Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,123 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from config.globals import SPEAKER_TYPES, initial_prompt
|
3 |
+
from langchain.prompts import PromptTemplate
|
4 |
+
from langchain.chains.question_answering import load_qa_chain
|
5 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
6 |
+
from langchain_community.vectorstores import Chroma
|
7 |
+
from langchain_google_genai import GoogleGenerativeAIEmbeddings, ChatGoogleGenerativeAI
|
8 |
+
from dotenv import load_dotenv
|
9 |
+
import PyPDF2
|
10 |
+
import os
|
11 |
+
import io
|
12 |
+
from langchain.document_loaders import PyPDFDirectoryLoader
|
13 |
+
from langchain.embeddings import SentenceTransformerEmbeddings
|
14 |
+
from langchain_core.output_parsers import StrOutputParser
|
15 |
+
from langchain_core.runnables import RunnablePassthrough
|
16 |
+
|
17 |
+
# --- Your RAG chatbot logic ---
|
18 |
+
source_data_folder = "MyData"
|
19 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
20 |
+
separators=["\n\n", "\n", ". ", " ", ""],
|
21 |
+
chunk_size=2000,
|
22 |
+
chunk_overlap=200
|
23 |
+
)
|
24 |
+
embeddings_model = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
|
25 |
+
path_db = "/content/VectorDB"
|
26 |
+
llm = ChatGoogleGenerativeAI(model="gemini-1.5-pro", google_api_key="AIzaSyAnsIVS4x_7lJLe9AYXGLV8FRwUTQkB-1w")
|
27 |
+
|
28 |
+
# --- Streamlit app starts here ---
|
29 |
+
# Set up the Streamlit app configuration
|
30 |
+
st.set_page_config(
|
31 |
+
page_title="Gemini Pro RAG App",
|
32 |
+
page_icon="π",
|
33 |
+
layout="wide",
|
34 |
+
initial_sidebar_state="expanded",
|
35 |
+
)
|
36 |
+
|
37 |
+
# Initialize session state for chat history and vectorstore (PDF context)
|
38 |
+
if 'chat_history' not in st.session_state:
|
39 |
+
st.session_state.chat_history = [initial_prompt]
|
40 |
+
if 'vectorstore' not in st.session_state:
|
41 |
+
st.session_state.vectorstore = None
|
42 |
+
|
43 |
+
# Function to clear chat history
|
44 |
+
def clear_chat_history():
|
45 |
+
st.session_state.chat_history = [initial_prompt]
|
46 |
+
|
47 |
+
# Extract text from PDF
|
48 |
+
def extract_text_from_pdf(pdf_bytes):
|
49 |
+
pdf_reader = PyPDF2.PdfReader(io.BytesIO(pdf_bytes))
|
50 |
+
text = ""
|
51 |
+
for page in pdf_reader.pages:
|
52 |
+
text += page.extract_text()
|
53 |
+
return text
|
54 |
+
|
55 |
+
# Initialize vectorstore
|
56 |
+
def initialize_vector_index(text):
|
57 |
+
docs = [{'page_content': text}]
|
58 |
+
splits = text_splitter.split_documents(docs)
|
59 |
+
vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings_model, persist_directory=path_db)
|
60 |
+
return vectorstore
|
61 |
+
|
62 |
+
# Sidebar configuration
|
63 |
+
with st.sidebar:
|
64 |
+
st.title('π Gemini RAG Chatbot')
|
65 |
+
st.write('This chatbot uses the Gemini Pro API with RAG capabilities.')
|
66 |
+
st.button('Clear Chat History', on_click=clear_chat_history, type='primary')
|
67 |
+
uploaded_file = st.file_uploader("Upload a PDF file", type=["pdf"], help="Upload your PDF file here to start the analysis.")
|
68 |
+
if uploaded_file is not None:
|
69 |
+
st.success("PDF File Uploaded Successfully!")
|
70 |
+
text = extract_text_from_pdf(uploaded_file.read())
|
71 |
+
vectorstore = initialize_vector_index(text)
|
72 |
+
st.session_state.vectorstore = vectorstore
|
73 |
+
|
74 |
+
# Main interface
|
75 |
+
st.header('Gemini Pro RAG Chatbot')
|
76 |
+
st.subheader('Upload a PDF and ask questions about its content!')
|
77 |
+
|
78 |
+
# Display the welcome prompt if chat history is only the initial prompt
|
79 |
+
if len(st.session_state.chat_history) == 1:
|
80 |
+
with st.chat_message(SPEAKER_TYPES.BOT, avatar="π"):
|
81 |
+
st.write(initial_prompt['content'])
|
82 |
+
|
83 |
+
# Get user input
|
84 |
+
prompt = st.chat_input("Ask a question about the PDF content:", key="user_input")
|
85 |
+
|
86 |
+
# Function to get a response from RAG chain
|
87 |
+
def get_rag_response(prompt):
|
88 |
+
retriever = st.session_state.vectorstore.as_retriever() # Use the stored vectorstore retriever
|
89 |
+
rag_chain = (
|
90 |
+
{"context": retriever | format_docs, "question": RunnablePassthrough()}
|
91 |
+
| prompt
|
92 |
+
| llm
|
93 |
+
| StrOutputParser()
|
94 |
+
)
|
95 |
+
response = rag_chain.invoke(prompt)
|
96 |
+
return response
|
97 |
+
|
98 |
+
# Handle the user prompt and generate response
|
99 |
+
if prompt:
|
100 |
+
# Add user prompt to chat history
|
101 |
+
st.session_state.chat_history.append({'role': SPEAKER_TYPES.USER, 'content': prompt})
|
102 |
+
|
103 |
+
# Display chat messages from the chat history
|
104 |
+
for message in st.session_state.chat_history[1:]:
|
105 |
+
with st.chat_message(message["role"], avatar="π€" if message['role'] == SPEAKER_TYPES.USER else "π"):
|
106 |
+
st.write(message["content"])
|
107 |
+
|
108 |
+
# Get the response using the RAG chain
|
109 |
+
with st.spinner(text='Generating response...'):
|
110 |
+
response_text = get_rag_response(prompt)
|
111 |
+
st.session_state.chat_history.append({'role': SPEAKER_TYPES.BOT, 'content': response_text})
|
112 |
+
|
113 |
+
# Display the bot response
|
114 |
+
with st.chat_message(SPEAKER_TYPES.BOT, avatar="π"):
|
115 |
+
st.write(response_text)
|
116 |
+
|
117 |
+
# Add footer for additional information or credits
|
118 |
+
st.markdown("""
|
119 |
+
<hr>
|
120 |
+
<div style="text-align: center;">
|
121 |
+
<small>Powered by Gemini Pro API | Developed by Christian Thomas BADOLO</small>
|
122 |
+
</div>
|
123 |
+
""", unsafe_allow_html=True)
|