app.py

# app.py
import gradio as gr
import torch
from threading import Thread
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

# Choose any chat model with a chat template; Zephyr works well:
MODEL_NAME = "google/gemma-3-270m-it"

# Load model + tokenizer
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype="auto",
    device_map="auto",
)

def build_chat(system_message: str, history: list[tuple[str, str]], user_message: str):
    """Convert Gradio history into a list of chat messages for apply_chat_template."""
    messages = []
    if system_message:
        messages.append({"role": "system", "content": system_message})

    for u, a in history:
        if u:
            messages.append({"role": "user", "content": u})
        if a:
            messages.append({"role": "assistant", "content": a})

    messages.append({"role": "user", "content": user_message})
    return messages

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # 1) Build chat messages and tokenize using the model's chat template
    messages = build_chat(system_message, history, message)

    inputs = tokenizer.apply_chat_template(
        messages,
        add_generation_prompt=True,
        tokenize=True,
        return_tensors="pt",
    )

    inputs = inputs.to(model.device)

    # 2) Stream generation
    streamer = TextIteratorStreamer(
        tokenizer,
        skip_prompt=True,
        skip_special_tokens=True,
    )

    gen_kwargs = dict(
        input_ids=inputs,
        max_new_tokens=int(max_tokens),
        do_sample=True,
        temperature=float(temperature),
        top_p=float(top_p),
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.eos_token_id,
        streamer=streamer,
    )

    # Run generate() in a background thread while we yield chunks
    thread = Thread(target=model.generate, kwargs=gen_kwargs)
    thread.start()

    response = ""
    for new_text in streamer:
        response += new_text
        yield response

    thread.join()

# Gradio UI (same controls as your example)
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()