-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathconfig.py
177 lines (165 loc) · 9.16 KB
/
config.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
import argparse
UNK_IDX = 0
UNK_WORD = "UUUNKKK"
MULTI_BLEU_PERL = 'multi-bleu.perl'
def str2bool(v):
return v.lower() in ('yes', 'true', 't', '1', 'y')
def get_base_parser():
parser = argparse.ArgumentParser(
description='Dialogue Summarization using PyTorch')
parser.register('type', 'bool', str2bool)
basic_group = parser.add_argument_group('basics')
# Basics
basic_group.add_argument('--debug', type="bool", default=False,
help='activation of debug mode (default: False)')
basic_group.add_argument('--auto_disconnect', type="bool", default=True,
help='for slurm (default: True)')
basic_group.add_argument('--save_prefix', type=str, default="experiments",
help='saving path prefix')
basic_group.add_argument('--gen_prefix', type=str, default="gen",
help='generation saving path prefix')
basic_group.add_argument('--model_dir', type=str, default=None,
help='model directory')
basic_group.add_argument('--model_type', type=str, default="reformer",
help='model type')
basic_group.add_argument('--tokenizer_type', type=str, default="reformer",
help='model type')
basic_group.add_argument('--ckpt_name', type=str, default="best.ckpt",
help='checkpoint to load')
data_group = parser.add_argument_group('data')
# Data file
data_group.add_argument('--train_path', type=str, default=None,
help='data file')
data_group.add_argument('--train_pos_path', type=str, default=None,
help='data file')
data_group.add_argument('--vocab_file', type=str, default=None,
help='bpe vocabulary file')
data_group.add_argument('--glossary_file', type=str, default=None,
help='bpe glossary file')
data_group.add_argument('--dev_path', type=str, default=None,
help='data file')
data_group.add_argument('--test_path', type=str, default=None,
help='data file')
data_group.add_argument('--dev_pos_path', type=str, default=None,
help='data file')
data_group.add_argument('--test_pos_path', type=str, default=None,
help='data file')
config_group = parser.add_argument_group('model_configs')
config_group.add_argument('-lr', '--learning_rate',
dest='lr',
type=float,
default=1e-3,
help='learning rate')
config_group.add_argument('-nlayer', '--num_hidden_layer',
dest='nlayer',
type=int,
default=6,
help='number of hidden layers')
config_group.add_argument('-elayer', '--num_encoder_layer',
dest='elayer',
type=int,
default=6,
help='number of encoder layers')
config_group.add_argument('-dlayer', '--num_decoder_layer',
dest='dlayer',
type=int,
default=6,
help='number of decoder layers')
config_group.add_argument('-hsize', '--hidden_size',
dest='hsize',
type=int,
default=512,
help='size of hidden layers')
config_group.add_argument('-adim1', '--axial_pos_embds_dim1',
dest='adim1',
type=int,
default=128,
help='axial_pos_embds_dim 1')
config_group.add_argument('-adim2', '--axial_pos_embds_dim2',
dest='adim2',
type=int,
default=384,
help='axial_pos_embds_dim 2')
config_group.add_argument('-wstep', '--warmup_steps',
dest='wstep', type=int, default=0,
help='learning rate warmup steps')
config_group.add_argument('-nhash', '--num_hashes',
dest='nhash', type=int, default=2,
help='number of hashes')
config_group.add_argument('-mf', '--mask_fraction',
dest='mf', type=float, default=0.0,
help='fraction of decoder input tokens being masked out')
config_group.add_argument('--eps',
type=float,
default=1e-5,
help='safty for avoiding numerical issues')
config_group.add_argument('-gclip', '--grad_clip',
dest='gclip',
type=float, default=1.0,
help='gradient clipping threshold')
config_group.add_argument('--l2', type=float, default=0.,
help='l2 regularization')
config_group.add_argument('-gcs', '--gradient_accumulation_steps',
dest='gcs', type=int, default=1,
help='gradient accumulation steps')
config_group.add_argument('-lsh_dp', '--lsh_attention_probs_dropout_prob',
dest='lsh_dp', type=float, default=0.05,
help='lsh attention dropout prob')
config_group.add_argument('-hdp', '--hidden_dropout_prob',
dest='hdp', type=float, default=0.05,
help='hidden state dropout prob')
config_group.add_argument('-ls', '--label_smoothing',
dest='ls', type=float, default=0.0,
help='label smoothing')
config_group.add_argument('-fns', '--feedforward_hidden_size',
dest='fns', type=int, default=2,
help='multiplier for feedforward neural network hidden size')
config_group.add_argument('-gelu', '--use_gelu',
dest='gelu', type="bool", default=False,
help='whether to use gelu activation')
config_group.add_argument('-fc', '--force_copy',
dest='fc', type="bool", default=False,
help='whether to force copy')
setup_group = parser.add_argument_group('train_setup')
setup_group.add_argument('--n_epoch', type=int, default=5,
help='number of epochs')
setup_group.add_argument('--batch_size', type=int, default=20,
help='batch size')
setup_group.add_argument('--eval_batch_size', type=int, default=20,
help='batch size')
setup_group.add_argument('--max_src_len', type=int, default=None,
help='maximum length')
setup_group.add_argument('--max_src_entry', type=int, default=None,
help='maximum length')
setup_group.add_argument('--max_tot_src_len', type=int, default=None,
help='maximum length')
setup_group.add_argument('--max_tgt_len', type=int, default=None,
help='maximum length')
setup_group.add_argument('--random_perm', type="bool", default=False,
help='whether to randomly permute anonymized entity ids')
setup_group.add_argument('--opt', type=str, default='adam',
choices=['sadam', 'adam', 'sgd', 'rmsprop', 'adamw'],
help='types of optimizer: adam (default), \
sgd, rmsprop')
setup_group.add_argument('--beam_size', type=int, default=5,
help='beam size')
setup_group.add_argument('--max_gen_len', type=int, default=800,
help='maximum generation length')
setup_group.add_argument('--min_gen_len', type=int, default=0,
help='minimum generation length')
setup_group.add_argument('--trigram_blocking', type="bool", default=False,
help='trigram blocking')
setup_group.add_argument('--gradient_checkpointing', type="bool", default=False,
help='gradient checkpointing')
misc_group = parser.add_argument_group('misc')
# misc
misc_group.add_argument('--print_every', type=int, default=500,
help='print training details after \
this number of iterations')
misc_group.add_argument('--eval_every', type=int, default=5000,
help='evaluate model after \
this number of iterations')
misc_group.add_argument('--save_every', type=int, default=2000,
help='save model after \
this number of iterations')
return parser