-
Notifications
You must be signed in to change notification settings - Fork 1
/
generate_CREATOR.py
208 lines (183 loc) · 9.95 KB
/
generate_CREATOR.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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
import json
import re
from utils import process_code, web_chat, check_code_valid
from grader import grade_answer
from cprint import *
# ================================ #
# Choose from "MATH", "TabMWP", "Creation"
task = "MATH"
mode = None
if task == "MATH":
fields = ["algebra", "counting_and_probability", "geometry", "intermediate_algebra", "number_theory", "prealgebra", "precalculus"]
sys_msg = "You are a helpful assistant in answering math competition problems."
elif task == "TabMWP":
fields = ["tabmwp"]
sys_msg = "You are a helpful assistant in answering questions with tabular contents."
elif task == "Creation":
fields = ["CC"]
sys_msg = "You are a helpful assistant in answering newly defined questions."
# For Creation Challenge, you can choose from "normal", "utilityhint" and "allhint"
mode = "allhint"
else:
raise Exception("Wrong task name!")
# ================================ #
start_key = 0
temperature = 0.3
prompt_creation = f"{task}/prompt_lib/prompt_CREATOR_creation.md"
prompt_decision = f"{task}/prompt_lib/prompt_CREATOR_decision.md"
prompt_rectification = f"{task}/prompt_lib/prompt_rectification.md"
code_file = "code_exec/tmp0"
gen_func = web_chat
rectify_limit = 3
# ================================ #
if mode == "utilityhint":
prompt_creation = f"{task}/prompt_lib/prompt_CREATOR_creation_utilityhint.md"
prompt_decision = f"{task}/prompt_lib/prompt_CREATOR_decision_utilityhint.md"
if mode == "allhint":
prompt_creation = f"{task}/prompt_lib/prompt_CREATOR_creation_allhint.md"
prompt_decision = f"{task}/prompt_lib/prompt_CREATOR_decision_allhint.md"
# ================================ #
for field in fields:
save_file = f"{task}/results/results_{field}_CREATOR_temp{temperature}.md"
open(save_file, "w").close()
f = open(prompt_creation, "r")
prompt1 = f.read().strip()
f.close()
f = open(prompt_decision, "r")
prompt2 = f.read().strip()
f.close()
f = open(prompt_rectification, "r")
err_prompt = f.read().strip()
f.close()
lines = []
f = open(f"{task}/dataset/{field}.jsonl", "r")
all_lines = f.readlines()
f.close()
for line in all_lines:
lines.append(json.loads(line.strip()))
correct = 0
wrong = 0
exec_err = 0
one_time_pass = 0
for line in lines[:]:
q = line["question"].strip()
try:
correct_ans = line["answer"]
env1 = prompt1.replace("===qst===", line["question"])
if task == "TabMWP":
env1 = env1.replace("===table===", line["table"])
if task == "Creation":
env1 = env1.replace("===constants===", line["constant"].strip())
if mode == "utilityhint":
hints = "You might use the following hints about what tool to create:\n"
hints += "Utility:\n" + line["utility"].strip() + "\n"
env1 = env1.replace("===hints===", hints.strip())
if mode == "allhint":
hints = "You might use the following hints about what tool to create\n"
hints += "Utility:\n" + line["utility"].strip() + "\n"
hints += "Args:\n" + line["args"].strip() + "\n"
hints += "Return:\n" + line["return"].strip() + "\n"
env1 = env1.replace("===hints===", hints.strip())
response1 = gen_func(env1, sys_msg, temperature=temperature)
tool_time = 0
while tool_time < 3:
# check the validity of the tool format
if (check_code_valid(response1, code_file)):
break
print("!!! Wrong tool format generated !!!")
response1 = gen_func(env1, sys_msg, temperature=temperature)
tool_time += 1
env2 = prompt2.replace("===qst===", line["question"]).replace("===tool===", response1)
if task == "TabMWP":
env2 = env2.replace("===table===", line["table"])
if task == "Creation":
env2 = env2.replace("===constants===", line["constant"].strip())
if mode == "utilityhint" or mode == "allhint":
env2 = env2.replace("===hints===", hints.strip())
response2: str = gen_func(env2, sys_msg, temperature=temperature)
# To further enhance the validity of the tool format
# if "```" not in response1:
# response1 = "```python\n" + response1 + "\n```"
# if "```" not in response2:
# response2 = "```python\n" + response2 + "\n```"
# response = response1 + "\n" + response2
if not response1.startswith("```python"):
response1 = "```python\n" + response1.strip()
if not response1.endswith("```"):
response1 = response1.strip() + "\n```"
if not response2.startswith("```python"):
response2 = "```python\n" + response2.strip()
if not response2.endswith("```"):
response2 = response2.strip() + "\n```"
response = response1 + "\n" + response2
time = 0
while True:
time += 1
if_succ, info = process_code(response, code_file)
if not if_succ and time <= rectify_limit:
print(f"!!! Error in execution, time {time} !!!")
f = open(save_file, "a")
if task == "TabMWP":
f.write("### Table\n" + line["table"] + "\n")
f.write("### Question\n" + line["question"] + "\n### Respose\n" + response.strip() + "\n\n### Back info:\n" + info.strip() + f"\n\nThe {time} time you try!\nThe correct answer should be {correct_ans}\n\n")
f.close()
rectify_env = err_prompt.replace("===qst===", line["question"]).replace("===ori===", response).replace("===err===", info)
if task == "TabMWP":
rectify_env = rectify_env.replace("===table===", line["table"])
if task == "Creation":
rectify_env = rectify_env.replace("===constants===", line["constant"].strip())
response = gen_func(rectify_env, sys_msg, temperature=temperature)
if "```" not in response:
response = "```python\n" + response + "\n```"
continue
elif not if_succ and time > rectify_limit:
print("!!! Times of execution error exceeds limit, skip !!!")
f = open(save_file, "a")
if task == "TabMWP":
f.write("### Table\n" + line["table"] + "\n")
f.write("### Question\n" + line["question"] + "\n### Respose\n" + response.strip() + "\n\n### Back info:\n" + info.strip() + f"\n\nExceed rectify limit!\nThe correct answer should be {correct_ans}")
f.write("\n\n=============================split case=============================\n\n")
f.close()
exec_err += 1
break
elif if_succ:
print("~~~ Runing successfully ~~~")
if "Final Answer:" in info:
model_ans = [info.split("Final Answer:")[1].strip()]
else:
cprint.info("Getting Answer by Directly Extracting Number ...")
model_ans = re.findall(r'-?\d+\.?\d*', info)
correct_flag = False
for ans in model_ans:
if grade_answer(str(ans), str(correct_ans)):
print("~~~ Correct Answer ~~~")
correct_flag = True
f = open(save_file, "a")
if task == "TabMWP":
f.write("### Table\n" + line["table"] + "\n")
f.write("### Question\n" + line["question"] + "\n### Respose\n" + response.strip() + "\n\n### Back info:\n" + info.strip() + f"\n\nCorrect Answer!\nThe correct answer should be {correct_ans}")
f.write("\n\n=============================split case=============================\n\n")
f.close()
correct += 1
if time == 1:
one_time_pass += 1
break
if not correct_flag:
print("!!! Wrong Answer !!!")
f = open(save_file, "a")
if task == "TabMWP":
f.write("### Table\n" + line["table"] + "\n")
f.write("### Question\n" + line["question"] + "\n### Respose\n" + response.strip() + "\n\n### Back info:\n" + info.strip() + f"\n\nWrong Answer!\nThe correct answer should be {correct_ans}")
f.write("\n\n=============================split case=============================\n\n")
f.close()
wrong += 1
break
# T_T you shouldn't reach here
break
print("Correct:", correct, "Wrong:", wrong, "Exec Error:", exec_err, "One Time Pass:", one_time_pass)
except:
exec_err += 1
f = open(save_file, "a")
total_num = correct + wrong + exec_err
f.write("Correct: " + str(correct) + " " + str(correct/total_num * 100) + "\nWrong: " + str(wrong) + " " + str(wrong/total_num * 100) + "\nExec Error: " + str(exec_err) + " " + str(exec_err/total_num * 100) + "\nOne Time Pass: " + str(one_time_pass) + "\nTotal: " + str(total_num) + "\n\n")
f.close()