app.py

import gradio as gr
from huggingface_hub import InferenceClient

client = InferenceClient("Qwen/Qwen1.5-4B-Chat")

def respond(message, history: list[tuple[str, str]]):
    system_message = "You are a friendly Chatbot. Respond only in bisaya language. No english translation."
    max_tokens = 4096
    temperature = 0.6
    top_p = 0.95

    messages = [{"role": "system", "content": system_message}]
    for user_text, assistant_text in history:
        if user_text:
            messages.append({"role": "user", "content": user_text})
        if assistant_text:
            messages.append({"role": "assistant", "content": assistant_text})
    messages.append({"role": "user", "content": message})

    response = ""
    previous_response = ""
    for token_message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = token_message.choices[0].delta.get("content", "")
        if not token:
            break
        response += token
        # Only yield if new content was added
        if response != previous_response:
            yield response
            previous_response = response

        # Optional: break out if the response is too long to avoid infinite loops
        if len(response) > 3000:  # adjust threshold as needed
            break

demo = gr.ChatInterface(respond)

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