# 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.
# ============================================================================
"""boost"""
from __future__ import absolute_import
import threading
from mindspore.nn.optim import SGD
from mindspore.boost.less_batch_normalization import LessBN
from mindspore.boost.grad_freeze import GradientFreeze
from mindspore.boost.base import OptimizerProcess, ParameterProcess
from mindspore.boost.base import _get_local_pca_mat_path
__all__ = ["AutoBoost"]
_boost_config_mode = ["auto", "manual", "enable_all", "disable_all"]
_boost_config_level = {
"O0": {
"less_bn": False,
"grad_freeze": False,
"adasum": False,
"grad_accumulation": False,
"dim_reduce": False,
'loss_scale_group': False},
"O1": {
"less_bn": True,
"grad_freeze": True,
"adasum": False,
"grad_accumulation": False,
"dim_reduce": False,
'loss_scale_group': False},
"O2": {
"less_bn": True,
"grad_freeze": True,
"adasum": True,
"grad_accumulation": False,
"dim_reduce": False,
'loss_scale_group': False}
}
[文档]class AutoBoost:
r"""
Provide auto accelerating for network.
Args:
level (str): Boost config level. Default: "O0".
boost_config_dict (dict): User config hyperparameter dict, recommended config format:
.. code-block::
{
"boost": {
"mode": "auto",
"less_bn": False,
"grad_freeze": False,
"adasum": False,
"grad_accumulation": False,
"dim_reduce": False,
"loss_scale_group": False
},
"common": {
"gradient_split_groups": [50, 100],
"device_number": 8
},
"less_bn": {
"fn_flag": True,
"gc_flag": True
},
"grad_freeze": {
"param_groups": 10,
"freeze_type": 1,
"freeze_p": 0.7,
"total_steps": 65536
}
"grad_accumulation": {
"grad_accumulation_step": 1
},
"dim_reduce": {
"rho": 0.55,
"gamma": 0.9,
"alpha": 0.001,
"sigma": 0.4,
"n_components": 32,
"pca_mat_path": None,
"weight_load_dir": None,
"timeout": 1800
}
}
- boost:
- mode (str): How to set the boost. Supports ["auto", "manual", "enable_all", "disable_all"].
Default: "auto".
- auto: Depend on the argument "boost_level" in class Model.
- manual: Depend on "boost_config_dict".
- enable_all: Set all boost functions true.
- disable_all: Set all boost functions false.
- less_bn (bool): Whether to apply less_bn function. Default: False.
- grad_freeze: (bool): Whether to apply grad_freeze function. Default: False.
- adasum (bool): Whether to apply adasum function. Default: False.
- grad_accumulation (bool): Whether to apply grad_accumulation function. Default: False.
- dim_reduce (bool): Whether to apply dim_reduce function. Default: False.
- loss_scale_group (bool): Whether to apply loss_scale_group function. Default: False.
If set dim_reduce true, other functions will be false.
If set grad_freeze true and dim_reduce false, other functions will be false.
- common:
- gradient_split_groups (list): The gradient split point of this network. Default: [50, 100].
- device_number (int): Device number. Default: 8.
- less_bn:
- fn_flag (bool): Whether changing fc to fn. Default: True.
- gc_flag (bool): Whether to apply gc. Default: True.
- grad_freeze:
- param_groups (int): The number of parameter groups. Default: 10.
- freeze_type (int): Gradient freeze grouping strategy, select from [0, 1]. Default: 1.
- freeze_p (float): Gradient freezing probability. Default: 0.7.
- total_steps (int): Total training steps. Default: 65536.
- grad_accumulation:
- grad_accumulation_step (int): Steps to accumulate gradients. Default: 1.
- dim_reduce:
The leading principles of dim_reduce:
.. math::
\begin{align}
grad\_k &= pca\_mat \cdot grad\\
dk &= - bk \cdot grad\_k\\
sk &= rho ^ m \cdot dk\\
delta\_loss &= sigma \cdot grad\_k.T \cdot sk
\end{align}
Here:
- pca_mat (array): Shape :math:`(k*n)`, k is part of n_components, n is the size of weight.
- bk (array): Shape :math:`(k*k)`, is the symmetric positive definite matrix in Quasi-Newton method.
we need to find the m satisfy:
.. math::
new\_loss < old\_loss + delta\_loss
Then, get delta_grad to update the weights for model:
.. math::
\begin{align}
grad\_k\_proj &= pca\_mat.T \cdot grad\_k\\
new\_grad\_momentum &= gamma \cdot old\_grad\_momentum + grad - grad\_k\_proj\\
delta\_grad &= alpha \cdot new\_grad\_momentum - pca\_mat.T \cdot sk
\end{align}
- rho (float): Generally, it does not need to be modified. Default: 0.55.
- gamma (float): Generally, it does not need to be modified. Default: 0.9.
- alpha (float): Generally, it does not need to be modified. Default: 0.001.
- sigma (float): Generally, it does not need to be modified. Default: 0.4.
- n_components (int): PCA component. Default: 32.
- pca_mat_path (str): The path to load pca mat. Default: None.
- weight_load_dir (str): The directory to load weight files saved as ckpt. Default: None.
- timeout (int): Waiting time to load local pca mat. Default: 1800 (second).
User can load the config through the JSON file or use the dictionary directly.
The unconfigured parameters will adopt the default values.
Raises:
ValueError: The boost mode not in ["auto", "manual", "enable_all", "disable_all"].
Supported Platforms:
``Ascend``
Examples:
>>> from mindspore.boost import AutoBoost
>>> #1) when configuring the dict directly:
>>> boost_config_dict = {"boost": {"mode": "auto"}}
>>> boost = AutoBoost("O1", boost_config_dict)
>>>
>>> #2) when loading the dict from a json file:
>>> import json
>>> boost_json = "/path/boost_config.json"
>>> with open(boost_json, 'r') as fp:
>>> boost_config_dict = json.load(fp)
>>> boost = AutoBoost("O1", boost_config_dict)
"""
_instance_lock = threading.Lock()
_instance = None
# pylint: disable=unused-argument
def __new__(cls, *args, **kwargs):
if AutoBoost._instance is None:
with AutoBoost._instance_lock:
if AutoBoost._instance is None:
AutoBoost._instance = object.__new__(cls)
AutoBoost._instance.level = None
AutoBoost._instance.boost_config_dict = None
return AutoBoost._instance
def __init__(self, level="O0", boost_config_dict=""):
if level not in _boost_config_level.keys():
level = "O0"
if self._instance.level is None:
self.level = level
self.boost_config_dict = boost_config_dict
self._fn_flag = True
self._gc_flag = True
self._param_groups = 10
self._freeze_type = 1
self._freeze_p = 0.7
self._total_steps = 65536
self.gradient_groups = None
self.device_number = 8
self.grad_accumulation_step = 1
self.rho = 0.55
self.gamma = 0.9
self.alpha = 0.001
self.sigma = 0.4
self.n_components = 32
self.pca_mat_path = None
self.weight_load_dir = None
self.local_pca_mat_path = None
self.timeout = 1800
self.boost_config = self._get_configuration(level, self.boost_config_dict)
self._param_processer = ParameterProcess()
[文档] def network_auto_process_train(self, network, optimizer):
r"""
Boost network train.
Args:
network (Cell): The training network.
optimizer (Cell): Optimizer for updating the weights.
"""
if self.boost_config.get("dim_reduce"):
self.local_pca_mat_path = _get_local_pca_mat_path(self.weight_load_dir, self.pca_mat_path,
self.n_components, self.device_number, network)
optimizer = SGD(network.trainable_params(), learning_rate=1)
setattr(optimizer, "dim_reduce", True)
return network, optimizer
if self.boost_config.get("less_bn"):
network = LessBN(network, fn_flag=self._fn_flag)
optimizer_process = OptimizerProcess(optimizer)
group_params = self._param_processer.assign_parameter_group(network.trainable_params(),
self.gradient_groups)
optimizer_process.origin_params = \
ParameterProcess.generate_group_params(group_params, optimizer_process.origin_params)
if self._gc_flag:
optimizer_process.add_grad_centralization(network)
optimizer = optimizer_process.generate_new_optimizer()
if self.boost_config.get("grad_freeze"):
freeze_processer = GradientFreeze(self._param_groups, self._freeze_type,
self._freeze_p, self._total_steps)
network, optimizer = freeze_processer.freeze_generate(network, optimizer)
if self.boost_config.get("adasum"):
setattr(optimizer, "adasum", True)
return network, optimizer
[文档] def network_auto_process_eval(self, network):
r"""
Boost network eval.
Args:
network (Cell): The inference network.
"""
if self.boost_config.get("dim_reduce"):
return network
if self.boost_config.get("less_bn"):
network = LessBN(network)
return network
def _set_fn_flag(self, fn_flag):
self._fn_flag = fn_flag
def _set_gc_flag(self, gc_flag):
self._gc_flag = gc_flag
def _set_param_groups(self, param_groups):
self._param_groups = param_groups
def _set_freeze_type(self, freeze_type):
self._freeze_type = freeze_type
def _set_freeze_p(self, freeze_p):
self._freeze_p = freeze_p
def _set_total_steps(self, total_steps):
self._total_steps = total_steps
def _set_device_number(self, device_number):
self.device_number = device_number
def _set_grad_accumulation_step(self, grad_accumulation_step):
self.grad_accumulation_step = grad_accumulation_step
def _set_gradient_split_groups(self, gradient_groups):
if not isinstance(gradient_groups, (list, int)):
raise ValueError(f"gradient_groups `{gradient_groups}` is not in (list, int)")
if isinstance(gradient_groups, int):
gradient_groups = list(gradient_groups)
self.gradient_groups = gradient_groups
def _set_rho(self, rho):
self.rho = rho
def _set_gamma(self, gamma):
self.gamma = gamma
def _set_alpha(self, alpha):
self.alpha = alpha
def _set_sigma(self, sigma):
self.sigma = sigma
def _set_n_components(self, n_components):
self.n_components = n_components
def _set_pca_mat_path(self, pca_mat_path):
self.pca_mat_path = pca_mat_path
def _set_weight_load_dir(self, weight_load_dir):
self.weight_load_dir = weight_load_dir
def _set_timeout(self, timeout):
self.timeout = timeout
def _get_configuration(self, level, boost_config_dict):
"""Get configuration."""
level_config = _boost_config_level.get(level)
if not boost_config_dict:
return level_config
mode = "auto"
if 'boost' in boost_config_dict and 'mode' in boost_config_dict['boost']:
mode = boost_config_dict['boost']['mode']
if mode not in _boost_config_mode:
raise ValueError("The boost mode must be in {}, but got {}".format(_boost_config_mode, mode))
if mode == "manual":
for key, value in boost_config_dict["boost"].items():
if key in level_config:
level_config[key] = value
elif mode == "enable_all":
level_config = {key: True for key in level_config}
elif mode == "disable_all":
level_config = {key: False for key in level_config}
self._do_new_config_func(boost_config_dict, level_config)
return level_config
def _do_new_config_func(self, boost_config_dict, level_config):
valid_boost_each_mode_config = []
for key, boost_each_mode_config in boost_config_dict.items():
if key in level_config.keys() and level_config[key] or key == "common":
valid_boost_each_mode_config.append(boost_each_mode_config)
for boost_each_mode_config in valid_boost_each_mode_config:
for key_s in boost_each_mode_config.keys():
if key_s in self._boost_config_func_map:
self._boost_config_func_map[key_s](self, boost_each_mode_config[key_s])
_boost_config_func_map = {
"fn_flag": _set_fn_flag,
"gc_flag": _set_gc_flag,
"param_groups": _set_param_groups,
"freeze_type": _set_freeze_type,
"freeze_p": _set_freeze_p,
"total_steps": _set_total_steps,
"device_number": _set_device_number,
"gradient_split_groups": _set_gradient_split_groups,
"grad_accumulation_step": _set_grad_accumulation_step,
"rho": _set_rho,
"gamma": _set_gamma,
"alpha": _set_alpha,
"sigma": _set_sigma,
"n_components": _set_n_components,
"pca_mat_path": _set_pca_mat_path,
"weight_load_dir": _set_weight_load_dir,
"timeout": _set_timeout
}