-
Notifications
You must be signed in to change notification settings - Fork 39
/
Copy pathoptimizer.py
166 lines (138 loc) · 4.77 KB
/
optimizer.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
# --------------------------------------------------------
# SimMIM
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Ze Liu
# Modified by MX
# --------------------------------------------------------
from functools import partial
from torch import optim as optim
from utils import print_on_rank_zero
def build_optimizer(hparams, model, is_pretrain):
if is_pretrain:
return build_pretrain_optimizer(hparams, model)
else:
return build_finetune_optimizer(hparams, model)
def build_pretrain_optimizer(hparams, model):
skip = {}
skip_keywords = {}
if hasattr(model, 'no_weight_decay'):
skip = model.no_weight_decay()
if hasattr(model, 'no_weight_decay_keywords'):
skip_keywords = model.no_weight_decay_keywords()
parameters = get_pretrain_param_groups(model, skip, skip_keywords)
opt_lower = hparams.optim_type.lower()
optimizer = None
if opt_lower == 'sgd':
optimizer = optim.SGD(parameters, momentum=0.9, nesterov=True,
lr=hparams.lr, weight_decay=hparams.weight_decay)
elif opt_lower == 'adamw':
optimizer = optim.AdamW(parameters, betas=(0.9, 0.999),
lr=hparams.lr, weight_decay=hparams.weight_decay)
return optimizer
def get_pretrain_param_groups(model, skip_list=(), skip_keywords=()):
has_decay = []
no_decay = []
has_decay_name = []
no_decay_name = []
for name, param in model.named_parameters():
if not param.requires_grad:
continue
if len(param.shape) == 1 or name.endswith(".bias") or (name in skip_list) or \
check_keywords_in_name(name, skip_keywords):
no_decay.append(param)
no_decay_name.append(name)
else:
has_decay.append(param)
has_decay_name.append(name)
print_on_rank_zero(f'params_no_decay_name: {no_decay_name} \n params_decay_name: {has_decay_name}')
return [{'params': no_decay, 'weight_decay': 0.},
{'params': has_decay},]
def build_finetune_optimizer(hparams, model):
if hparams.arch == 'mvit':
if hparams.layer_decay == 1:
get_layer_func = None
scales = None
else:
num_layers = 16
get_layer_func = partial(get_mvit_layer, num_layers=num_layers + 2)
scales = list(hparams.layer_decay ** i for i in reversed(range(num_layers + 2))) #layer_decay=1 disable
else:
return build_pretrain_optimizer(hparams, model)
skip = {}
skip_keywords = {}
if hasattr(model, 'no_weight_decay'):
skip = model.no_weight_decay()
if hasattr(model, 'no_weight_decay_keywords'):
skip_keywords = model.no_weight_decay_keywords()
parameters = get_finetune_param_groups(
model, hparams.lr, hparams.weight_decay,
get_layer_func, scales, skip, skip_keywords)
opt_lower = hparams.optim_type.lower()
optimizer = None
if opt_lower == 'sgd':
optimizer = optim.SGD(parameters, momentum=0.9, nesterov=True,
lr=hparams.lr, weight_decay=hparams.weight_decay)
elif opt_lower == 'adamw':
optimizer = optim.AdamW(parameters, betas=(0.9, 0.999),
lr=hparams.lr, weight_decay=hparams.weight_decay)
return optimizer
def get_mvit_layer(name, num_layers):
layer_name = name.replace('mvit.', '')
layer_name = layer_name.replace('model.', '')
if layer_name in ("mask_token"):
return 0
elif layer_name.startswith("patch_embed") or layer_name.startswith('cls_positional_encoding'):
return 0
elif layer_name.startswith("blocks"):
layer_id = int(layer_name.split('.')[1])
return layer_id + 1
else:
return num_layers - 1
def get_finetune_param_groups(model, lr, weight_decay, get_layer_func, scales, skip_list=(), skip_keywords=()):
parameter_group_names = {}
parameter_group_vars = {}
for name, param in model.named_parameters():
if not param.requires_grad:
continue
if len(param.shape) == 1 or name.endswith(".bias") or (name in skip_list) or \
check_keywords_in_name(name, skip_keywords):
group_name = "no_decay"
this_weight_decay = 0.
else:
group_name = "decay"
this_weight_decay = weight_decay
if get_layer_func is not None:
layer_id = get_layer_func(name)
group_name = "layer_%d_%s" % (layer_id, group_name)
#print(name, group_name)
else:
layer_id = None
if group_name not in parameter_group_names:
if scales is not None:
scale = scales[layer_id]
else:
scale = 1.
parameter_group_names[group_name] = {
"group_name": group_name,
"weight_decay": this_weight_decay,
"params": [],
"lr": lr * scale,
"lr_scale": scale,
}
parameter_group_vars[group_name] = {
"group_name": group_name,
"weight_decay": this_weight_decay,
"params": [],
"lr": lr * scale,
"lr_scale": scale
}
parameter_group_vars[group_name]["params"].append(param)
parameter_group_names[group_name]["params"].append(name)
return list(parameter_group_vars.values())
def check_keywords_in_name(name, keywords=()):
isin = False
for keyword in keywords:
if keyword in name:
isin = True
return isin