sarvam_api / app.py
Akshayram1's picture
Update app.py
1dbbef2 verified
import streamlit as st
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain.chains import ConversationalRetrievalChain
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.llms import HuggingFaceEndpoint
from langchain.memory import ConversationBufferMemory
from pathlib import Path
import chromadb
from unidecode import unidecode
import re
import os
# List of available LLMs
list_llm = [
"mistralai/Mistral-7B-Instruct-v0.2", "mistralai/Mixtral-8x7B-Instruct-v0.1", "mistralai/Mistral-7B-Instruct-v0.1",
"google/gemma-7b-it", "google/gemma-2b-it", "HuggingFaceH4/zephyr-7b-beta", "HuggingFaceH4/zephyr-7b-gemma-v0.1",
"meta-llama/Llama-2-7b-chat-hf", "microsoft/phi-2", "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "mosaicml/mpt-7b-instruct",
"tiiuae/falcon-7b-instruct", "google/flan-t5-xxl"
]
list_llm_simple = [os.path.basename(llm) for llm in list_llm]
# Load and split PDF document
def load_doc(file_paths, chunk_size, chunk_overlap):
loaders = [PyPDFLoader(fp) for fp in file_paths]
pages = [page for loader in loaders for page in loader.load()]
text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
return text_splitter.split_documents(pages)
# Create vector database
def create_db(docs, collection_name):
embedding = HuggingFaceEmbeddings()
client = chromadb.EphemeralClient()
return Chroma.from_documents(documents=docs, embedding=embedding, client=client, collection_name=collection_name)
# Initialize LLM chain
def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
if llm_model in ["mistralai/Mixtral-8x7B-Instruct-v0.1", "HuggingFaceH4/zephyr-7b-gemma-v0.1", "mosaicml/mpt-7b-instruct"]:
raise ValueError("LLM model is too large to be loaded automatically on free inference endpoint")
model_kwargs = {"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k}
llm = HuggingFaceEndpoint(repo_id=llm_model, **model_kwargs)
memory = ConversationBufferMemory(memory_key="chat_history", output_key='answer', return_messages=True)
retriever = vector_db.as_retriever()
return ConversationalRetrievalChain.from_llm(llm, retriever=retriever, chain_type="stuff", memory=memory, return_source_documents=True, verbose=False)
# Generate collection name for vector database
def create_collection_name(filepath):
collection_name = Path(filepath).stem
collection_name = unidecode(collection_name)
collection_name = re.sub('[^A-Za-z0-9]+', '-', collection_name)
collection_name = collection_name[:50]
if len(collection_name) < 3:
collection_name = collection_name + 'xyz'
if not collection_name[0].isalnum():
collection_name = 'A' + collection_name[1:]
if not collection_name[-1].isalnum():
collection_name = collection_name[:-1] + 'Z'
return collection_name
# Initialize database
def initialize_database(files, chunk_size, chunk_overlap):
file_paths = [file.name for file in files]
collection_name = create_collection_name(file_paths[0])
doc_splits = load_doc(file_paths, chunk_size, chunk_overlap)
vector_db = create_db(doc_splits, collection_name)
return vector_db, collection_name, "Complete!"
# Initialize LLM
def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db):
llm_name = list_llm[llm_option]
qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db)
return qa_chain, "Complete!"
# Format chat history
def format_chat_history(message, chat_history):
return [f"User: {user_message}\nAssistant: {bot_message}" for user_message, bot_message in chat_history]
# Handle conversation
def conversation(qa_chain, message, history):
formatted_chat_history = format_chat_history(message, history)
response = qa_chain({"question": message, "chat_history": formatted_chat_history})
response_answer = response["answer"].split("Helpful Answer:")[-1] if "Helpful Answer:" in response["answer"] else response["answer"]
response_sources = response["source_documents"]
response_source1 = response_sources[0].page_content.strip()
response_source2 = response_sources[1].page_content.strip()
response_source3 = response_sources[2].page_content.strip()
response_source1_page = response_sources[0].metadata["page"] + 1
response_source2_page = response_sources[1].metadata["page"] + 1
response_source3_page = response_sources[2].metadata["page"] + 1
new_history = history + [(message, response_answer)]
return qa_chain, "", new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
# Streamlit app
def main():
st.title("PDF-based Chatbot")
st.write("Upload your PDF documents and interact with the chatbot to get insights from your PDFs.")
uploaded_files = st.file_uploader("Upload PDF files", type="pdf", accept_multiple_files=True)
if uploaded_files:
chunk_size = st.slider("Chunk Size", 100, 1000, 600)
chunk_overlap = st.slider("Chunk Overlap", 10, 200, 40)
vector_db, collection_name, db_status = initialize_database(uploaded_files, chunk_size, chunk_overlap)
st.write(f"Vector Database Initialized: {db_status}")
llm_option = st.selectbox("Select LLM Model", options=list_llm_simple)
llm_temperature = st.slider("Temperature", 0.01, 1.0, 0.7)
max_tokens = st.slider("Max Tokens", 224, 4096, 1024)
top_k = st.slider("Top-K Samples", 1, 10, 3)
qa_chain, llm_status = initialize_LLM(list_llm_simple.index(llm_option), llm_temperature, max_tokens, top_k, vector_db)
st.write(f"QA Chain Initialized: {llm_status}")
st.write("Chat with the bot:")
chat_history = []
user_message = st.text_input("Your Message:")
if st.button("Submit"):
if user_message:
qa_chain, _, chat_history, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page = conversation(qa_chain, user_message, chat_history)
st.write(f"**Bot's Response:** {chat_history[-1][1]}")
st.write(f"**Reference 1:** {doc_source1} (Page {source1_page})")
st.write(f"**Reference 2:** {doc_source2} (Page {source2_page})")
st.write(f"**Reference 3:** {doc_source3} (Page {source3_page})")
if __name__ == "__main__":
main()