-
Notifications
You must be signed in to change notification settings - Fork 462
/
GPT4Generator.py
174 lines (143 loc) · 6.77 KB
/
GPT4Generator.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
import asyncio
import os
from dotenv import load_dotenv
from collections.abc import Iterator
from goldenverba.components.generation.interface import Generator
load_dotenv()
class GPT4Generator(Generator):
"""
GPT4 Generator.
"""
def __init__(self):
super().__init__()
self.name = "GPT4Generator"
self.description = "Generator using OpenAI's GPT-4-1106-preview model"
self.requires_library = ["openai"]
self.requires_env = ["OPENAI_API_KEY"]
self.streamable = True
self.model_name = os.getenv("OPENAI_MODEL","gpt-4-1106-preview")
self.context_window = 10000
async def generate(
self,
queries: list[str],
context: list[str],
conversation: dict = None,
) -> str:
"""Generate an answer based on a list of queries and list of contexts, and includes conversational context
@parameter: queries : list[str] - List of queries
@parameter: context : list[str] - List of contexts
@parameter: conversation : dict - Conversational context
@returns str - Answer generated by the Generator.
"""
if conversation is None:
conversation = {}
messages = self.prepare_messages(queries, context, conversation)
try:
import openai
openai.api_key = os.getenv("OPENAI_API_KEY")
if "OPENAI_API_TYPE" in os.environ:
openai.api_type = os.getenv("OPENAI_API_TYPE")
if "OPENAI_API_BASE" in os.environ:
openai.api_base = os.getenv("OPENAI_API_BASE")
if "OPENAI_API_VERSION" in os.environ:
openai.api_version = os.getenv("OPENAI_API_VERSION")
chat_completion_arguments = {
"model":self.model_name,
"messages":messages
}
if openai.api_type=="azure":
chat_completion_arguments["deployment_id"]=self.model_name
base_url = os.environ.get("OPENAI_BASE_URL", "")
if base_url:
openai.api_base = base_url
completion = await asyncio.to_thread(
openai.ChatCompletion.create, **chat_completion_arguments
)
system_msg = str(completion["choices"][0]["message"]["content"])
except Exception:
raise
return system_msg
async def generate_stream(
self,
queries: list[str],
context: list[str],
conversation: dict = None,
) -> Iterator[dict]:
"""Generate a stream of response dicts based on a list of queries and list of contexts, and includes conversational context
@parameter: queries : list[str] - List of queries
@parameter: context : list[str] - List of contexts
@parameter: conversation : dict - Conversational context
@returns Iterator[dict] - Token response generated by the Generator in this format {system:TOKEN, finish_reason:stop or empty}.
"""
if conversation is None:
conversation = {}
messages = self.prepare_messages(queries, context, conversation)
try:
import openai
openai.api_key = os.getenv("OPENAI_API_KEY")
base_url = os.environ.get("OPENAI_BASE_URL", "")
if base_url:
openai.api_base = base_url
if "OPENAI_API_TYPE" in os.environ:
openai.api_type = os.getenv("OPENAI_API_TYPE")
if "OPENAI_API_BASE" in os.environ:
openai.api_base = os.getenv("OPENAI_API_BASE")
if "OPENAI_API_VERSION" in os.environ:
openai.api_version = os.getenv("OPENAI_API_VERSION")
chat_completion_arguments = {
"model":self.model_name,
"messages":messages,
"stream":True,
"temperature":0.0
}
if openai.api_type=="azure":
chat_completion_arguments["deployment_id"]=self.model_name
completion = await openai.ChatCompletion.acreate(
**chat_completion_arguments
)
try:
while True:
chunk = await completion.__anext__()
if len(chunk["choices"]) > 0:
if "content" in chunk["choices"][0]["delta"]:
yield {
"message": chunk["choices"][0]["delta"]["content"],
"finish_reason": chunk["choices"][0]["finish_reason"],
}
else:
yield {
"message": "",
"finish_reason": chunk["choices"][0]["finish_reason"],
}
except StopAsyncIteration:
pass
except Exception:
raise
def prepare_messages(
self, queries: list[str], context: list[str], conversation: dict[str, str]
) -> dict[str, str]:
"""
Prepares a list of messages formatted for a Retrieval Augmented Generation chatbot system, including system instructions, previous conversation, and a new user query with context.
@parameter queries: A list of strings representing the user queries to be answered.
@parameter context: A list of strings representing the context information provided for the queries.
@parameter conversation: A list of previous conversation messages that include the role and content.
@returns A list of message dictionaries formatted for the chatbot. This includes an initial system message, the previous conversation messages, and the new user query encapsulated with the provided context.
Each message in the list is a dictionary with 'role' and 'content' keys, where 'role' is either 'system' or 'user', and 'content' contains the relevant text. This will depend on the LLM used.
"""
messages = [
{
"role": "system",
"content": "You are a Retrieval Augmented Generation chatbot. Please answer user queries only their provided context. If the provided documentation does not provide enough information, say so. If the answer requires code examples encapsulate them with ```programming-language-name ```. Don't do pseudo-code.",
}
]
for message in conversation:
messages.append({"role": message.type, "content": message.content})
query = " ".join(queries)
user_context = " ".join(context)
messages.append(
{
"role": "user",
"content": f"Please answer this query: '{query}' with this provided context: {user_context}",
}
)
return messages