app.py

import streamlit as st
import os

# Correct imports for newer LangChain versions
from langchain.chains import ConversationChain
from langchain.memory import ConversationSummaryBufferMemory
from langchain_community.llms import LlamaCpp
from huggingface_hub import hf_hub_download

# Page Config
st.set_page_config(page_title="Gemma Free Chat", page_icon="🦙")

# --- Constants ---
# We use a quantized (compressed) version of Gemma 2 (2B parameters)
# This allows it to run on the FREE Hugging Face CPU tier.
REPO_ID = "bartowski/gemma-2-2b-it-GGUF"
FILENAME = "gemma-2-2b-it-Q5_K_M.gguf"

@st.cache_resource
def load_model():
    """
    Downloads and loads the model into memory.
    Cached so it doesn't reload on every interaction.
    """
    print(f"Downloading {FILENAME} from {REPO_ID}...")
    model_path = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)
    
    # Initialize LlamaCpp (The engine that runs the model locally)
    llm = LlamaCpp(
        model_path=model_path,
        temperature=0.7,
        max_tokens=512,
        top_p=0.9,
        # Context window size (how much it remembers in one go)
        n_ctx=2048, 
        # Important for free tier: turn off verbose logging to save buffer space
        verbose=True,
    )
    return llm

# --- UI Layout ---
st.title("🦙 Gemma 2 (2B) - Local & Free")
st.markdown(
    """
    This chatbot runs **entirely inside this Space** using your CPU.
    * **No API Key required.**
    * **Model:** Gemma-2-2B-it (Quantized GGUF)
    * **Speed:** Might be slower than API models because it runs on free hardware.
    """
)

# --- Initialize Model & State ---
try:
    with st.spinner("Loading AI Model (this takes a minute first time)..."):
        llm = load_model()
except Exception as e:
    st.error(f"Failed to load model: {e}")
    st.stop()

# Initialize Chat History
if "messages" not in st.session_state:
    st.session_state.messages = [
        {"role": "assistant", "content": "Hello! I'm running locally on Gemma 2B. How can I help?"}
    ]

# Initialize Chain with Memory
if "conversation_chain" not in st.session_state:
    # Summary Buffer: Keeps recent messages, summarizes old ones to save RAM/Time
    memory = ConversationSummaryBufferMemory(
        llm=llm,
        max_token_limit=500, # Summarize when history exceeds ~500 tokens
        return_messages=True
    )
    
    st.session_state.conversation_chain = ConversationChain(
        llm=llm,
        memory=memory,
        verbose=True
    )

# --- Chat Interface ---

# 1. Display existing messages
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])

# 2. Handle User Input
if prompt := st.chat_input("Type your message..."):
    # Add user message to state and UI
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)

    # Generate Response
    if st.session_state.conversation_chain:
        with st.chat_message("assistant"):
            with st.spinner("Thinking... (CPU working hard 🐢)"):
                try:
                    response = st.session_state.conversation_chain.predict(input=prompt)
                    st.markdown(response)
                    st.session_state.messages.append({"role": "assistant", "content": response})
                except Exception as e:
                    st.error(f"Error during generation: {e}")