ChatWithPDF / app.py
afatuddin's picture
Update app.py
3a4a56f verified
Raw History Blame Contribute Delete
8.57 kB
import streamlit as st
from PyPDF2 import PdfReader
import os
from dotenv import load_dotenv
# Load environment variables from .env file
load_dotenv()
# --- UPDATED IMPORTS (LANGCHAIN v1.1 COMPATIBLE) ---
from langchain_core.documents import Document
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_core.prompts import ChatPromptTemplate
from langchain_huggingface import HuggingFaceEndpoint, HuggingFaceEndpointEmbeddings, ChatHuggingFace
# ... rest of your code ...
@st.cache_data
def get_pdf_text(_pdf_docs):
"""
Extracts text from PDFs and creates Document objects with metadata.
Returns a list of Document objects, not a string.
"""
documents = []
for pdf in _pdf_docs:
pdf_reader = PdfReader(pdf)
pdf_filename = pdf.name
for i, page in enumerate(pdf_reader.pages):
text = page.extract_text()
if text:
# Create a Document with metadata (page number starts at 1)
doc = Document(
page_content=text,
metadata={"page": i + 1, "source": pdf_filename}
)
documents.append(doc)
return documents
@st.cache_data
def get_text_chunks(_documents):
"""
Splits Document objects into chunks while preserving metadata.
"""
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=2000, # 10k is often too big for precise page tracking; 2k is better
chunk_overlap=200
)
# split_documents ensures metadata (page num) is copied to new chunks
chunks = text_splitter.split_documents(_documents)
return chunks
# --- Hugging Face Model Functions ---
@st.cache_resource
def get_hf_embeddings():
hf_token = os.getenv("HF_API_TOKEN") or st.secrets.get("HF_API_TOKEN")
if not hf_token:
st.error("HF_API_TOKEN not found in environment variables or secrets.toml.")
return None
return HuggingFaceEndpointEmbeddings(
model="BAAI/bge-small-en-v1.5",
huggingfacehub_api_token=hf_token
)
@st.cache_resource
def get_hf_llm():
hf_token = os.getenv("HF_API_TOKEN") or st.secrets.get("HF_API_TOKEN")
if not hf_token:
return None
# Get model selection from session state
model_choice = st.session_state.get("selected_model", "mistralai/Mistral-7B-Instruct-v0.2")
llm = HuggingFaceEndpoint(
huggingfacehub_api_token=hf_token,
repo_id=model_choice,
temperature=0.3,
max_new_tokens=512,
task="conversational"
)
return ChatHuggingFace(llm=llm)
# --- Vector Store Functions (UPDATED) ---
def get_vector_store(_text_chunks):
"""Creates FAISS vector store from Document objects and stores in session state."""
if not _text_chunks:
st.error("No text chunks found.")
return
embeddings = get_hf_embeddings()
if embeddings is None:
return
try:
# Create vector store in memory
vector_store = FAISS.from_documents(documents=_text_chunks, embedding=embeddings)
# Store in session state (persists during the session)
st.session_state.vector_store = vector_store
st.success("Vector store created and loaded successfully.")
except Exception as e:
st.error(f"Error creating vector store: {e}")
def load_faiss_index():
"""Load vector store from session state."""
try:
# Check if vector store exists in session state
if "vector_store" in st.session_state and st.session_state.vector_store is not None:
return st.session_state.vector_store
return None
except Exception as e:
# Silent fail on startup if index doesn't exist yet
return None
# --- Conversational Chain ---
def get_conversational_chain():
model = get_hf_llm()
if model is None:
return None
system_prompt = """You are a helpful AI assistant that answers questions based on the provided document content.
### Context from Documents:
{context}
### Instructions:
1. Read the context carefully and extract relevant information to answer the user's question.
2. Base your answer ONLY on the information provided in the context above.
3. If the information needed to answer the question is not in the context, respond with: "I couldn't find this information in the provided documents. Please upload documents that contain the relevant information."
4. Be concise and direct in your answers.
5. If asking for clarification helps, you can ask follow-up questions.
6. Always cite which page or section your information comes from when possible."""
prompt = ChatPromptTemplate.from_messages([
("system", system_prompt),
("human", "{input}")
])
# Create a simple chain using the pipe operator (|)
chain = prompt | model
return chain
# --- User Input Handler (UPDATED) ---
def user_input(user_question, db):
if db is None:
st.warning("Vector store is not loaded.")
return
try:
# 1. Get relevant documents
docs = db.similarity_search(user_question, k=4)
# 2. Get the response
chain = get_conversational_chain()
if chain is None:
st.error("Check API Key")
return
response = chain.invoke({
"context": docs,
"input": user_question
})
# 3. Display Response
st.markdown("### Reply:")
# Extract just the content text from the response
if hasattr(response, 'content'):
# Response is an AIMessage object
st.write(response.content)
elif isinstance(response, dict) and "answer" in response:
st.write(response["answer"])
elif isinstance(response, dict) and "content" in response:
st.write(response["content"])
else:
# Try to get string representation
st.write(str(response))
# 4. Display Retrieved Documents (Debug Info)
except Exception as e:
st.error(f"Error: {e}")
import traceback
st.error(traceback.format_exc())
# --- Main App ---
def main():
st.set_page_config(page_title="Chat PDF (Hugging Face)")
st.header("Chat with PDF using Hugging Face Models")
# Initialize session state variables
if "faiss_db" not in st.session_state:
st.session_state.faiss_db = None
if "vector_store" not in st.session_state:
st.session_state.vector_store = None
if "selected_model" not in st.session_state:
st.session_state.selected_model = "mistralai/Mistral-7B-Instruct-v0.2"
with st.sidebar:
st.title("Menu:")
# Model Selection Dropdown
model_options = {
"Mistral 7B": "mistralai/Mistral-7B-Instruct-v0.2",
"Llama 3.2 3B": "meta-llama/Llama-3.2-3B-Instruct"
}
selected_model_name = st.selectbox(
"Choose Model:",
options=list(model_options.keys()),
index=0
)
st.session_state.selected_model = model_options[selected_model_name]
st.caption(f"Selected: `{st.session_state.selected_model}`")
st.divider()
pdf_docs = st.file_uploader("Upload PDFs", accept_multiple_files=True)
if st.button("Process PDFs"):
if pdf_docs:
with st.spinner("Processing..."):
# Clear previous cache to avoid using old vector stores
st.cache_data.clear()
st.cache_resource.clear()
# 1. Get Documents (not string)
raw_documents = get_pdf_text(pdf_docs)
# 2. Split Documents (preserves metadata)
text_chunks = get_text_chunks(raw_documents)
# 3. Create Vector Store (stored in session state)
get_vector_store(text_chunks)
st.session_state.faiss_db = load_faiss_index()
else:
st.warning("Upload a PDF first.")
user_question = st.text_input("Ask a Question")
if user_question:
if st.session_state.faiss_db:
user_input(user_question, st.session_state.faiss_db)
else:
st.warning("Process PDFs first.")
if __name__ == "__main__":
main()