app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Initialize the client with the model ID
client = InferenceClient("13Aluminium/gemma-3.1")

def format_chat_history(history, system_message):
    """Convert the chat history to the format expected by Gemma"""
    formatted_prompt = f"<system>\n{system_message}\n</system>\n\n"
    
    for user_msg, assistant_msg in history:
        if user_msg:
            formatted_prompt += f"<user>\n{user_msg}\n</user>\n\n"
        if assistant_msg:
            formatted_prompt += f"<assistant>\n{assistant_msg}\n</assistant>\n\n"
    
    return formatted_prompt

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Format the history into a text prompt that Gemma understands
    prompt = format_chat_history(history, system_message)
    
    # Add the current message
    prompt += f"<user>\n{message}\n</user>\n\n<assistant>\n"
    
    response = ""
    
    # Use text generation instead of chat completion
    for token in client.text_generation(
        prompt,
        max_new_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        response += token
        yield response

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()