app.py

import streamlit as st
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer, pipeline
from threading import Thread

# Model Initialization
model_id = "rasyosef/Llama-3.2-180M-Amharic-Instruct"

st.title("Llama 3.2 180M Amharic Chatbot Demo")
st.write("""
This chatbot was created using [Llama-3.2-180M-Amharic-Instruct](https://huggingface.co/rasyosef/Llama-3.2-180M-Amharic-Instruct), 
a finetuned version of the 180 million parameter Llama 3.2 Amharic transformer model.
""")

# Load the tokenizer and model
@st.cache_resource
def load_model():
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForCausalLM.from_pretrained(model_id)
    llama_pipeline = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        pad_token_id=tokenizer.pad_token_id,
        eos_token_id=tokenizer.eos_token_id
    )
    return tokenizer, llama_pipeline

tokenizer, llama_pipeline = load_model()

# Generate text
def generate_response(prompt, chat_history, max_new_tokens):
    history = []

    # Build chat history
    for sent, received in chat_history:
        history.append({"role": "user", "content": sent})
        history.append({"role": "assistant", "content": received})

    history.append({"role": "user", "content": prompt})

    if len(tokenizer.apply_chat_template(history)) > 512:
        return "Chat history is too long."
    else:
        streamer = TextIteratorStreamer(
            tokenizer=tokenizer,
            skip_prompt=True,
            skip_special_tokens=True,
            timeout=300.0
        )
        thread = Thread(target=llama_pipeline, kwargs={
            "text_inputs": history,
            "max_new_tokens": max_new_tokens,
            "repetition_penalty": 1.15,
            "streamer": streamer
        })
        thread.start()

        generated_text = ""
        for word in streamer:
            generated_text += word
            response = generated_text.strip()
            yield response

# Sidebar: Configuration
st.sidebar.header("Chatbot Configuration")
max_tokens = st.sidebar.slider("Maximum new tokens", min_value=8, max_value=256, value=64, help="Larger values result in longer responses.")

# Examples
examples = [
    "ሰላም፣ እንዴት ነህ?",
    "የኢትዮጵያ ዋና ከተማ ስም ምንድን ነው?",
    "የኢትዮጵያ የመጨረሻው ንጉስ ማን ነበሩ?",
    "የአማርኛ ግጥም ፃፍልኝ",
    "ተረት ንገረኝ\n\nጅብና አንበሳ",
    "አንድ አስቂኝ ቀልድ ንገረኝ",
    "የተሰጠው ጽሑፍ አስተያየት ምን አይነት ነው? 'አዎንታዊ'፣ 'አሉታዊ' ወይም 'ገለልተኛ' የሚል ምላሽ ስጥ። 'አሪፍ ፊልም ነበር'",
    "የፈረንሳይ ዋና ከተማ ስም ምንድን ነው?",
    "አሁን የአሜሪካ ፕሬዚዳንት ማን ነው?",
    "ሶስት የአፍሪካ ሀገራት ጥቀስልኝ",
    "3 የአሜሪካ መሪዎችን ስም ጥቀስ",
    "5 የአሜሪካ ከተማዎችን ጥቀስ",
    "አምስት የአውሮፓ ሀገሮችን ጥቀስልኝ",
    "በ ዓለም ላይ ያሉትን 7 አህጉራት ንገረኝ"
]

st.subheader("Chat with the Amharic Chatbot")
chat_history = st.session_state.get("chat_history", [])

# Example selector
example = st.selectbox("Choose an example:", ["Type your own message"] + examples)

# User Input
user_input = st.text_input("Your message:", value=example if example != "Type your own message" else "", placeholder="Type your message here...")

if st.button("Send"):
    if user_input:
        st.session_state.chat_history = st.session_state.get("chat_history", [])
        st.session_state.chat_history.append((user_input, ""))
        responses = generate_response(user_input, st.session_state.chat_history, max_tokens)

        # Stream output
        with st.spinner("Generating response..."):
            final_response = ""
            for response in responses:
                final_response = response
                st.session_state.chat_history[-1] = (user_input, final_response)
                st.rerun()

# Display Chat History
if "chat_history" in st.session_state:
    for i, (user_msg, bot_response) in enumerate(st.session_state.chat_history):
        st.write(f"**User {i+1}:** {user_msg}")
        st.write(f"**Bot:** {bot_response}")