self.enable_dim_reduce:
loss = F.depend(loss, self.dim_reduce(loss, grads, sens, self.weights, self.weights_clone, *inputs))
elif self.enable_adasum:
loss = F.depend(loss, self.adasum_process(loss, grads))
else:
loss = F.depend(loss, self.optimizer(grads))
def __init_dim_reduce(self):
"""dim reduce algorithm init method."""
local_pca_mat_path = self.auto_boost.local_pca_mat_path
rho = self.auto_boost.rho
gamma = self.auto_boost.gamma
alpha = self.auto_boost.alpha
sigma = self.auto_boost.sigma
_rank = _get_global_rank()
_rank_size = 1 if self.parallel_mode == ParallelMode.STAND_ALONE else get_group_size()
n_components = self.auto_boost.n_components
timeout = self.auto_boost.timeout
pca_mat = _load_local_pca_mat(local_pca_mat_path, timeout)
self.weights_clone = ParameterTuple(self.weights).clone(prefix="weights_clone", init="same")
self.dim_reduce = DimReduce(self.network, self.optimizer, self.weights, pca_mat, n_components, rho, gamma,
alpha, sigma, _rank, _rank_size)
def __init_adasum(self):
"""adasum algorithm init method."""
_rank = _get_global_rank()
_rank_size = get_group_size()
_device_number = self.auto_boost.device_number
self.device_number = _device_number
group_number = _rank_size // _device_number
self.server_rank = _rank % _device_number
parameter_rank_number = len(self.weights) // _device_number
self.start = [x * parameter_rank_number for x in range(_device_number)]
self.end = [(x + 1) * parameter_rank_number for x in range(_device_number)]
self.end[-1] = len(self.weights)
current_weights = self.weights[self.start[self.server_rank]: self.end[self.server_rank]]
self.grad_clone = ParameterTuple(current_weights).clone(prefix="delta_weight")
self.adasum = AdaSum(_rank, _device_number, group_number, self.grad_clone)
self.degree = int(self.degree // group_number)
group_list = [list(range(x * self.degree, (x + 1) * self.degree)) for x in range(group_number)]
current_index = _rank // _device_number
server_group_name = "allreduce_" + str(current_index)
create_group(server_group_name, group_list[current_index])
self.grad_reducer = DistributedGradReducer(self.weights, self.mean, self.degree, group=server_group_name)
[文档]class BoostTrainOneStepWithLossScaleCell(BoostTrainOneStepCell):
r"""
Boost Network training with loss scaling.
This is a training step with loss scaling. It takes a network, an optimizer and possibly a scale update
Cell as args. The loss scale value can be updated in both host side or device side. The
BoostTrainOneStepWithLossScaleCell will be compiled to be graph which takes `*inputs` as input data.
The Tensor type of `scale_sense` is acting as loss scaling value. If you want to update it on host side,
the value must be provided. If the Tensor type of `scale_sense` is not given, the loss scale update logic
must be provide by Cell type of `scale_sense`.
Args:
network (Cell): The training network. The network only supports single output.
optimizer (Cell): Optimizer for updating the weights.
scale_sense (Union[Tensor, Cell]): If this value is Cell type, the loss scaling update logic cell.If this value
is Tensor type, :func:`mindspore.nn.TrainOneStepWithLossScaleCell.set_sense_scale` can be called to update
loss scale factor, Tensor with shape :math:`()` or :math:`(1,)`.
Inputs:
- **\*inputs** (Tuple(Tensor)) - Tuple of input tensors with shape :math:`(N, \ldots)`.
Outputs:
Tuple of 3 Tensor, the loss, overflow flag and current loss scaling value.
- **loss** (Tensor) - Tensor with shape :math:`()`.
- **overflow** (Tensor) - Tensor with shape :math:`()`, type is bool.
- **loss scaling value** (Tensor) - Tensor with shape :math:`()`
Raises:
TypeError: If `scale_sense` is neither Cell nor Tensor.
ValueError: If shape of `scale_sense` is neither :math:`(1,)` nor :math:`()`.
Supported Platforms:
``Ascend`` ``GPU``
Examples:
>>> import numpy as np
>>> from mindspore import Tensor, Parameter, nn
>>> import mindspore.ops as ops
>>> from mindspore.nn import WithLossCell
>>> from mindspore import dtype as mstype
>>> from mindspore import boost
>>>
>>> class Net(nn.Cell):
... def __init__(self, in_features, out_features):
... super(Net, self).__init__()
... self.weight = Parameter(Tensor(np.ones([in_features, out_features]).astype(np.float32)),
... name='weight')
... self.matmul = ops.MatMul()
...
... def construct(self, x):
... output = self.matmul(x, self.weight)
... return output
...
>>> size, in_features, out_features = 16, 16, 10
>>> #1) when the type of scale_sense is Cell:
>>> net = Net(in_features, out_features)
>>> loss = nn.MSELoss()
>>> optimizer = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> net_with_loss = WithLossCell(net, loss)
>>> manager = nn.DynamicLossScaleUpdateCell(loss_scale_value=2**12, scale_factor=2, scale_window=1000)
>>> train_network = boost.BoostTrainOneStepWithLossScaleCell(net_with_loss, optimizer, scale_sense=manager)
>>> input = Tensor(np.ones([out_features, in_features]), mstype.float32)
>>> labels = Tensor(np.ones([out_features,]), mstype.float32)
>>> output = train_network(input, labels)
>>>
>>> #2) when the type of scale_sense is Tensor:
>>> net = Net(in_features, out_features)
>>> loss = nn.MSELoss()
>>> optimizer = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> net_with_loss = WithLossCell(net, loss)
>>> inputs = Tensor(np.ones([size, in_features]).astype(np.float32))
>>> label = Tensor(np.zeros([size, out_features]).astype(np.float32))
>>> scaling_sens = Tensor(np.full((1), np.finfo(np.float32).max), dtype=mstype.float32)
>>> train_network = boost.BoostTrainOneStepWithLossScaleCell(net_with_loss, optimizer, scale_sense=scaling_sens)
>>> output = train_network(inputs, label)
"""
def __init__(self, network, optimizer, scale_sense):
super(BoostTrainOneStepWithLossScaleCell, self).__init__(network, optimizer, sens=None)
self.base = Tensor(1, mstype.float32)
self.reduce_sum = P.ReduceSum(keep_dims=False)
self.less_equal = P.LessEqual()
self.allreduce = P.AllReduce()
self.is_distributed = (self.parallel_mode != ParallelMode.STAND_ALONE)
self.gpu_target = (context.get_context("device_target") == "GPU")
self.loss_scaling_manager = None
self.base0 = Tensor(0, mstype.int32)
self.reduce_all = P.ReduceAll(keep_dims=False)
self.logic_not = P.LogicalNot()
self.equal = P.Equal()
if self.auto_boost.boost_config.get("loss_scale_group", False):
self.enable_enhanced_amp = True
if not isinstance(scale_sense, Cell) or not hasattr(scale_sense, "set_loss_scale_status"):
raise TypeError("The scale_sense must be enhanced amp Cell, bug got {}".format(type(scale_sense)))
self.loss_scaling_manager = scale_sense
self.loss_scale_groups = scale_sense.loss_scale_groups
self._init_enhanced_amp()
self._do_keep_mix_fp32(self.network)
else:
self.enable_enhanced_amp = False
if isinstance(scale_sense, Cell):
self.loss_scaling_manager = scale_sense
self.scale_sense = Parameter(Tensor(scale_sense.get_loss_scale(), dtype=mstype.float32),
name="scale_sense")
elif isinstance(scale_sense, Tensor):
if scale_sense.shape == (1,) or scale_sense.shape == ():
self.scale_sense = Parameter(scale_sense, name='scale_sense')
else:
raise ValueError("The shape of scale_sense must be (1,) or (), \
but got {}".format(scale_sense.shape))
else:
raise TypeError("The scale_sense must be Cell or Tensor, but got {}".format(type(scale_sense)))
def construct(self, *inputs):
weights = self.weights
loss = self.network(*inputs)
if self.enable_enhanced_amp:
scaling_sens = F.fill(loss.dtype, loss.shape, 1)
grads = self.grad(self.network, weights)(*inputs, scaling_sens)
grads = self.grad_reducer(grads)
cond, scaling_sens = self._enhanced_amp_process_overflow_status(grads)
else:
scaling_sens = self.scale_sense
status, scaling_sens = self._start_overflow_check(loss, scaling_sens)
scaling_sens_filled = C.ones_like(loss) * F.cast(scaling_sens, F.dtype(loss))
grads = self.grad(self.network, weights)(*inputs, scaling_sens_filled)
grads = self.hyper_map(F.partial(_grad_scale, scaling_sens), grads)
grads = self.grad_reducer(grads)
# get the overflow buffer
cond = self._get_overflow_status(status, grads)
overflow = self._process_loss_scale(cond)
# if there is no overflow, do optimize
if not overflow:
loss = self.__multi_update(loss, grads, scaling_sens_filled, *inputs)
return loss, cond, scaling_sens
def __multi_update(self, loss, grads, scaling_sens_filled, *inputs):
"""enable multi-algorithm's process"""
if self.use_grad_accumulation:
loss = self.gradient_accumulation_process(loss, grads, scaling_sens_filled, *inputs)
else:
if self.enable_dim_reduce:
loss = F.depend(loss, self.dim_reduce(loss, grads, scaling_sens_filled, self.weights,
self.weights_clone, *inputs))
elif self.enable_adasum:
loss = F.depend(loss, self.adasum_process(loss, grads))
else:
loss = F.depend(loss, self.optimizer(grads))
return loss
def _get_dynamic_overflow_status(self, param):
"""
Judge whether the current network overflows.
Inputs:
- **param** (Tensor) - Whether the overflow occurs or not.
Outputs:
bool, overflow value.
float, update ratio.
"""
flag_sum = self.equal(self.base0, param)
if self.reducer_flag:
flag_reduce = self.allreduce(flag_sum)
overflow = self.logic_not(self.reduce_all(flag_reduce))
else:
overflow = self.logic_not(self.reduce_all(flag_sum))
if overflow:
update_ratio = self.reduce_ratio
else:
update_ratio = self.growth_ratio
return overflow, update_ratio
def _enhanced_amp_process_overflow_status(self, grads):
"""
Enhanced hybrid precision update loss scale and update weights.
Inputs:
- **grads** (Tuple(Tensor)) - Tuple of gradients.
Outputs:
bool, overflow value.
float, loss scale value.
"""
overflow_global_flag = Tensor(0, mstype.int32)
layer = 0
loss_scale_temp = ()
for param in self.overflow_status_list:
overflow, update_ratio = self._get_dynamic_overflow_status(param)
if overflow:
overflow_global_flag += 1
new_loss_scale_value = self.loss_scaling_manager.update_loss_scale_status(layer, update_ratio)
loss_scale_temp += (new_loss_scale_value,) * self.optimizer_loss_scale[layer]
layer += 1
if P.Less()(overflow_global_flag, self.base):
grads = self.hyper_map(F.partial(_grad_scale), loss_scale_temp, grads)
overflow_global_flag = F.depend(overflow_global_flag, self.optimizer(grads))
return overflow_global_flag, loss_scale_temp[0]
def _set_sense_scale(self, sens):
"""
If the user has set the sens in the training process and wants to reassign the value, he can call
this function again to make modification, and sens needs to be of type Tensor.
Inputs:
- **sens** (Tensor) - The new sense whose shape and type are the same with original `scale_sense`.
"""
if self.scale_sense and isinstance(sens, Tensor):
self.scale_sense.set_data(sens)
else:
raise TypeError("The input type must be Tensor, but got {}".format(type(sens)))
def _start_overflow_check(self, pre_cond, compute_input):
"""
Start floating-point overflow detection. Create and clear the overflow detection state.
Specify the argument 'pre_cond' and 'compute_input' to make sure overflow status is cleared at the right time.
Taking this situation as an example, we need to execute state clearing after loss calculation and then detect
overflow in the process of gradient calculation. In this case, pre_cond should be the output of the loss
function, and compute_input should be the input of gradients-computing function.
Inputs:
- **pre_cond** (Tensor) - A precondition for starting overflow detection. It determines the executing order
of overflow state clearing and prior processions. It makes sure that the function 'start_overflow'
clears status after finishing the process of precondition.
- **compute_input** (object) - The input of subsequent process. Overflow detection should be performed on a
certain computation. Set `compute_input` as the input of the computation, to ensure overflow status is
cleared before executing the computation.
Outputs:
Tuple[object, object], the first value is False for GPU backend, while it is an instance of
NPUAllocFloatStatus for other backend. The status is used to detect overflow during overflow detection.
The second value is the same as the input of `compute_input`, but contains some information about the
execution order.
"""
status = Tensor([0] * 8, mstype.int32)
if not self.gpu_target:
status = F.depend(status, pre_cond)
# clear overflow buffer
clear_status = NPUClearFloatStatusV2()(status)
compute_input = F.depend(compute_input, clear_status)
return status, compute_input
def _get_overflow_status(self, status, compute_output):
"""
Get floating-point overflow status.
Get overflow results after executing the target process for overflow detection.
Inputs:
- **status** (object) - A status instance used to detect the overflow.
- **compute_output** - Overflow detection should be performed on a certain computation. Set `compute_output`
as the output of the computation, to ensure overflow status is acquired before executing the
computation.
Outputs:
bool, whether the overflow occurs or not.
"""
if not self.gpu_target:
status = F.depend(status, compute_output)
get_status = NPUGetFloatStatusV2()(status)
if self.is_distributed:
# sum overflow flag over devices
flag_reduce = self.allreduce(get_status)
# get_status not equal to [0]*8 means overflow
flag = self.equal(self.base0, flag_reduce)
status = F.depend(status, flag)
# distributed needs to skip allreduce to avoid its overflow affecting the next step
clear_status = NPUClearFloatStatusV2()(status)
flag = F.depend(flag, clear_status)
overall_finite = self.reduce_all(flag)
else:
status = F.depend(status, get_status)
clear_status = NPUClearFloatStatusV2()(status)
get_status = F.depend(get_status, clear_status)
flag = self.equal(self.base0, get_status)
overall_finite = self.reduce_all(flag)
overflow = self.logic_not(overall_finite)
else:
flag_sum = self.hyper_map(F.partial(_grad_overflow), compute_output)
flag_sum = P.AddN()(flag_sum)
# convert flag_sum to scalar
flag_sum = P.Reshape()(flag_sum, (()))
if self.is_distributed:
# sum overflow flag over devices
flag_reduce = self.allreduce(flag_sum)
overflow = self.less_equal(self.base, flag_reduce)
else:
overflow = self.less_equal(self.base, flag_sum)
return overflow
def _process_loss_scale(self, overflow):
"""
Calculate loss scale according to the overflow.
Inputs:
- **overflow** (bool) - Whether the overflow occurs or not.
Outputs:
bool, overflow value.
"""
if self.loss_scaling_manager is not None:
return self.loss_scaling_manager(self.scale_sense, overflow)
return overflow
def _init_enhanced_amp(self):
"""
Init enhanced hybrid precision.
"""
self.params_len = len(self.optimizer.params)
self.parent = list(range(self.params_len))
self.layer_rank = [0 for _ in range(self.params_len)]
index = 0
loss_scale_number = len(self.loss_scale_groups)
for loss_scale_group in self.loss_scale_groups:
for i, _ in enumerate(loss_scale_group):
if i == 0:
index += 1
continue
self._union(index - 1, index)
index += 1
parent_set = list(set(self.parent))
self.optimizer_loss_scale = [self.parent.count(x) for x in parent_set]
self.reduce_ratio = Tensor(1.0 / (2 ** 0.5), mstype.float32)
self.growth_ratio = Tensor(2 ** (1.0 / 1000.0), mstype.float32)
self.overflow_status_list = ParameterTuple(Parameter(Tensor(np.zeros(shape=[8]), mstype.int32),
name='mix_layer_status_{}'.format(x), requires_grad=False)
for x in range(loss_scale_number))
self.loss_scaling_manager.set_loss_scale_status(loss_scale_number, self.loss_scaling_manager.get_loss_scale())
def _get_root(self, i):
"""
Get parent id.
Args:
i (int): the current parameters's id.
Returns:
Number, the parent id.
"""
if self.parent[i] != self.parent[self.parent[i]]:
self.parent[i] = self.get_root(self.parent[i])
return self.parent[i]
def _union(self, i, j):
"""
Aggregate parameters of the same category.
Args:
i (int): the last parameters's id.
j (int): the current parameters's id.
"""
i_root = self._get_root(i)
j_root = self._get_root(j)
if self.layer_rank[i_root] == self.layer_rank[j_root]:
self.parent[j_root] = i_root
self.layer_rank[i_root] += 1
elif self.layer_rank[i_root] > self.layer_rank[j_root]:
self.parent[j_root] = i_root
else:
self.parent[i_root] = j_root
def _do_keep_mix_fp32(self, network):
"""
Keep enhanced amp cell of type float32.
Args:
network (Cell): The training network.
"""
cells = network.name_cells()
change = False
for name in cells:
subcell = cells[name]
if subcell == network:
continue
if "GroupLossScaleManager" in subcell.cls_name:
network._cells[name] = _OutputToFloat16(subcell.to_float(mstype.float32)) # pylint: disable=W0212
change = True
else:
self._do_keep_mix_fp32(subcell)
if isinstance(network, SequentialCell) and change:
network.cell_list = list(network.cells())