Spaces:
Runtime error
Runtime error
first
Browse files- app.py +55 -0
- utils/__init__.py +1 -0
- utils/__pycache__/__init__.cpython-310.pyc +0 -0
- utils/__pycache__/bot.cpython-310.pyc +0 -0
- utils/bot.py +168 -0
app.py
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from utils import Bot
|
3 |
+
|
4 |
+
# Define the initialization function for the bot
|
5 |
+
def initialize_bot(openai_api_key, table_descriptions, text_documents):
|
6 |
+
global bot
|
7 |
+
bot = Bot(openai_api_key=openai_api_key, table_descriptions=table_descriptions, text_documents=text_documents, verbose=True)
|
8 |
+
|
9 |
+
# Define the function to generate the bot's response
|
10 |
+
def generate_response(question):
|
11 |
+
response = bot(question)
|
12 |
+
return response
|
13 |
+
|
14 |
+
# Define the interface components
|
15 |
+
openai_api_key_input = gr.inputs.Textbox(label="OpenAI API Key")
|
16 |
+
table_descriptions_input = gr.inputs.Textbox(label="Table Descriptions")
|
17 |
+
text_documents_input = gr.inputs.Textbox(label="Text Documents")
|
18 |
+
|
19 |
+
submit_button = gr.outputs.Button(label="Submit")
|
20 |
+
|
21 |
+
question_input = gr.inputs.Textbox(label="Ask a question")
|
22 |
+
submit_question_button = gr.outputs.Button(label="Submit")
|
23 |
+
|
24 |
+
# Define the interface layout
|
25 |
+
input_components = [
|
26 |
+
openai_api_key_input,
|
27 |
+
table_descriptions_input,
|
28 |
+
text_documents_input,
|
29 |
+
submit_button,
|
30 |
+
]
|
31 |
+
|
32 |
+
output_components = [
|
33 |
+
question_input,
|
34 |
+
submit_question_button,
|
35 |
+
]
|
36 |
+
|
37 |
+
interface = gr.Interface(
|
38 |
+
initialize_bot,
|
39 |
+
input_components,
|
40 |
+
show_input=False,
|
41 |
+
outputs=output_components,
|
42 |
+
title="Chat with your Bot",
|
43 |
+
)
|
44 |
+
|
45 |
+
# Define the callback function for the submit button
|
46 |
+
def submit_callback(openai_api_key, table_descriptions, text_documents):
|
47 |
+
initialize_bot(openai_api_key, table_descriptions, text_documents)
|
48 |
+
|
49 |
+
# Define the callback function for the submit question button
|
50 |
+
def submit_question_callback(question):
|
51 |
+
response = generate_response(question)
|
52 |
+
return response
|
53 |
+
|
54 |
+
# Add the callbacks to the interface
|
55 |
+
interface.launch(inline=False, submit_button=submit_callback, submit_question_button=submit_question_callback)
|
utils/__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
from .bot import Bot
|
utils/__pycache__/__init__.cpython-310.pyc
ADDED
Binary file (166 Bytes). View file
|
|
utils/__pycache__/bot.cpython-310.pyc
ADDED
Binary file (5.72 kB). View file
|
|
utils/bot.py
ADDED
@@ -0,0 +1,168 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import langchain
|
2 |
+
from langchain.agents import create_csv_agent
|
3 |
+
from langchain.schema import HumanMessage
|
4 |
+
from langchain.chat_models import ChatOpenAI
|
5 |
+
from langchain.embeddings import OpenAIEmbeddings
|
6 |
+
from langchain.vectorstores import Chroma
|
7 |
+
from typing import List, Dict
|
8 |
+
from langchain.agents import AgentType
|
9 |
+
from langchain.chains.conversation.memory import ConversationBufferWindowMemory
|
10 |
+
|
11 |
+
|
12 |
+
class Bot:
|
13 |
+
|
14 |
+
def __init__(
|
15 |
+
self,
|
16 |
+
openai_api_key: str,
|
17 |
+
table_descriptions: List[Dict[str, any]],
|
18 |
+
text_documents: List[langchain.schema.Document],
|
19 |
+
verbose: bool = False
|
20 |
+
):
|
21 |
+
self.verbose = verbose
|
22 |
+
self.table_descriptions = table_descriptions
|
23 |
+
|
24 |
+
self.llm = ChatOpenAI(
|
25 |
+
openai_api_key=openai_api_key,
|
26 |
+
temperature=0,
|
27 |
+
model_name="gpt-3.5-turbo"
|
28 |
+
)
|
29 |
+
|
30 |
+
embeddings = OpenAIEmbeddings(openai_api_key=openai_api_key)
|
31 |
+
vector_store = Chroma.from_documents(text_documents, embeddings)
|
32 |
+
self.text_retriever = langchain.chains.RetrievalQAWithSourcesChain.from_chain_type(
|
33 |
+
llm=self.llm,
|
34 |
+
chain_type='stuff',
|
35 |
+
retriever=vector_store.as_retriever()
|
36 |
+
)
|
37 |
+
self.text_search_tool = langchain.agents.Tool(
|
38 |
+
func=self._text_search,
|
39 |
+
description="Use this tool when searching for text information",
|
40 |
+
name="search text information"
|
41 |
+
)
|
42 |
+
|
43 |
+
def __call__(
|
44 |
+
self,
|
45 |
+
question: str
|
46 |
+
):
|
47 |
+
self.tools = []
|
48 |
+
self.tools.append(self.text_search_tool)
|
49 |
+
table = self._define_appropriate_table(question)
|
50 |
+
if table != "None of the tables":
|
51 |
+
number = int(table[table.find('№')+1:])
|
52 |
+
table_description = [x for x in self.table_descriptions if x['number'] == number][0]
|
53 |
+
table_path = table_description['path']
|
54 |
+
|
55 |
+
self.csv_agent = create_csv_agent(
|
56 |
+
llm=self.llm,
|
57 |
+
path=table_path,
|
58 |
+
verbose=self.verbose
|
59 |
+
)
|
60 |
+
|
61 |
+
self._init_tabular_search_tool(table_description)
|
62 |
+
self.tools.append(self.tabular_search_tool)
|
63 |
+
|
64 |
+
self._init_chatbot()
|
65 |
+
print(table)
|
66 |
+
response = self.agent(question)
|
67 |
+
return response
|
68 |
+
|
69 |
+
def _init_chatbot(self):
|
70 |
+
|
71 |
+
conversational_memory = ConversationBufferWindowMemory(
|
72 |
+
memory_key='chat_history',
|
73 |
+
k=5,
|
74 |
+
return_messages=True
|
75 |
+
)
|
76 |
+
|
77 |
+
self.agent = langchain.agents.initialize_agent(
|
78 |
+
agent=AgentType.CHAT_CONVERSATIONAL_REACT_DESCRIPTION,
|
79 |
+
tools=self.tools,
|
80 |
+
llm=self.llm,
|
81 |
+
verbose=self.verbose,
|
82 |
+
max_iterations=5,
|
83 |
+
early_stopping_method='generate',
|
84 |
+
memory=conversational_memory
|
85 |
+
)
|
86 |
+
sys_msg = (
|
87 |
+
"You are an expert summarizer and deliverer of information. "
|
88 |
+
"Yet, the reason you are so intelligent is that you make complex "
|
89 |
+
"information incredibly simple to understand. It's actually rather incredible."
|
90 |
+
"When users ask information you refer to the relevant tools."
|
91 |
+
"if one of the tools helped you with only a part of the necessary information, you must "
|
92 |
+
"try to find the missing information using another tool"
|
93 |
+
"if you can't find the information using the provided tools, you MUST "
|
94 |
+
"say 'I don't know'. Don't try to make up an answer."
|
95 |
+
)
|
96 |
+
prompt = self.agent.agent.create_prompt(
|
97 |
+
system_message=sys_msg,
|
98 |
+
tools=self.tools
|
99 |
+
)
|
100 |
+
self.agent.agent.llm_chain.prompt = prompt
|
101 |
+
|
102 |
+
def _text_search(
|
103 |
+
self,
|
104 |
+
query: str
|
105 |
+
) -> str:
|
106 |
+
query = self.text_retriever.prep_inputs(query)
|
107 |
+
res = self.text_retriever(query)['answer']
|
108 |
+
return res
|
109 |
+
|
110 |
+
def _tabular_search(
|
111 |
+
self,
|
112 |
+
query: str
|
113 |
+
) -> str:
|
114 |
+
res = self.csv_agent.run(query)
|
115 |
+
return res
|
116 |
+
|
117 |
+
def _init_tabular_search_tool(
|
118 |
+
self,
|
119 |
+
table_description: Dict[str, any]
|
120 |
+
) -> None:
|
121 |
+
|
122 |
+
columns = table_description["columns"]
|
123 |
+
columns = '"' + '", "'.join(columns) + '"'
|
124 |
+
tittle = table_description["tittle"]
|
125 |
+
|
126 |
+
description = f"""
|
127 |
+
Use this tool when searching for tabular information.
|
128 |
+
With this tool you could get access to table.
|
129 |
+
This table tittle is "{tittle}" and the names of the columns in this table: {columns}
|
130 |
+
"""
|
131 |
+
|
132 |
+
self.tabular_search_tool = langchain.agents.Tool(
|
133 |
+
func=self._tabular_search,
|
134 |
+
description=description,
|
135 |
+
name="search tabular information"
|
136 |
+
)
|
137 |
+
|
138 |
+
def _define_appropriate_table(
|
139 |
+
self,
|
140 |
+
question: str
|
141 |
+
) -> str:
|
142 |
+
''' Определяет по описаниям таблиц в какой из них может содержаться ответ на во��рос.
|
143 |
+
Возвращает номер таблицы по шаблону "Table №1" или "None of the tables" '''
|
144 |
+
|
145 |
+
message = 'I have list of table descriptions: \n'
|
146 |
+
k = 0
|
147 |
+
for description in self.table_descriptions:
|
148 |
+
k += 1
|
149 |
+
number = description["number"]
|
150 |
+
columns = description["columns"]
|
151 |
+
columns = '"' + '", "'.join(columns) + '"'
|
152 |
+
tittle = description["tittle"]
|
153 |
+
str_description = f""" {k}) description for Table №{number}:
|
154 |
+
a) table consist of columns with names: {columns};
|
155 |
+
b) table tittle: {tittle}.\n"""
|
156 |
+
message += str_description
|
157 |
+
|
158 |
+
question = f""" How do you think, which table can help answer the question: "{question}" .
|
159 |
+
Your answer MUST be specific,
|
160 |
+
for example if you think that Table №2 can help answer the question, you MUST just write "Table №2".
|
161 |
+
If you think that none of the tables can help answer the question just write "None of the tables"
|
162 |
+
Don't include to answer information about your thinking.
|
163 |
+
"""
|
164 |
+
message += question
|
165 |
+
|
166 |
+
res = self.llm([HumanMessage(content=message)])
|
167 |
+
return res.content[:-1]
|
168 |
+
|