Husnain
commited on
Commit
•
f41554f
1
Parent(s):
e9f6b1e
💎 [Feature] Enable gpt-3.5 in chat_api
Browse files- networks/openai_streamer.py +219 -0
networks/openai_streamer.py
ADDED
@@ -0,0 +1,219 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import copy
|
2 |
+
import json
|
3 |
+
import re
|
4 |
+
import tiktoken
|
5 |
+
import uuid
|
6 |
+
|
7 |
+
from curl_cffi import requests
|
8 |
+
from tclogger import logger
|
9 |
+
|
10 |
+
from constants.envs import PROXIES
|
11 |
+
from constants.headers import OPENAI_GET_HEADERS, OPENAI_POST_DATA
|
12 |
+
from constants.models import TOKEN_LIMIT_MAP, TOKEN_RESERVED
|
13 |
+
|
14 |
+
from messagers.message_outputer import OpenaiStreamOutputer
|
15 |
+
|
16 |
+
|
17 |
+
class OpenaiRequester:
|
18 |
+
def __init__(self):
|
19 |
+
self.init_requests_params()
|
20 |
+
|
21 |
+
def init_requests_params(self):
|
22 |
+
self.api_base = "https://chat.openai.com/backend-anon"
|
23 |
+
self.api_me = f"{self.api_base}/me"
|
24 |
+
self.api_models = f"{self.api_base}/models"
|
25 |
+
self.api_chat_requirements = f"{self.api_base}/sentinel/chat-requirements"
|
26 |
+
self.api_conversation = f"{self.api_base}/conversation"
|
27 |
+
self.uuid = str(uuid.uuid4())
|
28 |
+
self.requests_headers = copy.deepcopy(OPENAI_GET_HEADERS)
|
29 |
+
extra_headers = {
|
30 |
+
"Oai-Device-Id": self.uuid,
|
31 |
+
}
|
32 |
+
self.requests_headers.update(extra_headers)
|
33 |
+
|
34 |
+
def log_request(self, url, method="GET"):
|
35 |
+
logger.note(f"> {method}:", end=" ")
|
36 |
+
logger.mesg(f"{url}", end=" ")
|
37 |
+
|
38 |
+
def log_response(self, res: requests.Response, stream=False, verbose=False):
|
39 |
+
status_code = res.status_code
|
40 |
+
status_code_str = f"[{status_code}]"
|
41 |
+
|
42 |
+
if status_code == 200:
|
43 |
+
logger_func = logger.success
|
44 |
+
else:
|
45 |
+
logger_func = logger.warn
|
46 |
+
|
47 |
+
logger_func(status_code_str)
|
48 |
+
|
49 |
+
if verbose:
|
50 |
+
if stream:
|
51 |
+
if not hasattr(self, "content_offset"):
|
52 |
+
self.content_offset = 0
|
53 |
+
|
54 |
+
for line in res.iter_lines():
|
55 |
+
line = line.decode("utf-8")
|
56 |
+
line = re.sub(r"^data:\s*", "", line)
|
57 |
+
if re.match(r"^\[DONE\]", line):
|
58 |
+
logger.success("\n[Finished]")
|
59 |
+
break
|
60 |
+
line = line.strip()
|
61 |
+
if line:
|
62 |
+
try:
|
63 |
+
data = json.loads(line, strict=False)
|
64 |
+
message_role = data["message"]["author"]["role"]
|
65 |
+
message_status = data["message"]["status"]
|
66 |
+
if (
|
67 |
+
message_role == "assistant"
|
68 |
+
and message_status == "in_progress"
|
69 |
+
):
|
70 |
+
content = data["message"]["content"]["parts"][0]
|
71 |
+
delta_content = content[self.content_offset :]
|
72 |
+
self.content_offset = len(content)
|
73 |
+
logger_func(delta_content, end="")
|
74 |
+
except Exception as e:
|
75 |
+
logger.warn(e)
|
76 |
+
else:
|
77 |
+
logger_func(res.json())
|
78 |
+
|
79 |
+
def get_models(self):
|
80 |
+
self.log_request(self.api_models)
|
81 |
+
res = requests.get(
|
82 |
+
self.api_models,
|
83 |
+
headers=self.requests_headers,
|
84 |
+
proxies=PROXIES,
|
85 |
+
timeout=10,
|
86 |
+
impersonate="chrome120",
|
87 |
+
)
|
88 |
+
self.log_response(res)
|
89 |
+
|
90 |
+
def auth(self):
|
91 |
+
self.log_request(self.api_chat_requirements, method="POST")
|
92 |
+
res = requests.post(
|
93 |
+
self.api_chat_requirements,
|
94 |
+
headers=self.requests_headers,
|
95 |
+
proxies=PROXIES,
|
96 |
+
timeout=10,
|
97 |
+
impersonate="chrome120",
|
98 |
+
)
|
99 |
+
self.chat_requirements_token = res.json()["token"]
|
100 |
+
self.log_response(res)
|
101 |
+
|
102 |
+
def transform_messages(self, messages: list[dict]):
|
103 |
+
def get_role(role):
|
104 |
+
if role in ["system", "user", "assistant"]:
|
105 |
+
return role
|
106 |
+
else:
|
107 |
+
return "system"
|
108 |
+
|
109 |
+
new_messages = [
|
110 |
+
{
|
111 |
+
"author": {"role": get_role(message["role"])},
|
112 |
+
"content": {"content_type": "text", "parts": [message["content"]]},
|
113 |
+
"metadata": {},
|
114 |
+
}
|
115 |
+
for message in messages
|
116 |
+
]
|
117 |
+
return new_messages
|
118 |
+
|
119 |
+
def chat_completions(self, messages: list[dict], verbose=False):
|
120 |
+
extra_headers = {
|
121 |
+
"Accept": "text/event-stream",
|
122 |
+
"Openai-Sentinel-Chat-Requirements-Token": self.chat_requirements_token,
|
123 |
+
}
|
124 |
+
requests_headers = copy.deepcopy(self.requests_headers)
|
125 |
+
requests_headers.update(extra_headers)
|
126 |
+
|
127 |
+
post_data = copy.deepcopy(OPENAI_POST_DATA)
|
128 |
+
extra_data = {
|
129 |
+
"messages": self.transform_messages(messages),
|
130 |
+
"websocket_request_id": str(uuid.uuid4()),
|
131 |
+
}
|
132 |
+
post_data.update(extra_data)
|
133 |
+
|
134 |
+
self.log_request(self.api_conversation, method="POST")
|
135 |
+
s = requests.Session()
|
136 |
+
res = s.post(
|
137 |
+
self.api_conversation,
|
138 |
+
headers=requests_headers,
|
139 |
+
json=post_data,
|
140 |
+
proxies=PROXIES,
|
141 |
+
timeout=10,
|
142 |
+
impersonate="chrome120",
|
143 |
+
stream=True,
|
144 |
+
)
|
145 |
+
if verbose:
|
146 |
+
self.log_response(res, stream=True, verbose=True)
|
147 |
+
return res
|
148 |
+
|
149 |
+
|
150 |
+
class OpenaiStreamer:
|
151 |
+
def __init__(self):
|
152 |
+
self.model = "gpt-3.5"
|
153 |
+
self.message_outputer = OpenaiStreamOutputer(owned_by="openai", model="gpt-3.5")
|
154 |
+
self.tokenizer = tiktoken.get_encoding("cl100k_base")
|
155 |
+
|
156 |
+
def count_tokens(self, messages: list[dict]):
|
157 |
+
token_count = sum(
|
158 |
+
len(self.tokenizer.encode(message["content"])) for message in messages
|
159 |
+
)
|
160 |
+
logger.note(f"Prompt Token Count: {token_count}")
|
161 |
+
return token_count
|
162 |
+
|
163 |
+
def check_token_limit(self, messages: list[dict]):
|
164 |
+
token_limit = TOKEN_LIMIT_MAP[self.model]
|
165 |
+
token_redundancy = int(
|
166 |
+
token_limit - TOKEN_RESERVED - self.count_tokens(messages)
|
167 |
+
)
|
168 |
+
if token_redundancy <= 0:
|
169 |
+
raise ValueError(f"Prompt exceeded token limit: {token_limit}")
|
170 |
+
return True
|
171 |
+
|
172 |
+
def chat_response(self, messages: list[dict]):
|
173 |
+
self.check_token_limit(messages)
|
174 |
+
requester = OpenaiRequester()
|
175 |
+
requester.auth()
|
176 |
+
return requester.chat_completions(messages, verbose=False)
|
177 |
+
|
178 |
+
def chat_return_generator(self, stream_response: requests.Response):
|
179 |
+
content_offset = 0
|
180 |
+
is_finished = False
|
181 |
+
|
182 |
+
for line in stream_response.iter_lines():
|
183 |
+
line = line.decode("utf-8")
|
184 |
+
line = re.sub(r"^data:\s*", "", line)
|
185 |
+
line = line.strip()
|
186 |
+
|
187 |
+
if not line:
|
188 |
+
continue
|
189 |
+
|
190 |
+
if re.match(r"^\[DONE\]", line):
|
191 |
+
content_type = "Finished"
|
192 |
+
delta_content = ""
|
193 |
+
logger.success("\n[Finished]")
|
194 |
+
is_finished = True
|
195 |
+
else:
|
196 |
+
content_type = "Completions"
|
197 |
+
try:
|
198 |
+
data = json.loads(line, strict=False)
|
199 |
+
message_role = data["message"]["author"]["role"]
|
200 |
+
message_status = data["message"]["status"]
|
201 |
+
if message_role == "assistant" and message_status == "in_progress":
|
202 |
+
content = data["message"]["content"]["parts"][0]
|
203 |
+
if not len(content):
|
204 |
+
continue
|
205 |
+
delta_content = content[content_offset:]
|
206 |
+
content_offset = len(content)
|
207 |
+
logger.success(delta_content, end="")
|
208 |
+
else:
|
209 |
+
continue
|
210 |
+
except Exception as e:
|
211 |
+
logger.warn(e)
|
212 |
+
|
213 |
+
output = self.message_outputer.output(
|
214 |
+
content=delta_content, content_type=content_type
|
215 |
+
)
|
216 |
+
yield output
|
217 |
+
|
218 |
+
if not is_finished:
|
219 |
+
yield self.message_outputer.output(content="", content_type="Finished")
|