-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathadam_win.py
192 lines (156 loc) · 8.74 KB
/
adam_win.py
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
# Copyright 2022 Garena Online Private Limited
#
# 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.
import math
import torch
from torch.optim.optimizer import Optimizer
class Adam_Win(Optimizer):
r"""Implements Win- and Win2-accelerated Adam algorithm.
The original Adam algorithm was proposed in `Adam: A Method for Stochastic Optimization`_.
Arguments:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float, optional): learning rate (default: 1e-3)
betas (Tuple[float, float], optional): two coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))
reckless_steps (Tuple[float, float], optional): two coefficients used as the multiples
of the reckless stepsizes over the conservative stepsize in Win and Win2 (default: (2.0, 8.0))
eps (float, optional): term added to the denominator to improve
numerical stability (default: 1e-8)
weight_decay (float, optional): weight decay coefficient (default: 1e-2)
amsgrad (boolean, optional): whether to use the AMSGrad variant of this
algorithm from the paper `On the Convergence of Adam and Beyond`_
(default: False)
max_grad_norm (float, optional): value used to clip global grad norm (default: 0.0, no gradient clip)
acceleration_mode (string, optional): win or win2 or none (vanilla AdamW)
"""
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), reckless_steps=(2.0, 8.0), eps=1e-8,
weight_decay=0, amsgrad=False, max_grad_norm=0.0, acceleration_mode=None):
if not 0.0 <= lr:
raise ValueError("Invalid learning rate: {}".format(lr))
if not 0.0 <= eps:
raise ValueError("Invalid epsilon value: {}".format(eps))
if not 0.0 <= betas[0] < 1.0:
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
if not 0.0 <= betas[1] < 1.0:
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
if reckless_steps[0] < 0.0:
raise ValueError("Invalid reckless_steps parameter at index 0: {}".format(reckless_steps[0]))
if reckless_steps[1] < 0.0:
raise ValueError("Invalid reckless_steps parameter at index 1: {}".format(reckless_steps[1]))
defaults = dict(lr=lr, betas=betas, reckless_steps=reckless_steps, eps=eps,
weight_decay=weight_decay, amsgrad=amsgrad,
acceleration_mode=acceleration_mode, max_grad_norm=max_grad_norm)
super(Adam_Win, self).__init__(params, defaults)
def __setstate__(self, state):
super(Adam_Win, self).__setstate__(state)
for group in self.param_groups:
group.setdefault('amsgrad', False)
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()
## whether perform gradient clip
if self.defaults['max_grad_norm'] > 1e-8:
device = self.param_groups[0]['params'][0].device
global_grad_norm = torch.zeros(1, device=device)
max_grad_norm = torch.tensor(self.defaults['max_grad_norm'], device=device)
for group in self.param_groups:
for p in group['params']:
if p.grad is not None:
grad = p.grad
global_grad_norm.add_(grad.pow(2).sum())
global_grad_norm = torch.sqrt(global_grad_norm)
clip_global_grad_norm = torch.clamp(max_grad_norm /(global_grad_norm + group['eps']) , max=1.0)
# parameter update
for group in self.param_groups:
for p in group['params']:
if p.grad is None:
continue
# Perform optimization step
if self.defaults['max_grad_norm'] > 1e-8:
grad = p.grad.mul_(clip_global_grad_norm)
else:
grad = p.grad
if grad.is_sparse:
raise RuntimeError('Adam does not support sparse gradients, please consider SparseAdam instead')
amsgrad = group['amsgrad']
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)
# extra variables for acceleration
if group['acceleration_mode'] == 'win':
state['x'] = torch.zeros_like(p)
state['x'].add_(p.data.clone(), alpha=1)
elif group['acceleration_mode'] == 'win2':
state['x'] = torch.zeros_like(p)
state['x'].add_(p.data.clone(), alpha=1)
state['y'] = torch.zeros_like(p)
state['y'].add_(p.data.clone(), alpha=1)
if amsgrad:
# Maintains max of all exp. moving avg. of sq. grad. values
state['max_exp_avg_sq'] = torch.zeros_like(p.data)
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
if amsgrad:
max_exp_avg_sq = state['max_exp_avg_sq']
beta1, beta2 = group['betas']
state['step'] += 1
if group['weight_decay'] != 0:
grad.add_(group['weight_decay'], p.data)
# Decay the first and second moment running average coefficient
exp_avg.mul_(beta1).add_(1 - beta1, grad)
exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
if amsgrad:
# Maintains the maximum of all 2nd moment running avg. till now
torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
# Use the max. for normalizing running avg. of gradient
denom = max_exp_avg_sq.sqrt().add_(group['eps'])
else:
denom = exp_avg_sq.sqrt().add_(group['eps'])
bias_correction1 = 1 - beta1 ** state['step']
bias_correction2 = 1 - beta2 ** state['step']
## Win and Win2 acceleration for parameter update
if 'win' in group['acceleration_mode']:
beta3, beta4 = group['reckless_steps']
## compute parameter update
update = (exp_avg / denom) * (math.sqrt(bias_correction2) / bias_correction1)
## update x
lr_x = group['lr']
state['x'].add_(update, alpha= -lr_x)
lr_y = beta3 * group['lr']
gamma = 1.0 / (1.0 + lr_y / lr_x)
if group['acceleration_mode'] == 'win':
## update y
p.data.mul_(gamma).add_(state['x'], alpha = (lr_y / lr_x) * gamma).add_(update, alpha= - lr_y * gamma)
elif group['acceleration_mode'] == 'win2':
## update y
state['y'].data.mul_(gamma).add_(state['x'], alpha = (lr_y / lr_x) * gamma).add_(update, alpha= - lr_y * gamma)
## update z
lr_z = beta4 * group['lr']
gamma = 1.0 / (1.0 + lr_z / lr_x + lr_z / lr_y)
p.data.mul_(gamma).add_(update, alpha= - lr_z * gamma)
p.data.add_(state['x'], alpha = (lr_z / lr_x) * gamma).add_(state['y'], alpha = (lr_z / lr_y) * gamma)
else: ## vanilla Adam optimizer
step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
p.data.addcdiv_(-step_size, exp_avg, denom)
return loss