-
Notifications
You must be signed in to change notification settings - Fork 8
/
llm_exp_long.py
282 lines (251 loc) · 13.3 KB
/
llm_exp_long.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
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
import os
import json
import requests, time
import re
import string
import collections
from tqdm import tqdm
api_key = ''
api_url = 'http://localhost:8000/v1/completions'
api_model_name = 'longchat-13b-16k' # 'gpt-3.5-turbo-16k'
headers={
# 'Authorization': f'Bearer {api_key}',
'Authorization': f'{api_key}',
'Content-Type': 'application/json'
}
def get_response(tgt):
data={
"model": api_model_name,
# "messages":[
# {
# "role":"user",
# "content":tgt
# }
# ],
"prompt": tgt,
"max_tokens": 50,
"temperature": 0.5,
}
response = requests.post(api_url, data=json.dumps(data), headers=headers)
if response.status_code == 200:
result = response.json()
else:
print(f"Error: {response.status_code} - {response.text}")
result = 'error'
return result
dataset_type = 'hotpot'
llm_type = 'open_source' # open_source or openai
examples_num = 3 # few-shots
output_file_path = f"example_{dataset_type}_{llm_type}.jsonl"
output_file = open(output_file_path, 'w')
with open(f'prompts/few-shots_{dataset_type}.txt') as f:
prompt_template_few = f.read().rstrip("\n")
with open(f'prompts/few-shots_{dataset_type}_br.txt') as f:
prompt_template_few_br = f.read().rstrip("\n")
if dataset_type == 'hotpot':
pred_data = json.load(open('hotpot_pred_best_retr.json'))
dev_data = json.load(open('data/hotpotqa/hotpot_dev_distractor_v1.json'))
test_subsampled_data = open('data/processed_data/hotpotqa/test_subsampled.jsonl').readlines()
test_subsampled_data = [json.loads(item) for item in test_subsampled_data]
test_subsampled_data_ids = [item['question_id'] for item in test_subsampled_data]
test_subsampled_data = [item for item in dev_data if item['_id'] in test_subsampled_data_ids]
for item in tqdm(test_subsampled_data):
formatted_documents, formatted_documents_br = [], []
question = item['question']
documents, documents_br = [], []
pred = pred_data[item['_id']]
for idx, (title, sents) in enumerate(item['context']):
document = {'title': title, 'text': "".join(sents)}
documents.append(document)
if idx in pred:
documents_br.append(document)
for document_index, document in enumerate(documents):
formatted_documents.append(f"Document [{document_index+1}](Title: {document['title']}) {document['text']}")
for document_index, document in enumerate(documents_br):
formatted_documents_br.append(f"Document [{document_index+1}](Title: {document['title']}) {document['text']}")
content = prompt_template_few.format(search_results="\n".join(formatted_documents), question=question)
content_br = prompt_template_few_br.format(search_results="\n".join(formatted_documents_br), question=question)
input = content
try:
response = get_response(input)
if response != 'error':
print("prompt tokens num:", response['usage']['prompt_tokens'])
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
else:
print('error wait 20 seconds...')
time.sleep(20)
response = get_response(input)
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
except Exception as e:
print(e)
print('wait 20 seconds...')
time.sleep(20)
response = get_response(input)
if response != 'error':
print("prompt tokens num:", response['usage']['prompt_tokens'])
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
print(item['question'])
print(item['answer'])
print(response)
input = content_br
try:
response2 = get_response(input)
if response2 != 'error':
print("prompt tokens num:", response2['usage']['prompt_tokens'])
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
else:
print('error wait 20 seconds...')
time.sleep(20)
response2 = get_response(input)
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
except Exception as e:
print(e)
print('wait 20 seconds...')
time.sleep(20)
response2 = get_response(input)
if response2 != 'error':
print("prompt tokens num:", response2['usage']['prompt_tokens'])
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
print(response2)
print()
output_file.write(json.dumps({'id': item['_id'], 'response': response, 'response2': response2}))
output_file.write('\n')
elif dataset_type == 'musique':
retr_url = 'musique_pred_best.json'
dev_data = open('data/musique_ans_v1.0_dev.jsonl').readlines()
dev_data = [json.loads(item) for item in dev_data]
pred_data = json.load(open(retr_url))
test_subsampled_data = open('data/processed_data/musique/test_subsampled.jsonl').readlines()
test_subsampled_data = [json.loads(item) for item in test_subsampled_data]
# test_subsampled_data_ids = [item['question_id'] for item in test_subsampled_data]
# test_subsampled_data = [item for item in dev_data if item['id'] in test_subsampled_data_ids]
for item in tqdm(test_subsampled_data):
formatted_documents, formatted_documents_br = [], []
question = item['question_text']
documents, documents_br = [], []
pred = pred_data[item['question_id']]
for idx, para in enumerate(item['contexts']):
document = {'title': para['title'], 'text': para['paragraph_text']}
documents.append(document)
if idx in pred:
documents_br.append(document)
for document_index, document in enumerate(documents):
formatted_documents.append(f"Document [{document_index+1}](Title: {document['title']}) {document['text']}")
for document_index, document in enumerate(documents_br):
formatted_documents_br.append(f"Document [{document_index+1}](Title: {document['title']}) {document['text']}")
content = prompt_template_few.format(search_results="\n".join(formatted_documents), question=question)
content_br = prompt_template_few_br.format(search_results="\n".join(formatted_documents_br), question=question)
input = content
try:
response = get_response(input)
if response != 'error':
print("prompt tokens num:", response['usage']['prompt_tokens'])
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
else:
print('error wait 20 seconds...')
time.sleep(20)
response = get_response(input)
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
except Exception as e:
print(e)
print('wait 20 seconds...')
time.sleep(20)
response = get_response(input)
if response != 'error':
print("prompt tokens num:", response['usage']['prompt_tokens'])
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
print(item['question_id'])
print(item['answers_objects'][0]['spans'][0])
print(response)
input = content_br
try:
response2 = get_response(input)
if response2 != 'error':
print("prompt tokens num:", response2['usage']['prompt_tokens'])
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
else:
print('error wait 20 seconds...')
time.sleep(20)
response2 = get_response(input)
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
except Exception as e:
print(e)
print('wait 20 seconds...')
time.sleep(20)
response2 = get_response(input)
if response2 != 'error':
print("prompt tokens num:", response2['usage']['prompt_tokens'])
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
print(response2)
print()
output_file.write(json.dumps({'id': item['question_id'], 'response': response, 'response2': response2}))
output_file.write('\n')
elif dataset_type == '2wiki':
pred_data = json.load(open('2wiki_pred_best.json'))
dev_data = json.load(open('data/2wiki/data_ids/dev.json'))
test_subsampled_data = open('data/processed_data/2wikimultihopqa/test_subsampled.jsonl').readlines()
test_subsampled_data = [json.loads(item) for item in test_subsampled_data]
test_subsampled_data_ids = [item['question_id'] for item in test_subsampled_data]
test_subsampled_data = [item for item in dev_data if item['_id'] in test_subsampled_data_ids]
for item in tqdm(test_subsampled_data):
formatted_documents, formatted_documents_br = [], []
question = item['question']
documents, documents_br = [], []
pred = pred_data[item['_id']]
for idx, (title, sents) in enumerate(item['context']):
document = {'title': title, 'text': "".join(sents)}
documents.append(document)
if idx in pred:
documents_br.append(document)
for document_index, document in enumerate(documents):
formatted_documents.append(f"Document [{document_index+1}](Title: {document['title']}) {document['text']}")
for document_index, document in enumerate(documents_br):
formatted_documents_br.append(f"Document [{document_index+1}](Title: {document['title']}) {document['text']}")
content = prompt_template_few.format(search_results="\n".join(formatted_documents), question=question)
content_br = prompt_template_few_br.format(search_results="\n".join(formatted_documents_br), question=question)
input = content
try:
response = get_response(input)
if response != 'error':
print("prompt tokens num:", response['usage']['prompt_tokens'])
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
else:
print('error wait 20 seconds...')
time.sleep(20)
response = get_response(input)
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
except Exception as e:
print(e)
print('wait 20 seconds...')
time.sleep(20)
response = get_response(input)
if response != 'error':
print("prompt tokens num:", response['usage']['prompt_tokens'])
response = response['choices'][0]['message']['content'] if 'message' in response['choices'][0] else response['choices'][0]['text']
print(item['question'])
print(item['answer'])
print(response)
input = content_br
try:
response2 = get_response(input)
if response2 != 'error':
print("prompt tokens num:", response2['usage']['prompt_tokens'])
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
else:
print('error wait 20 seconds...')
time.sleep(20)
response2 = get_response(input)
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
except Exception as e:
print(e)
print('wait 20 seconds...')
time.sleep(20)
response2 = get_response(input)
if response2 != 'error':
print("prompt tokens num:", response2['usage']['prompt_tokens'])
response2 = response2['choices'][0]['message']['content'] if 'message' in response2['choices'][0] else response2['choices'][0]['text']
print(response2)
print()
output_file.write(json.dumps({'id': item['_id'], 'response': response, 'response2': response2}))
output_file.write('\n')
output_file.close()