app.py

import gradio as gr
from huggingface_hub import InferenceClient
from typing import List, Tuple, Dict

client = InferenceClient("AuriLab/gpt-bi-instruct-cesar")

def format_messages(history: List[Tuple[str, str]], system_message: str, user_message: str) -> List[Dict[str, str]]:
    messages = [{"role": "system", "content": system_message}]
    messages.extend([
        {"role": "user" if i % 2 == 0 else "assistant", "content": str(msg)}  # Convert msg to string
        for turn in history
        for i, msg in enumerate(turn)
        if msg is not None
    ])
    messages.append({"role": "user", "content": str(user_message)})  # Convert user_message to string
    return messages

async def respond(message: str, history: List[Tuple[str, str]]) -> str:
    system_message = "You are a helpful AI assistant."
    max_tokens = 1000
    temperature = 0.7
    top_p = 0.85
    
    messages = format_messages(history, system_message, message)
    response = ""
    
    try:
        async for msg in client.chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
        ):
            if hasattr(msg.choices[0].delta, 'content'):
                token = msg.choices[0].delta.content
                if token is not None:
                    response += token
                    yield response
        
        if not response:  # Handle empty response case
            yield "No response generated."
            
    except Exception as e:
        yield f"Error: {str(e)}"

# Update the ChatInterface to use async function
demo = gr.ChatInterface(
    fn=respond,
    title="Demo GPT-BI instruct",
    examples=["nola duzu izena?", "Nola egiten duzu?"]
)

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