app.py

import gradio as gr
import asyncio
from huggingface_hub import InferenceClient

client = InferenceClient("microsoft/Phi-3.5-mini-instruct")


async def respond(
    message,
    history: list[dict],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Ensure history is in OpenAI-style 'role' and 'content' format
    messages = [{"role": "system", "content": system_message}]
    messages.extend(history)  # Add existing history

    # Add the user's latest message
    messages.append({"role": "user", "content": message})

    response = ""
    async for message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content
        response += token
        yield response


demo = gr.ChatInterface(
    fn=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)",
        ),
    ],
    chatbot=gr.Chatbot(type="messages"),  # Specify the 'messages' format
)

if __name__ == "__main__":
    demo.queue().launch()  # Simply call queue without `concurrency_count`