optimizer.py

# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# 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.

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals

from paddle import optimizer as optim


class Momentum(object):
    """
    Simple Momentum optimizer with velocity state.
    Args:
        learning_rate (float|Variable) - The learning rate used to update parameters.
            Can be a float value or a Variable with one float value as data element.
        momentum (float) - Momentum factor.
        regularization (WeightDecayRegularizer, optional) - The strategy of regularization.
    """

    def __init__(
        self, learning_rate, momentum, weight_decay=None, grad_clip=None, **args
    ):
        super(Momentum, self).__init__()
        self.learning_rate = learning_rate
        self.momentum = momentum
        self.weight_decay = weight_decay
        self.grad_clip = grad_clip

    def __call__(self, model):
        train_params = [
            param for param in model.parameters() if param.trainable is True
        ]
        opt = optim.Momentum(
            learning_rate=self.learning_rate,
            momentum=self.momentum,
            weight_decay=self.weight_decay,
            grad_clip=self.grad_clip,
            parameters=train_params,
        )
        return opt


class Adam(object):
    def __init__(
        self,
        learning_rate=0.001,
        beta1=0.9,
        beta2=0.999,
        epsilon=1e-08,
        parameter_list=None,
        weight_decay=None,
        grad_clip=None,
        name=None,
        lazy_mode=False,
        **kwargs
    ):
        self.learning_rate = learning_rate
        self.beta1 = beta1
        self.beta2 = beta2
        self.epsilon = epsilon
        self.parameter_list = parameter_list
        self.learning_rate = learning_rate
        self.weight_decay = weight_decay
        self.grad_clip = grad_clip
        self.name = name
        self.lazy_mode = lazy_mode
        self.group_lr = kwargs.get("group_lr", False)
        self.training_step = kwargs.get("training_step", None)

    def __call__(self, model):
        if self.group_lr:
            if self.training_step == "LF_2":
                import paddle

                if isinstance(model, paddle.DataParallel):  # multi gpu
                    mlm = model._layers.head.MLM_VRM.MLM.parameters()
                    pre_mlm_pp = (
                        model._layers.head.MLM_VRM.Prediction.pp_share.parameters()
                    )
                    pre_mlm_w = (
                        model._layers.head.MLM_VRM.Prediction.w_share.parameters()
                    )
                else:  # single gpu
                    mlm = model.head.MLM_VRM.MLM.parameters()
                    pre_mlm_pp = model.head.MLM_VRM.Prediction.pp_share.parameters()
                    pre_mlm_w = model.head.MLM_VRM.Prediction.w_share.parameters()

                total = []
                for param in mlm:
                    total.append(id(param))
                for param in pre_mlm_pp:
                    total.append(id(param))
                for param in pre_mlm_w:
                    total.append(id(param))

                group_base_params = [
                    param for param in model.parameters() if id(param) in total
                ]
                group_small_params = [
                    param for param in model.parameters() if id(param) not in total
                ]
                train_params = [
                    {"params": group_base_params},
                    {
                        "params": group_small_params,
                        "learning_rate": self.learning_rate.values[0] * 0.1,
                    },
                ]

            else:
                print("group lr currently only support VisionLAN in LF_2 training step")
                train_params = [
                    param for param in model.parameters() if param.trainable is True
                ]
        else:
            train_params = [
                param for param in model.parameters() if param.trainable is True
            ]

        opt = optim.Adam(
            learning_rate=self.learning_rate,
            beta1=self.beta1,
            beta2=self.beta2,
            epsilon=self.epsilon,
            weight_decay=self.weight_decay,
            grad_clip=self.grad_clip,
            name=self.name,
            lazy_mode=self.lazy_mode,
            parameters=train_params,
        )
        return opt


class RMSProp(object):
    """
    Root Mean Squared Propagation (RMSProp) is an unpublished, adaptive learning rate method.
    Args:
        learning_rate (float|Variable) - The learning rate used to update parameters.
            Can be a float value or a Variable with one float value as data element.
        momentum (float) - Momentum factor.
        rho (float) - rho value in equation.
        epsilon (float) - avoid division by zero, default is 1e-6.
        regularization (WeightDecayRegularizer, optional) - The strategy of regularization.
    """

    def __init__(
        self,
        learning_rate,
        momentum=0.0,
        rho=0.95,
        epsilon=1e-6,
        weight_decay=None,
        grad_clip=None,
        **args
    ):
        super(RMSProp, self).__init__()
        self.learning_rate = learning_rate
        self.momentum = momentum
        self.rho = rho
        self.epsilon = epsilon
        self.weight_decay = weight_decay
        self.grad_clip = grad_clip

    def __call__(self, model):
        train_params = [
            param for param in model.parameters() if param.trainable is True
        ]
        opt = optim.RMSProp(
            learning_rate=self.learning_rate,
            momentum=self.momentum,
            rho=self.rho,
            epsilon=self.epsilon,
            weight_decay=self.weight_decay,
            grad_clip=self.grad_clip,
            parameters=train_params,
        )
        return opt


class Adadelta(object):
    def __init__(
        self,
        learning_rate=0.001,
        epsilon=1e-08,
        rho=0.95,
        parameter_list=None,
        weight_decay=None,
        grad_clip=None,
        name=None,
        **kwargs
    ):
        self.learning_rate = learning_rate
        self.epsilon = epsilon
        self.rho = rho
        self.parameter_list = parameter_list
        self.learning_rate = learning_rate
        self.weight_decay = weight_decay
        self.grad_clip = grad_clip
        self.name = name

    def __call__(self, model):
        train_params = [
            param for param in model.parameters() if param.trainable is True
        ]
        opt = optim.Adadelta(
            learning_rate=self.learning_rate,
            epsilon=self.epsilon,
            rho=self.rho,
            weight_decay=self.weight_decay,
            grad_clip=self.grad_clip,
            name=self.name,
            parameters=train_params,
        )
        return opt


class AdamW(object):
    def __init__(
        self,
        learning_rate=0.001,
        beta1=0.9,
        beta2=0.999,
        epsilon=1e-8,
        weight_decay=0.01,
        multi_precision=False,
        grad_clip=None,
        no_weight_decay_name=None,
        one_dim_param_no_weight_decay=False,
        name=None,
        lazy_mode=False,
        **args
    ):
        super().__init__()
        self.learning_rate = learning_rate
        self.beta1 = beta1
        self.beta2 = beta2
        self.epsilon = epsilon
        self.grad_clip = grad_clip
        self.weight_decay = 0.01 if weight_decay is None else weight_decay
        self.grad_clip = grad_clip
        self.name = name
        self.lazy_mode = lazy_mode
        self.multi_precision = multi_precision
        self.no_weight_decay_name_list = (
            no_weight_decay_name.split() if no_weight_decay_name else []
        )
        self.one_dim_param_no_weight_decay = one_dim_param_no_weight_decay

    def __call__(self, model):
        parameters = [param for param in model.parameters() if param.trainable is True]

        self.no_weight_decay_param_name_list = [
            p.name
            for n, p in model.named_parameters()
            if any(nd in n for nd in self.no_weight_decay_name_list)
        ]

        if self.one_dim_param_no_weight_decay:
            self.no_weight_decay_param_name_list += [
                p.name for n, p in model.named_parameters() if len(p.shape) == 1
            ]

        opt = optim.AdamW(
            learning_rate=self.learning_rate,
            beta1=self.beta1,
            beta2=self.beta2,
            epsilon=self.epsilon,
            parameters=parameters,
            weight_decay=self.weight_decay,
            multi_precision=self.multi_precision,
            grad_clip=self.grad_clip,
            name=self.name,
            lazy_mode=self.lazy_mode,
            apply_decay_param_fun=self._apply_decay_param_fun,
        )
        return opt

    def _apply_decay_param_fun(self, name):
        return name not in self.no_weight_decay_param_name_list