Update app.py

app.py

@@ -2,7 +2,6 @@ import os
 import pandas as pd
 import logging
 from datasets import load_dataset
-from sentence_transformers import SentenceTransformer  # Optional: if you want to compute embeddings separately
 from langchain_huggingface import HuggingFaceEmbeddings, HuggingFacePipeline
 from langchain_chroma import Chroma
 from langchain_core.prompts import PromptTemplate
@@ -17,11 +16,10 @@ logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(
 # ------------------------------------------------------------------
 # 1. Load and Prepare the Bank FAQ Dataset
 # ------------------------------------------------------------------
-
-# Load the dataset from Hugging Face (your bank FAQs dataset)
+# Load the dataset from Hugging Face (Bank FAQs)
 ds = load_dataset("maxpro291/bankfaqs_dataset")
 train_ds = ds['train']
-data = train_ds[:]  # Get all examples from the training split
+data = train_ds[:]  # load all examples
 
 # Separate questions and answers from the 'text' field
 questions = []
@@ -32,11 +30,10 @@ for entry in data['text']:
     elif entry.startswith("A:"):
         answers.append(entry)
 
-# Create a DataFrame with the questions and answers
+# Create a DataFrame with questions and answers
 Bank_Data = pd.DataFrame({'question': questions, 'answer': answers})
 
-# Build context strings by combining question and answer for each entry.
-# These will be stored in the vector store.
+# 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']}"
@@ -45,32 +42,32 @@ for i in range(len(Bank_Data)):
 # ------------------------------------------------------------------
 # 2. Create the Vector Store for Retrieval
 # ------------------------------------------------------------------
-
-# Initialize the embedding model using LangChain's HuggingFaceEmbeddings
+# Initialize the embedding model
 embed_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
 
-# Create a Chroma vector store from the context data.
+# Create a Chroma vector store from the context data
 vectorstore = Chroma.from_texts(
     texts=context_data,
     embedding=embed_model,
-    persist_directory="./chroma_db_bank"  # Directory to persist the vector store
+    persist_directory="./chroma_db_bank"
 )
 
-# Create a retriever from the vector store.
+# 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.
 
-# Define the model name for your Hugging Face LLM.
-model_name = "meta-llama/Llama-2-7b-chat-hf"  # Change if you want a different model
-
-# Load the tokenizer and model.
+# Load tokenizer and model
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(model_name)
 
-# Create a Hugging Face pipeline for text-generation.
+# Create a text-generation pipeline
 pipe = pipeline(
     "text-generation",
     model=model,
@@ -81,14 +78,13 @@ pipe = pipeline(
     repetition_penalty=1.15
 )
 
-# Wrap the pipeline in LangChain's HuggingFacePipeline wrapper.
+# 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 model to use provided context.
+# 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. "
@@ -98,11 +94,7 @@ template = (
 )
 rag_prompt = PromptTemplate.from_template(template)
 
-# Build the RAG chain by piping:
-#   (a) the retriever providing context,
-#   (b) the prompt template formatting the question and context,
-#   (c) the LLM generating the answer,
-#   (d) and finally parsing the output.
+# Build the RAG chain by piping the retriever, prompt, LLM, and an output parser
 rag_chain = (
     {"context": retriever, "question": RunnablePassthrough()}
     | rag_prompt
@@ -113,15 +105,14 @@ rag_chain = (
 # ------------------------------------------------------------------
 # 5. Set Up the Gradio Chat Interface
 # ------------------------------------------------------------------
-
 def rag_memory_stream(message, history):
     partial_text = ""
-    # The chain will stream responses; yield the text incrementally.
+    # Stream the generated answer
     for new_text in rag_chain.stream(message):
         partial_text += new_text
         yield partial_text
 
-# Some example questions for the banking assistant
+# Example questions
 examples = [
     "I want to open an account", 
     "What is a savings account?",
@@ -135,7 +126,7 @@ description = (
     "Ask me anything, and I’ll do my best to assist you."
 )
 
-# Create the Gradio ChatInterface (chat-style UI)
+# Create a chat interface using Gradio
 demo = gr.ChatInterface(
     fn=rag_memory_stream,
     title=title,
@@ -147,6 +138,5 @@ demo = gr.ChatInterface(
 # ------------------------------------------------------------------
 # 6. Launch the App
 # ------------------------------------------------------------------
-
 if __name__ == "__main__":
     demo.launch(share=True)