app.py

import streamlit as st
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# ------------------------
# Streamlit UI
# ------------------------
st.set_page_config(page_title="LangChain Demo", page_icon=":robot:")
st.header("MHRV Chatbot")

# ------------------------
# Session memory
# ------------------------
if "conversation" not in st.session_state:
    st.session_state.conversation = []

# ------------------------
# Load model and tokenizer
# ------------------------
model_name = "bigscience/bloom-560m"  # CPU-friendly
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Text-generation pipeline
generator = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device=-1,  # CPU
    max_new_tokens=256,
    temperature=0.3
)

# ------------------------
# Helper functions
# ------------------------
def load_answer(user_input):
    # Add user input to session
    st.session_state.conversation.append({"role": "user", "content": user_input})

    # Build prompt for BLOOM
    system_instruction = (
        "You are a helpful, professional customer support assistant. "
        "Answer questions clearly, politely, and accurately. "
        "If the question is mathematical or factual, provide the correct answer. "
        "Do not repeat the user's message."
    )

    # Use last 3 messages (or fewer if new) to maintain context
    prompt = system_instruction + "\n"
    for msg in st.session_state.conversation[-3:]:
        if msg["role"] == "user":
            prompt += f"User: {msg['content']}\n"
        elif msg["role"] == "assistant":
            prompt += f"Assistant: {msg['content']}\n"
    prompt += "Assistant:"

    # Generate answer
    output = generator(prompt, max_new_tokens=128, do_sample=False)
    answer = output[0]["generated_text"][len(prompt):].strip()

    # Save answer in session
    st.session_state.conversation.append({"role": "assistant", "content": answer})
    return answer

# ------------------------
# Streamlit input
# ------------------------
user_input = st.text_input("You: ", key="input")
submit = st.button("Generate")

if submit and user_input:
    response = load_answer(user_input)
    st.subheader("Answer:")
    st.write(response)

# Optional: show conversation history
if st.checkbox("Show conversation history"):
    for msg in st.session_state.conversation:
        role = "You" if msg["role"] == "user" else "Bot"
        st.write(f"**{role}:** {msg['content']}")