File size: 8,953 Bytes
212111c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 |
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Modifications by Roshan Rao
#
# 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.
"""PyTorch optimization for BERT model."""
import logging
import math
import torch
from torch.optim import Optimizer # type: ignore
from torch.optim.lr_scheduler import LambdaLR
logger = logging.getLogger(__name__)
class ConstantLRSchedule(LambdaLR):
""" Constant learning rate schedule.
"""
def __init__(self, optimizer, last_epoch=-1):
super(ConstantLRSchedule, self).__init__(
optimizer, lambda _: 1.0, last_epoch=last_epoch)
class WarmupConstantSchedule(LambdaLR):
""" Linear warmup and then constant.
Linearly increases learning rate schedule from 0 to 1 over `warmup_steps`
training steps. Keeps learning rate schedule equal to 1. after warmup_steps.
"""
def __init__(self, optimizer, warmup_steps, last_epoch=-1):
self.warmup_steps = warmup_steps
super(WarmupConstantSchedule, self).__init__(
optimizer, self.lr_lambda, last_epoch=last_epoch)
def lr_lambda(self, step):
if step < self.warmup_steps:
return float(step) / float(max(1.0, self.warmup_steps))
return 1.
class WarmupLinearSchedule(LambdaLR):
""" Linear warmup and then linear decay.
Linearly increases learning rate from 0 to 1 over `warmup_steps` training steps.
Linearly decreases learning rate from 1. to 0. over remaining `t_total - warmup_steps`
steps.
"""
def __init__(self, optimizer, warmup_steps, t_total, last_epoch=-1):
self.warmup_steps = warmup_steps
self.t_total = t_total
super(WarmupLinearSchedule, self).__init__(
optimizer, self.lr_lambda, last_epoch=last_epoch)
def lr_lambda(self, step):
if step < self.warmup_steps:
return float(step) / float(max(1, self.warmup_steps))
return max(0.0, float(self.t_total - step) / float(
max(1.0, self.t_total - self.warmup_steps)))
class WarmupCosineSchedule(LambdaLR):
""" Linear warmup and then cosine decay.
Linearly increases learning rate from 0 to 1 over `warmup_steps` training steps.
Decreases learning rate from 1. to 0. over remaining `t_total - warmup_steps` steps
following a cosine curve. If `cycles` (default=0.5) is different from default, learning
rate follows cosine function after warmup.
"""
def __init__(self, optimizer, warmup_steps, t_total, cycles=.5, last_epoch=-1):
self.warmup_steps = warmup_steps
self.t_total = t_total
self.cycles = cycles
super(WarmupCosineSchedule, self).__init__(
optimizer, self.lr_lambda, last_epoch=last_epoch)
def lr_lambda(self, step):
if step < self.warmup_steps:
return float(step) / float(max(1.0, self.warmup_steps))
# progress after warmup
progress = float(step - self.warmup_steps) / float(
max(1, self.t_total - self.warmup_steps))
return max(0.0, 0.5 * (1. + math.cos(math.pi * float(self.cycles) * 2.0 * progress)))
class WarmupCosineWithHardRestartsSchedule(LambdaLR):
""" Linear warmup and then cosine cycles with hard restarts.
Linearly increases learning rate from 0 to 1 over `warmup_steps` training steps.
If `cycles` (default=1.) is different from default, learning rate follows `cycles` times
a cosine decaying learning rate (with hard restarts).
"""
def __init__(self, optimizer, warmup_steps, t_total, cycles=1., last_epoch=-1):
self.warmup_steps = warmup_steps
self.t_total = t_total
self.cycles = cycles
super(WarmupCosineWithHardRestartsSchedule, self).__init__(
optimizer, self.lr_lambda, last_epoch=last_epoch)
def lr_lambda(self, step):
if step < self.warmup_steps:
return float(step) / float(max(1, self.warmup_steps))
# progress after warmup
progress = float(step - self.warmup_steps) / float(
max(1, self.t_total - self.warmup_steps))
if progress >= 1.0:
return 0.0
return max(0.0, 0.5 * (1. + math.cos(
math.pi * ((float(self.cycles) * progress) % 1.0))))
class AdamW(Optimizer):
""" Implements Adam algorithm with weight decay fix.
Parameters:
lr (float): learning rate. Default 1e-3.
betas (tuple of 2 floats): Adams beta parameters (b1, b2). Default: (0.9, 0.999)
eps (float): Adams epsilon. Default: 1e-6
weight_decay (float): Weight decay. Default: 0.0
correct_bias (bool): can be set to False to avoid correcting bias in Adam
(e.g. like in Bert TF repository). Default True.
"""
def __init__(self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-6,
weight_decay=0.0,
correct_bias=True):
if lr < 0.0:
raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
if not 0.0 <= betas[0] < 1.0:
raise ValueError(f"Invalid beta parameter: {betas[0]} - should be in [0.0, 1.0)")
if not 0.0 <= betas[1] < 1.0:
raise ValueError(f"Invalid beta parameter: {betas[1]} - should be in [0.0, 1.0)")
if not 0.0 <= eps:
raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(eps))
defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay,
correct_bias=correct_bias)
super(AdamW, self).__init__(params, defaults)
def step(self, closure=None):
"""Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
loss = closure()
for group in self.param_groups:
for p in group['params']:
if p.grad is None:
continue
grad = p.grad.data
if grad.is_sparse:
raise RuntimeError('Adam does not support sparse gradients, '
'please consider SparseAdam instead')
state = self.state[p]
# State initialization
if len(state) == 0:
state['step'] = 0
# Exponential moving average of gradient values
state['exp_avg'] = torch.zeros_like(p.data)
# Exponential moving average of squared gradient values
state['exp_avg_sq'] = torch.zeros_like(p.data)
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
beta1, beta2 = group['betas']
state['step'] += 1
# Decay the first and second moment running average coefficient
# In-place operations to update the averages at the same time
exp_avg.mul_(beta1).add_(1.0 - beta1, grad)
exp_avg_sq.mul_(beta2).addcmul_(1.0 - beta2, grad, grad)
denom = exp_avg_sq.sqrt().add_(group['eps'])
step_size = group['lr']
if group['correct_bias']: # No bias correction for Bert
bias_correction1 = 1.0 - beta1 ** state['step']
bias_correction2 = 1.0 - beta2 ** state['step']
step_size = step_size * math.sqrt(bias_correction2) / bias_correction1
p.data.addcdiv_(-step_size, exp_avg, denom)
# Just adding the square of the weights to the loss function is *not*
# the correct way of using L2 regularization/weight decay with Adam,
# since that will interact with the m and v parameters in strange ways.
#
# Instead we want to decay the weights in a manner that doesn't interact
# with the m/v parameters. This is equivalent to adding the square
# of the weights to the loss with plain (non-momentum) SGD.
# Add weight decay at the end (fixed version)
if group['weight_decay'] > 0.0:
p.data.add_(-group['lr'] * group['weight_decay'], p.data)
return loss
|