-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathapproach-cooperate.py
180 lines (156 loc) · 8.68 KB
/
approach-cooperate.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
import json
import argparse
import lm_utils
import metrics
import random
from tqdm import tqdm
if __name__ == "__main__":
argParser = argparse.ArgumentParser()
argParser.add_argument("-m", "--model", help="which language model to use: \"mistral\", \"llama2_7/13/70b\", \"chatgpt\"")
argParser.add_argument("-d", "--dataset", help="which dataset in data/: \"mmlu\", \"knowledge_crosswords\", \"hellaswag\", \"propaganda\", \"ambigqa\", \"electionqa23\"")
argParser.add_argument("-o", "--portion", default = 1.0, help="portion of the dataset to use")
argParser.add_argument("-t", "--type", help="approach type, self or others") # "self", "others"
args = argParser.parse_args()
model_name = args.model
dataset = args.dataset
approach_type = args.type
portion = args.portion
lm_utils.llm_init(model_name)
correct_flags = []
abstain_flags = []
abstain_scores = []
with open("data/" + dataset + ".json", "r") as f:
data = json.load(f)
data["dev"] = data["dev"][:int(len(data["dev"])*float(portion))]
data["test"] = data["test"][:int(len(data["test"])*float(portion))]
answers = []
feedback_1 = []
feedback_2 = []
feedback_3 = []
# obtain correct flags
for d in tqdm(data["test"]):
original_prompt = "Question: " + d["question"] + "\n"
for key in d["choices"].keys():
original_prompt += (key + ": " + d["choices"][key] + "\n")
original_prompt += "Choose one answer from the above choices. The answer is"
response = lm_utils.llm_response(original_prompt, model_name, probs=False)
# print(response)
# print(lm_utils.answer_parsing(response))
if lm_utils.answer_parsing(response) == d["answer"]:
correct_flags.append(1)
else:
correct_flags.append(0)
answers.append(response)
# obtain feedbacks
prompt_feedback_list = []
for d, i in tqdm(zip(data["test"], range(len(data["test"])))):
original_prompt = "Question: " + d["question"] + "\n"
for key in d["choices"].keys():
original_prompt += (key + ": " + d["choices"][key] + "\n")
original_prompt += "Choose one answer from the above choices. The answer is"
prompt_feedback = original_prompt + " " + answers[i].strip() + "\nPlease review the proposed answer and provide feedback on its correctness.\nFeedback:"
prompt_feedback_list.append(prompt_feedback)
if approach_type == "self": # expert reviewers with self-specialized domains
for prompt_feedback in tqdm(prompt_feedback_list):
# generate knowledge from different expert's perspectives: facts, multi-hop, and commonsense
prompt_feedback_experts = []
for domain_name in ["factual information", "multi-hop reasoning", "commonsense knowledge"]:
expert_prompt = "Generate some knowledge about the question, focusing on " + domain_name + ".\n" + prompt_feedback.split("\n")[0] + "\nKnowledge:"
prompt_feedback_experts.append("Knowledge: " + lm_utils.llm_response(expert_prompt, model_name, probs=False, temperature=1).split("\n")[0].strip() + "\n" + prompt_feedback)
assert len(prompt_feedback_experts) == 3
response = lm_utils.llm_response(prompt_feedback_experts[0], model_name, probs=False, temperature=1)
feedback_1.append(response.split("\n")[0].strip())
response = lm_utils.llm_response(prompt_feedback_experts[1], model_name, probs=False, temperature=1)
feedback_2.append(response.split("\n")[0].strip())
response = lm_utils.llm_response(prompt_feedback_experts[2], model_name, probs=False, temperature=1)
feedback_3.append(response.split("\n")[0].strip())
elif approach_type == "others":
models_list = ["mistral", "llama2_70b", "chatgpt"]
# current model
for prompt_feedback in tqdm(prompt_feedback_list):
response = lm_utils.llm_response(prompt_feedback, model_name, probs=False, temperature=1)
feedback_1.append(response)
models_list.remove(model_name)
# other models
model_now = models_list[0]
lm_utils.wipe_model()
lm_utils.llm_init(model_now)
for prompt_feedback in tqdm(prompt_feedback_list):
response = lm_utils.llm_response(prompt_feedback, model_now, probs=False, temperature=1)
feedback_2.append(response)
model_now = models_list[1]
lm_utils.wipe_model()
lm_utils.llm_init(model_now)
for prompt_feedback in tqdm(prompt_feedback_list):
response = lm_utils.llm_response(prompt_feedback, model_now, probs=False, temperature=1)
feedback_3.append(response)
# obtain abstain flags and scores
prompt_area_chair_list = []
assert len(data["test"]) == len(answers) == len(feedback_1) == len(feedback_2) == len(feedback_3)
for i in range(len(data["test"])):
d = data["test"][i]
prompt_area_chair = "Question: " + d["question"] + "\n"
for key in d["choices"].keys():
prompt_area_chair += (key + ": " + d["choices"][key] + "\n")
prompt_area_chair += "Choose one answer from the above choices. The answer is " + answers[i].strip() + "\n\nFeedback 1: " + feedback_1[i].strip() + "\n\nFeedback 2: " + feedback_2[i].strip() + "\n\nFeedback 3: " + feedback_3[i].strip() + "\n\nBased on the feedback, the proposed answer is:\nA. True\nB. False\nThe answer is"
prompt_area_chair_list.append(prompt_area_chair)
# print(prompt_area_chair)
# print("--------------------")
assert len(prompt_area_chair_list) == len(data["test"])
if approach_type == "self":
for prompt_area_chair in tqdm(prompt_area_chair_list):
response, probs = lm_utils.llm_response(prompt_area_chair, model_name, probs=True)
if lm_utils.answer_parsing(response) == "A":
abstain_flags.append(0)
elif lm_utils.answer_parsing(response) == "B":
abstain_flags.append(1)
else:
print("Error: abstain flag not found")
abstain_flags.append(random.randint(0, 1))
try:
if abstain_flags[-1] == 0:
if "A" in probs.keys():
abstain_scores.append(probs["A"])
elif " A" in probs.keys():
abstain_scores.append(probs[" A"])
else:
if "B" in probs.keys():
abstain_scores.append(probs["B"])
elif " B" in probs.keys():
abstain_scores.append(probs[" B"])
except:
abstain_scores.append(0.5)
elif approach_type == "others":
lm_utils.wipe_model()
lm_utils.llm_init("chatgpt") # always area-charing with chatgpt
for prompt_area_chair in tqdm(prompt_area_chair_list):
response, probs = lm_utils.llm_response(prompt_area_chair, "chatgpt", probs=True)
if lm_utils.answer_parsing(response) == "A":
abstain_flags.append(0)
elif lm_utils.answer_parsing(response) == "B":
abstain_flags.append(1)
else:
print("Error: abstain flag not found")
abstain_flags.append(random.randint(0, 1))
# print(probs)
try:
if abstain_flags[-1] == 0:
if "A" in probs.keys():
abstain_scores.append(probs["A"])
elif " A" in probs.keys():
abstain_scores.append(probs[" A"])
else:
if "B" in probs.keys():
abstain_scores.append(probs["B"])
elif " B" in probs.keys():
abstain_scores.append(probs[" B"])
except:
abstain_scores.append(0.5)
# print(abstain_scores)
print("------------------")
print("Approach: Cooperate")
print("Model:", model_name)
print("Dataset:", dataset)
print("Type:", approach_type)
print(metrics.compute_metrics(correct_flags, abstain_flags, abstain_scores))
print("------------------")