STGE/config.py at main · ASCII-LAB/STGE · GitHub

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
import argparse
import os
from utils import set_seed, get_current_time
import torch


def get_args_parser():
    parser = argparse.ArgumentParser()

    parser.add_argument('--llm_name', default="llama3.1_8b_instruct", type=str, help="large language model.")
    parser.add_argument("--seed", default=42, type=int, help="random seed.")
    parser.add_argument("--task_type", default="eae", type=str, choices=['ner', 're', 'eae'],
                        help="task name.")
    parser.add_argument("--dataset_type", default="ace", type=str, choices=['rams', 'wiki', 'ace', 'nerd'],
                        help="dataset name.")
    parser.add_argument("--max_new_tokens", default=256, type=int, help="model max_new_tokens")
    parser.add_argument("--method_type", default="base", type=str, choices=["stge"],
                        help="select method type")
    parser.add_argument("--logit_strategy", default="llm", type=str, choices=["fusion"], help="logit strategy.")
    parser.add_argument("--window_size", default=250, type=int,
                        help="for document exceeding the length constraint, add a window centering at "
                             "the trigger word and drop words outside this window")
    parser.add_argument("--shot_num", default=20, type=int, help="shot number")
    parser.add_argument("--demo_num", default=2, type=int, choices=[0, 1, 2, 3], help="demo number")
    parser.add_argument('--value_num', type=int, default=2)
    parser.add_argument('--filter', action='store_true')
    parser.add_argument('--rule', action='store_true')


    # scorer_data_process argument
    parser.add_argument("--max_num", type=int, help="max number")
    parser.add_argument("--pred_max_num", type=int, help="pred max number")

    parser.add_argument("--train_save_path", type=str)
    parser.add_argument("--dev_save_path", type=str)
    parser.add_argument('--threshold', type=float, default=0.5)


    # scorer_train argument
    parser.add_argument("--scorer_train_data_path", default="",
                        type=str, help="train data")
    parser.add_argument("--scorer_dev_data_path", default="", type=str,
                        help="dev data")
    parser.add_argument("--scorer_test_data_path", default="", type=str,
                        help="test data")
    parser.add_argument("--load_scorer_path", default="", type=str, help="model path")
    parser.add_argument('--epoch', type=int, default=5)
    parser.add_argument('--batch_size', type=int, default=32)
    parser.add_argument('--warmup_proportion', type=int, default=0.1)
    parser.add_argument('--max_socrer_data_length', type=int, default=512)
    parser.add_argument('--model_name', type=str,
                        default="../../pretrain_model/roberta-large")
    parser.add_argument('--hidden_size', type=int, default=1024)
    parser.add_argument('--weight_decay', type=float, default=0.01)
    parser.add_argument('--learning_rate', type=float, default=2e-05)
    parser.add_argument('--save_step', type=int, default=1000000)
    parser.add_argument('--model_save', type=str, default="nerd_train_large_all")
    parser.add_argument("--train_type", default="supervise", type=str, choices=["supervise", "stge"],
                        help="select train type")

    args = parser.parse_args()
    if args.dataset_type == "rams":
        args.train_path = "../../dataset/RAMS_1.0/data/train.jsonlines"
        args.dev_path = "../../dataset/RAMS_1.0/data/dev.jsonlines"
        args.test_path = "../../dataset/RAMS_1.0/data/test.jsonlines"
        args.convert_path = "../../dataset/RAMS_1.0/data/convert/{}/".format(args.task_type)
        args.scorer_data_path = "../../dataset/RAMS_1.0/scorer_data/{}/".format(args.task_type)
        args.invalid_arg_num = 1
    elif args.dataset_type == "wiki":
        args.train_path = "../../dataset/WIKIEVENT/train.jsonl"
        args.dev_path = "../../dataset/WIKIEVENT/dev.jsonl"
        args.test_path = "../../dataset/WIKIEVENT/test.jsonl"
        args.convert_path = "../../dataset/WIKIEVENT/convert/{}/".format(args.task_type)
        args.scorer_data_path = "../../dataset/WIKIEVENT/scorer_data/{}/".format(args.task_type)
        args.invalid_arg_num = 0
        args.head_only = True
    elif args.dataset_type == "ace":
        args.train_path = "../../dataset/ACE2005/train_convert.json"
        args.dev_path = "../../dataset/ACE2005/dev_convert.json"
        args.test_path = "../../dataset/ACE2005/test_convert.json"
        args.convert_path = "../../dataset/ACE2005/convert/{}/".format(args.task_type)
        args.scorer_data_path = "../../dataset/ACE2005/scorer_data/{}/".format(args.task_type)
        args.invalid_arg_num = 0
    elif args.dataset_type == "nerd":
        args.convert_path = "../../dataset/Few-NERD/convert/{}/".format(args.task_type)
        args.scorer_data_path = "../../dataset/Few-NERD/scorer_data/{}/".format(args.task_type)
    args.template_path = args.convert_path + "template.json"
    args.shot_event_map_path = args.convert_path + "shot_event_map.json"
    args.current_time = get_current_time()
    args.save_result_path = "output/{}/{}_{}_{}_{}_{}_shot_result_{}.json".format(
        args.dataset_type, args.llm_name,
        args.method_type, args.task_type, args.logit_strategy,
        args.shot_num,
        args.current_time)

    args.shot_data_path = args.convert_path + "{}-shot_data.json".format(str(args.shot_num))

    if args.dataset_type == "ace" and args.task_type == "ner":
        args.value_num = 0


    args.scorer_test_data_path = args.scorer_data_path + "/test.tsv"
    if args.dataset_type == "nerd":
        if not os.path.exists("model/{}/".format(args.dataset_type)):
            os.makedirs("model/{}/".format(args.dataset_type))
        args.model_save_path = "model/{}/{}".format(args.dataset_type, args.model_save)
        args.scorer_test_data_path = None
    else:
        if "/" not in args.model_save:
            if not os.path.exists("model/{}/{}/".format(args.dataset_type, args.task_type)):
                os.makedirs("model/{}/{}/".format(args.dataset_type, args.task_type))
            args.model_save_path = "model/{}/{}/{}".format(args.dataset_type, args.task_type, args.model_save)
        else:
            args.model_save_path = args.model_save


    if args.scorer_train_data_path == "":
        args.scorer_train_data_path = args.scorer_data_path + "{}-shot_train_per.tsv".format(args.shot_num)

    if not os.path.exists("output/{}".format(args.dataset_type)):
        os.makedirs("output/{}".format(args.dataset_type))

    if not os.path.exists(args.scorer_data_path):
        os.makedirs(args.scorer_data_path)

    args.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    set_seed(args)
    return args