-
Notifications
You must be signed in to change notification settings - Fork 65
/
generate_question.py
130 lines (110 loc) · 7.07 KB
/
generate_question.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
"""
Generate questions for LLMs and save it as a task
"""
import argparse
import os
import sys
import json
from prompt.prompt_builder import prompt_factory
from utils.data_builder import load_data
from utils.enums import REPR_TYPE, EXAMPLE_TYPE, SELECTOR_TYPE, LLM
from utils.utils import cost_estimate
from tqdm import tqdm
PATH_DATA = "dataset/"
sys.path.append("./")
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--data_type", type=str, choices=["spider", "realistic", "bird"], default="spider")
parser.add_argument("--split", type=str, choices=["train", "test"], default="test", required=True)
parser.add_argument("--k_shot", type=int, default=0, help="Number of examples")
parser.add_argument("--prompt_repr", type=str, choices=[REPR_TYPE.CODE_REPRESENTATION,
REPR_TYPE.TEXT_REPRESENTATION,
REPR_TYPE.OPENAI_DEMOSTRATION,
REPR_TYPE.BASIC,
REPR_TYPE.ALPACA_SFT,
REPR_TYPE.OPENAI_DEMOSTRATION_WFK,
REPR_TYPE.BASIC_WOFK,
REPR_TYPE.TEXT_REPRESENTATION_WFK,
REPR_TYPE.ALPACA_SFT_WFK,
REPR_TYPE.OPENAI_DEMOSTRATION_WORULE,
REPR_TYPE.CODE_REPRESENTATION_WRULE,
REPR_TYPE.ALPACA_SFT_WRULE,
REPR_TYPE.TEXT_REPRESENTATION_WRULE,
REPR_TYPE.CODE_REPRESENTATION_COT,
REPR_TYPE.TEXT_REPRESENTATION_COT,
REPR_TYPE.OPENAI_DEMOSTRATION_COT,
REPR_TYPE.ALPACA_SFT_COT,
REPR_TYPE.CBR])
parser.add_argument("--example_type", type=str, choices=[EXAMPLE_TYPE.ONLY_SQL,
EXAMPLE_TYPE.QA,
EXAMPLE_TYPE.COMPLETE,
EXAMPLE_TYPE.QAWRULE,
EXAMPLE_TYPE.OPENAI_DEMOSTRATION_QA,
EXAMPLE_TYPE.BASIC_QA], default=None)
parser.add_argument("--selector_type", type=str, choices=[SELECTOR_TYPE.COS_SIMILAR,
SELECTOR_TYPE.RANDOM,
SELECTOR_TYPE.EUC_DISTANCE,
SELECTOR_TYPE.EUC_DISTANCE_THRESHOLD,
SELECTOR_TYPE.EUC_DISTANCE_SKELETON_SIMILARITY_THRESHOLD,
SELECTOR_TYPE.EUC_DISTANCE_QUESTION_MASK,
SELECTOR_TYPE.EUC_DISTANCE_PRE_SKELETON_SIMILARITY_THRESHOLD,
SELECTOR_TYPE.EUC_DISTANCE_PRE_SKELETON_SIMILARITY_PLUS,
SELECTOR_TYPE.EUC_DISTANCE_MASK_PRE_SKELETON_SIMILARITY_THRESHOLD,
SELECTOR_TYPE.EUC_DISTANCE_MASK_PRE_SKELETON_SIMILARITY_THRESHOLD_SHIFT
], default=None)
parser.add_argument("--max_seq_len", type=int, default=2048, help="The maximal length that LLM takes")
parser.add_argument("--max_ans_len", type=int, default=200, help="The maximal length that an answer takes")
parser.add_argument("--tokenizer", type=str, default="gpt-3.5-turbo")
parser.add_argument("--scope_factor", type=int, default=100, help="Times of the searching scope")
parser.add_argument("--pre_test_result", type=str, default=None)
args = parser.parse_args()
# load test dataset here
data = load_data(args.data_type, PATH_DATA, args.pre_test_result)
# Read all tables into a dict
databases = data.get_databases()
# select the prompt
prompt = prompt_factory(args.prompt_repr, args.k_shot, args.example_type, args.selector_type)(data=data, tokenizer=args.tokenizer)
# format all questions
questions = list()
token_cnt = 0
# choose split
func_name = f"get_{args.split}_json"
cross_domain = args.split == "train"
for question_json in tqdm(getattr(data, func_name)()):
question_format = prompt.format(target=question_json,
max_seq_len=args.max_seq_len,
max_ans_len=args.max_ans_len,
scope_factor=args.scope_factor,
cross_domain=cross_domain)
questions.append(question_format)
token_cnt += question_format["prompt_tokens"]
# cost estimated
token_cnt = float(token_cnt) / len(questions)
print(f"Total {len(questions)} questions, {token_cnt} tokens per prompt, {token_cnt / len(questions)} tokens per question")
n_total_tokens = len(questions) * args.max_ans_len + token_cnt
cost_gpt_35_turbo = cost_estimate(n_total_tokens, LLM.GPT_35_TURBO)
cost_text_davinci_003 = cost_estimate(n_total_tokens, LLM.TEXT_DAVINCI_003)
example_quality = prompt.get_example_quality()
# example_quality_each = prompt.get_example_quality_for_each()
pattern_similarity = prompt.get_pattern_similarity()
print(f"Example quality: {example_quality}")
print(f"Estimated cost for {LLM.GPT_4}: {cost_gpt_35_turbo*20}")
print(f"Estimated cost for {LLM.GPT_35_TURBO}: {cost_gpt_35_turbo}")
print(f"Estimated cost for {LLM.TEXT_DAVINCI_003}: {cost_text_davinci_003}")
# save questions
task = {
"args": vars(args),
"costs": {
"prompt_tokens_per_prompt": token_cnt,
"gpt-4": cost_gpt_35_turbo*20,
"gpt-3.5-turbo": cost_gpt_35_turbo,
"text-davinci-003": cost_text_davinci_003,
"example_quality": example_quality,
"pattern_similarity": pattern_similarity,
# "example_quality_for_each": example_quality_each
},
"questions": questions
}
path_generate = f"dataset/process/{args.data_type.upper()}-{args.split.upper()}_{prompt.name}_CTX-{args.max_ans_len}_ANS-{args.max_seq_len}"
os.makedirs(path_generate, exist_ok=True)
json.dump(task, open(os.path.join(path_generate, "questions.json"), "w"), indent=4)