app.py

import os
import pandas as pd
import logging
from datasets import load_dataset
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFacePipeline
from langchain_chroma import Chroma
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

# ------------------------------------------------------------------
# 1. Load and Prepare the Bank FAQ Dataset
# ------------------------------------------------------------------
# Load the dataset from Hugging Face (Bank FAQs)
ds = load_dataset("maxpro291/bankfaqs_dataset")
train_ds = ds['train']
data = train_ds[:]  # load all examples

# Separate questions and answers from the 'text' field
questions = []
answers = []
for entry in data['text']:
    if entry.startswith("Q:"):
        questions.append(entry)
    elif entry.startswith("A:"):
        answers.append(entry)

# Create a DataFrame with questions and answers
Bank_Data = pd.DataFrame({'question': questions, 'answer': answers})

# Build context strings (combining question and answer) for the vector store
context_data = []
for i in range(len(Bank_Data)):
    context = f"Question: {Bank_Data.iloc[i]['question']} Answer: {Bank_Data.iloc[i]['answer']}"
    context_data.append(context)

# ------------------------------------------------------------------
# 2. Create the Vector Store for Retrieval
# ------------------------------------------------------------------
# Initialize the embedding model
embed_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Create a Chroma vector store from the context data
vectorstore = Chroma.from_texts(
    texts=context_data,
    embedding=embed_model,
    persist_directory="./chroma_db_bank"
)

# Create a retriever from the vector store
retriever = vectorstore.as_retriever()

# ------------------------------------------------------------------
# 3. Initialize the LLM for Generation
# ------------------------------------------------------------------
# Note:
# The model "meta-llama/Llama-2-7b-chat-hf" is gated. If you have access,
# authenticate using `huggingface-cli login`. Otherwise, switch to a public model.
model_name = "gpt2"  # Replace with "meta-llama/Llama-2-7b-chat-hf" if you are authenticated.

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Create a text-generation pipeline
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_length=512,
    temperature=0.7,
    top_p=0.95,
    repetition_penalty=1.15
)

# Wrap the pipeline in LangChain's HuggingFacePipeline
huggingface_model = HuggingFacePipeline(pipeline=pipe)

# ------------------------------------------------------------------
# 4. Build the Retrieval-Augmented Generation (RAG) Chain
# ------------------------------------------------------------------
# Define a prompt template that instructs the assistant to use provided context
template = (
    "You are a helpful banking assistant. "
    "Use the provided context if it is relevant to answer the question. "
    "If not, answer using your general banking knowledge.\n"
    "Question: {question}\n"
    "Answer:"
)
rag_prompt = PromptTemplate.from_template(template)

# Build the RAG chain by piping the retriever, prompt, LLM, and an output parser
rag_chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | rag_prompt
    | huggingface_model
    | StrOutputParser()
)

# ------------------------------------------------------------------
# 5. Set Up the Gradio Chat Interface
# ------------------------------------------------------------------
def rag_memory_stream(message, history):
    partial_text = ""
    # Stream the generated answer
    for new_text in rag_chain.stream(message):
        partial_text += new_text
        yield partial_text

# Example questions
examples = [
    "I want to open an account", 
    "What is a savings account?",
    "How do I use an ATM?",
    "How can I resolve a bank account issue?"
]

title = "Your Personal Banking Assistant 💬"
description = (
    "Welcome! I’m here to answer your questions about banking and related topics. "
    "Ask me anything, and I’ll do my best to assist you."
)

# Create a chat interface using Gradio
demo = gr.ChatInterface(
    fn=rag_memory_stream,
    title=title,
    description=description,
    examples=examples,
    theme="glass",
)

# ------------------------------------------------------------------
# 6. Launch the App
# ------------------------------------------------------------------
if __name__ == "__main__":
    demo.launch(share=True)