-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathICL_gpt4.py
172 lines (154 loc) · 8 KB
/
ICL_gpt4.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
'''
This file is for evaluating the in-context learning performance of GPT3.5 and GPT4 on Concept-1K
'''
from openai import OpenAI
import json
import os
import random
import numpy as np
seed = 20240210
random.seed(seed)
client = OpenAI(api_key="")
role = "You're a helpful assistant."
# ============================================================================
prompt_1shot = "I will provide some knowledge as follows:\n \
Question: {0}\n Short Answer: {1}\n \
Please answer the following question according to the above knowledge:\n \
Question: {2}\n Short Answer: "
prompt_5shot = "I will provide some knowledge as follows:\n \
Question: {0}\n Short Answer: {1}\n \
Question: {2}\n Short Answer: {3}\n \
Question: {4}\n Short Answer: {5}\n \
Question: {6}\n Short Answer: {7}\n \
Question: {8}\n Short Answer: {9}\n \
Please answer the following question according to the above knowledge:\n \
Question: {10}\n Short Answer: "
# ============================================================================
n_subsample = 500
# model_name = 'gpt-3.5-turbo' # gpt-4-turbo-preview, gpt-4, gpt-3.5-turbo
# shot = 1
# method = 'same_concept' # same_instance, same_concept, rand
for model_name in ['gpt-3.5-turbo','gpt-4-turbo-preview']:
for shot in [1,5]:
for method in ['rand','same_concept']:
if model_name == 'gpt-3.5-turbo' and shot == 1 and method=='rand':
continue
save_name = '%dshot_%s'%(shot,method)
cur_dir = os.path.dirname(__file__)
with open(os.path.join(cur_dir,'dataset/concept_1k_task1/continual_data.json'), 'r', encoding='utf-8') as f:
data = json.load(f)
train_data = data['0']['train']
test_data = data['0']['test']
select_index = list(range(len(train_data['input'])))
random.shuffle(select_index)
select_index = select_index[:n_subsample]
cnt = 0
acc_list = []
for test_sample_id in select_index:
if shot == 1:
if method == 'same_instance':
prompted_test_input = prompt_1shot.format(
train_data['input'][test_sample_id],
train_data['target'][test_sample_id],
test_data['input'][test_sample_id],
)
elif method == 'same_concept':
same_concept_idx = np.where(np.array(train_data['concept_id'])==test_data['concept_id'][test_sample_id])[0]
random_index = np.random.choice(same_concept_idx,size=1)
prompted_test_input = prompt_1shot.format(
train_data['input'][random_index[0]],
train_data['target'][random_index[0]],
test_data['input'][test_sample_id],
)
elif method == 'rand':
random_index = np.random.choice(list(range(len(train_data['input']))),size=1)
prompted_test_input = prompt_1shot.format(
train_data['input'][random_index[0]],
train_data['target'][random_index[0]],
test_data['input'][test_sample_id],
)
else:
raise NotImplementedError()
elif shot==5:
if method == 'same_instance':
random_index = np.random.choice(list(range(len(train_data['input']))),size=4)
prompted_test_input = prompt_5shot.format(
train_data['input'][random_index[0]],
train_data['target'][random_index[0]],
train_data['input'][random_index[1]],
train_data['target'][random_index[1]],
train_data['input'][random_index[2]],
train_data['target'][random_index[2]],
train_data['input'][random_index[3]],
train_data['target'][random_index[3]],
train_data['input'][test_sample_id],
train_data['target'][test_sample_id],
test_data['input'][test_sample_id],
)
elif method == 'same_concept':
same_concept_idx = np.where(np.array(train_data['concept_id'])==test_data['concept_id'][test_sample_id])[0]
random_index = np.random.choice(same_concept_idx,size=5)
prompted_test_input = prompt_5shot.format(
train_data['input'][random_index[0]],
train_data['target'][random_index[0]],
train_data['input'][random_index[1]],
train_data['target'][random_index[1]],
train_data['input'][random_index[2]],
train_data['target'][random_index[2]],
train_data['input'][random_index[3]],
train_data['target'][random_index[3]],
train_data['input'][random_index[4]],
train_data['target'][random_index[4]],
test_data['input'][test_sample_id],
)
elif method == 'rand':
random_index = np.random.choice(list(range(len(train_data['input']))),size=5)
prompted_test_input = prompt_5shot.format(
train_data['input'][random_index[0]],
train_data['target'][random_index[0]],
train_data['input'][random_index[1]],
train_data['target'][random_index[1]],
train_data['input'][random_index[2]],
train_data['target'][random_index[2]],
train_data['input'][random_index[3]],
train_data['target'][random_index[3]],
train_data['input'][random_index[4]],
train_data['target'][random_index[4]],
test_data['input'][test_sample_id],
)
else:
raise NotImplementedError()
else:
raise NotImplementedError()
test_target = test_data['target'][test_sample_id]
completion = client.chat.completions.create(
model=model_name,
messages=[
{"role": "system", "content": role},
{"role": "user", "content": prompted_test_input}
]
)
response = completion.choices[0].message.content
response = response.strip().lower()
test_target = test_target.strip().lower()
acc_list.append(1.0 if response==test_target else 0.0)
cnt += 1
print('Progress %.2f%%[=%d/%d]: Test_ID=%s, Target=%s, Predict=%s, tmp_ACC=%.2f%%'%(
cnt/len(select_index)*100,
cnt,
len(select_index),
test_sample_id,
test_target,
response,
np.mean(acc_list)*100,
))
with open(os.path.join(cur_dir,'./seed%d_%s_%s.txt'%(seed,model_name,save_name)), 'a') as file:
file.write('Progress %.2f%%[=%d/%d]: Test_ID=%s, Target=%s, Predict=%s, tmp_ACC=%.2f%%\n'%(
cnt/len(select_index)*100,
cnt,
len(select_index),
test_sample_id,
test_target,
response,
np.mean(acc_list)*100,
))