-
Notifications
You must be signed in to change notification settings - Fork 1
/
demo.py
171 lines (137 loc) · 5.99 KB
/
demo.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
import argparse
import logging
import torch
import random
import time
from tqdm import tqdm
import os
from utils import *
import openai
import time
def print_stream(text):
for i in text:
print(i,end='',flush=True)
time.sleep(0.02)
print('\n',end='')
def main():
args = parse_arguments()
fix_seed(args.random_seed)
# Initialize decoder class (load model and tokenizer) ...
decoder = Decoder(args)
demo_F = create_demo_text(args, cot_flag=True)
demo_B = create_verifier_demo_text(args, cot_flag=True)
demo_F = demo_F.split('\n\n')[:-1]
demo_B = demo_B.split('\n\n')[:-1]
while True:
x = input("\033[0;34mWhat do you want to ask me? :)\n[QUESTION] \033[0m")
if 'break' == x.lower():
break
# Prepare question template ...
max_length = args.max_length_cot
declarative = ''
answers = []
x_F = "Q: " + x + "\n" + "A:"
x_F = '\n\n'.join(demo_F) + '\n\n' + x_F
print_stream("\033[0;33m[INFO] Setting Temperature = \033[0m{}".format(args.K))
print_stream("\033[0;33m[INFO] Setting Candidate Answer Number = \033[0m{}".format(args.N))
print('')
print_stream("\033[0;33m[INFO] Now Model is Generating Candidate Answers...\033[0m")
pred = decoder.decode(args, x_F, max_length, 0, args.K, args.N, '\n')
for p_it in pred:
p_item = answer_cleansing(args, p_it)
try:
p_item = str(float(p_item))
except:
pass
if p_item != '':
if p_item not in answers:
answers.append(p_item)
if len(answers) == 0:
print_stream("\033[0;31m[ERROR] The Candidate Answer is None!!!\033[0m")
print_stream("\033[0;31m[ERROR] Break Now! /(ㄒoㄒ)/~~\033[0m")
print('')
else:
if len(answers) == 1:
print_stream("\033[0;32m[ACCEPT] The Candidate Answer's number is 1, so the answer is:\033[0m")
print_stream("\033[0;32m[ACCEPT] {}\033[0m".format(answers[0]))
print('')
else:
print_stream("\033[0;33m[INFO] Now Model is Self-Verificating the Candidate Answers...\033[0m")
scores = {i: 0 for i in range(len(answers))}
pred_verifier = {i: [] for i in range(len(answers))}
for A in range(len(answers)):
decl, answer, declarative = question_turn_decalrative(args, x, answers[A], answers[0],
decoder.decode, declarative)
for d in range(len(decl)):
random.shuffle(demo_B)
x_B = '\n\n'.join(demo_B) + 'Q: ' + decl[d] + '\nA: '
try:
pred_v = decoder.decode(args, x_B, max_length, 0, 0.4, 10, '\n\n')
except:
pred_v = [''] * 10
answers_verifier = []
for p in range(len(pred_v)):
p_item_v = answer_cleansing_verifier(args, pred_v[p])
try:
answers_verifier.append(float(p_item_v))
except:
try:
answers_verifier.append(p_item_v)
except:
pass
try:
score = sum(np.array(answers_verifier) == np.array(float(answer[d])))
except:
try:
score = sum(np.array(answers_verifier) == np.array(answer[d]))
except:
score = 0
pred_verifier[A].append(pred_v)
scores[A] += score
try:
print_stream(f"\033[0;36m[Self-Verification SCORE] The Candidate Answer “{answers[A]}” :\033[0m")
print_stream(f"\033[0;36m[Self-Verification SCORE] {'█'*scores[A]}: {str(scores[A])}\033[0m")
except:
pass
verifier_scores = list(scores.values())
for i in range(len(verifier_scores)):
if verifier_scores[i] == max(verifier_scores):
print_stream(f"\033[0;32m[ACCEPT] The Best Answer is:\033[0m")
print_stream(f"\033[0;32m[ACCEPT] {answers[i]}\033[0m")
print('')
break
def parse_arguments():
parser = argparse.ArgumentParser(description="Reason with self-verification")
parser.add_argument("--random_seed", type=int, default=1, help="random seed")
parser.add_argument(
"--model", type=str, default="text-003",
choices=["gpt3", "gpt3-medium", "gpt3-large", "gpt3-xl", "codex", "codex-001","text-003"],
help="model used for decoding. Note that 'gpt3' are the smallest models."
)
parser.add_argument(
"--max_length_cot", type=int, default=168,
help="maximum length of output tokens by model for reasoning extraction"
)
parser.add_argument(
"--api_time_interval", type=float, default=4.0, help=""
)
parser.add_argument(
"--log_dir", type=str, default="./log/", help="log directory"
)
parser.add_argument(
"--N", type=int, default=5
)
parser.add_argument(
"--K", type=int, default=0.6
)
parser.add_argument(
"--FN", type=int, default=0, help="few-shot number"
)
args = parser.parse_args()
args.dataset = 'gsm8k'
args.direct_answer_trigger_for_fewshot = "The answer is"
args.method = 'verifier_cot'
args.verifier_text = " What is the answer of 'X'?"
return args
if __name__ == "__main__":
main()