Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Import required modules from 'langchain' for document processing, embeddings, Q&A, etc.
|
2 |
+
from langchain.document_loaders import PyPDFLoader
|
3 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
4 |
+
from langchain.vectorstores import Chroma
|
5 |
+
from langchain.embeddings import OpenAIEmbeddings
|
6 |
+
from langchain.chat_models import ChatOpenAI
|
7 |
+
from langchain.chains import RetrievalQA
|
8 |
+
|
9 |
+
# Importing Streamlit for creating the web app, and other necessary modules for file handling.
|
10 |
+
import streamlit as st
|
11 |
+
import tempfile
|
12 |
+
import os
|
13 |
+
|
14 |
+
# Import a handler for streaming outputs.
|
15 |
+
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
|
16 |
+
|
17 |
+
# Set the title of the Streamlit web application.
|
18 |
+
st.title("ChatPDF")
|
19 |
+
# Create a horizontal line for better visual separation in the app.
|
20 |
+
st.write("---")
|
21 |
+
|
22 |
+
# Provide an input box for users to enter their OpenAI API key.
|
23 |
+
openai_key = st.text_input('Enter OPEN_AI_API_KEY', type="password")
|
24 |
+
|
25 |
+
# Provide a file upload widget to let users upload their PDF files.
|
26 |
+
uploaded_file = st.file_uploader("Upload your PDF file!", type=['pdf'])
|
27 |
+
# Another visual separation after the file uploader.
|
28 |
+
st.write("---")
|
29 |
+
|
30 |
+
# Define a function that converts the uploaded PDF into a document format.
|
31 |
+
def pdf_to_document(uploaded_file):
|
32 |
+
# Create a temporary directory to store the uploaded PDF file temporarily.
|
33 |
+
temp_dir = tempfile.TemporaryDirectory()
|
34 |
+
# Join the directory path with the uploaded file name to get the complete path.
|
35 |
+
temp_filepath = os.path.join(temp_dir.name, uploaded_file.name)
|
36 |
+
|
37 |
+
# Write the content of the uploaded file into the temporary file path.
|
38 |
+
with open(temp_filepath, "wb") as f:
|
39 |
+
f.write(uploaded_file.getvalue())
|
40 |
+
|
41 |
+
# Use PyPDFLoader to read and split the PDF into individual pages.
|
42 |
+
loader = PyPDFLoader(temp_filepath)
|
43 |
+
pages = loader.load_and_split()
|
44 |
+
return pages
|
45 |
+
|
46 |
+
# Check if a file has been uploaded by the user.
|
47 |
+
if uploaded_file is not None:
|
48 |
+
# Convert the uploaded PDF into a document format.
|
49 |
+
pages = pdf_to_document(uploaded_file)
|
50 |
+
|
51 |
+
# Initialize a text splitter to break the document into smaller chunks.
|
52 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
53 |
+
# Define parameters for the splitter: chunk size, overlap, etc.
|
54 |
+
chunk_size = 300,
|
55 |
+
chunk_overlap = 20,
|
56 |
+
length_function = len
|
57 |
+
)
|
58 |
+
# Split the document pages into chunks.
|
59 |
+
texts = text_splitter.split_documents(pages)
|
60 |
+
|
61 |
+
# Initialize the OpenAIEmbeddings model for creating embeddings from texts using the provided API key.
|
62 |
+
embeddings_model = OpenAIEmbeddings(openai_api_key=openai_key)
|
63 |
+
|
64 |
+
# Load the textual chunks into Chroma after creating embeddings.
|
65 |
+
db = Chroma.from_documents(texts, embeddings_model)
|
66 |
+
|
67 |
+
# Define a custom handler to stream outputs to the Streamlit app.
|
68 |
+
from langchain.callbacks.base import BaseCallbackHandler
|
69 |
+
class StreamHandler(BaseCallbackHandler):
|
70 |
+
def __init__(self, container, initial_text=""):
|
71 |
+
self.container = container
|
72 |
+
self.text=initial_text
|
73 |
+
def on_llm_new_token(self, token: str, **kwargs) -> None:
|
74 |
+
self.text+=token
|
75 |
+
self.container.markdown(self.text)
|
76 |
+
|
77 |
+
# Display a header for the question section of the web app.
|
78 |
+
st.header("Ask the PDF a question!")
|
79 |
+
# Provide an input box for users to type in their questions.
|
80 |
+
question = st.text_input('Type your question')
|
81 |
+
|
82 |
+
# Check if the user has clicked on the 'Ask' button.
|
83 |
+
if st.button('Ask'):
|
84 |
+
# Show a spinner animation while processing the user's question.
|
85 |
+
with st.spinner('Processing...'):
|
86 |
+
# Create a space to display the answer.
|
87 |
+
chat_box = st.empty()
|
88 |
+
# Initialize a handler to stream outputs.
|
89 |
+
stream_hander = StreamHandler(chat_box)
|
90 |
+
# Initialize the ChatOpenAI model for Q&A tasks with streaming enabled.
|
91 |
+
llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0, openai_api_key=openai_key, streaming=True, callbacks=[stream_hander])
|
92 |
+
# Create a RetrievalQA chain that uses the ChatOpenAI model and Chroma retriever to answer the question.
|
93 |
+
qa_chain = RetrievalQA.from_chain_type(llm, retriever=db.as_retriever())
|
94 |
+
# Fetch the answer to the user's question.
|
95 |
+
qa_chain({"query": question})
|