-
Notifications
You must be signed in to change notification settings - Fork 2
/
finetune_config.py
153 lines (123 loc) · 6.96 KB
/
finetune_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
import torch
import argparse
from utils import str2bool
def create_finetune_nlp_parser():
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', default='movielens', type=str, help='dataset name')
parser.add_argument("--llm", type = str, default='tiny-bert', help = '' )
parser.add_argument("--sample", type = str2bool, default=False, help = "sample dataset")
# mixed precision
parser.add_argument("--mixed_precision", type = str2bool, default=False)
# parameter for pretrain
parser.add_argument("--model_path", type = str, help = "")
parser.add_argument("--temperature", type = str, help = "") # 0.07, 0.1
parser.add_argument("--use_mfm", type = str2bool, help = "")
parser.add_argument("--use_mlm", type = str2bool, help = "")
parser.add_argument("--pre_epochs", type = str, help = "")
parser.add_argument("--pre_lr", type = str, help = "")
# parameter for finetune
parser.add_argument("--batch_size", type = int, default=128, help = "")
parser.add_argument("--epochs", type = int, default=25, help = "")
parser.add_argument("--lr", type = float, default=5e-5, help = "")
parser.add_argument("--num_workers", type = int, default=4, help = "")
parser.add_argument("--weight_decay", type = float, default=1e-3, help = "")
parser.add_argument("--patience", type = int, default=3, help = "")
parser.add_argument("--factor", type = float, default=0.95, help = "")
parser.add_argument("--dropout", type = float, default=0.2 , help = "")
parser.add_argument("--use_special_token", type = str2bool, default=False)
parser.add_argument("--obs", type = str2bool, default=True, help = "")
args = parser.parse_args()
args.load_prefix_path = "./"
args.output_prefix_path = './'
# dp
args.batch_size = args.batch_size*torch.cuda.device_count()
args.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if args.dataset == 'movielens':
args.data_path = args.load_prefix_path+"data/ml-1m/"
args.max_length = 100
elif args.dataset == 'bookcrossing':
args.data_path = args.load_prefix_path+"data/BookCrossing/"
args.max_length = 100
elif args.dataset == 'goodreads':
args.data_path = args.load_prefix_path+"data/GoodReads/"
args.max_length = 180
args.text_path = args.data_path+"text.txt"
args.struct_path = args.data_path+"remap_data.csv"
if args.llm == 'tiny-bert':
args.text_encoder_model = args.load_prefix_path+"pretrained_models/tiny-bert-4l-en/"
args.text_tokenizer = args.load_prefix_path+"pretrained_models/tiny-bert-4l-en/"
args.text_embedding_dim = 312
elif args.llm == 'roberta':
args.text_encoder_model = args.load_prefix_path+"pretrained_models/roberta-base/"
args.text_tokenizer = args.load_prefix_path+"pretrained_models/roberta-base/"
args.text_embedding_dim = 768
elif args.llm == 'roberta-large':
args.text_encoder_model = args.load_prefix_path+"pretrained_models/roberta-large/"
args.text_tokenizer = args.load_prefix_path+"pretrained_models/roberta-large/"
args.text_embedding_dim = 1024
args.sample_ration = 0.01
for k,v in vars(args).items():
print(k,'=',v)
return args
def create_finetune_all_parser():
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', default='movielens', type=str, help='dataset name')
parser.add_argument("--llm", type = str, default='tiny-bert', help = '' )
parser.add_argument("--backbone", type = str, default='DeepFM', help = "")
parser.add_argument("--sample", type = str2bool, default=False, help = "sample dataset")
parser.add_argument("--mixed_precision", type = str2bool, default=False)
parser.add_argument("--model_path", type = str, help = "")
parser.add_argument("--temperature", type = str, help = "") # 0.07, 0.1
parser.add_argument("--use_mfm", type = str2bool, help = "")
parser.add_argument("--use_mlm", type = str2bool, help = "")
parser.add_argument("--pre_epochs", type = str, help = "")
parser.add_argument("--pre_lr", type = str, help = "")
parser.add_argument("--batch_size", type = int, default=128, help = "")
parser.add_argument("--epochs", type = int, default=25, help = "")
parser.add_argument("--lr", type = float, default=5e-5, help = "")
parser.add_argument("--num_workers", type = int, default=4, help = "")
parser.add_argument("--weight_decay", type = float, default=1e-3, help = "")
parser.add_argument("--patience", type = int, default=3, help = "")
parser.add_argument("--factor", type = float, default=0.95, help = "")
parser.add_argument("--use_mlp", type = str2bool, default=False)
parser.add_argument("--use_cls", type = str2bool, default=False)
parser.add_argument("--dropout", type = float, default=0.2 , help = "")
parser.add_argument("--obs", type = str2bool, default=True, help = "")
parser.add_argument("--alpha", type = float, default=0.2 , help = "")
parser.add_argument("--real_mask_index", type = int)
args = parser.parse_args()
args.load_prefix_path = "./"
args.output_prefix_path = './'
# dp
args.batch_size = args.batch_size*torch.cuda.device_count()
args.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if args.dataset == 'movielens':
args.text_path = args.load_prefix_path+"data/ml-1m/text.txt"
args.struct_path = args.load_prefix_path+"data/ml-1m/remap_data.csv"
args.meta_path = args.load_prefix_path+"data/ml-1m/meta.json"
args.max_length = 100
elif args.dataset == 'bookcrossing':
args.text_path = args.load_prefix_path+"data/BookCrossing/text.txt"
args.struct_path = args.load_prefix_path+"data/BookCrossing/remap_data.csv"
args.meta_path = args.load_prefix_path+"data/BookCrossing/meta.json"
args.max_length = 100
elif args.dataset == 'goodreads':
args.text_path = args.load_prefix_path+"data/GoodReads/text.txt"
args.struct_path = args.load_prefix_path+"data/GoodReads/remap_data.csv"
args.max_length = 180
if args.llm == 'tiny-bert':
args.text_encoder_model = args.load_prefix_path+"pretrained_models/tiny-bert-4l-en/"
args.text_tokenizer = args.load_prefix_path+"pretrained_models/tiny-bert-4l-en/"
args.text_embedding_dim = 312
elif args.llm == 'roberta':
args.text_encoder_model = args.load_prefix_path+"pretrained_models/roberta-base/"
args.text_tokenizer = args.load_prefix_path+"pretrained_models/roberta-base/"
args.text_embedding_dim = 768
elif args.llm == 'roberta-large':
args.text_encoder_model = args.load_prefix_path+"pretrained_models/roberta-large/"
args.text_tokenizer = args.load_prefix_path+"pretrained_models/roberta-large/"
args.text_embedding_dim = 1024
args.sample_ration = 0.01
for k,v in vars(args).items():
print(k,'=',v)
return args