app.py

import gradio as gr
from openai import OpenAI, APIError
import os
import tenacity
import asyncio

ACCESS_TOKEN = os.getenv("HF_TOKEN")

client = OpenAI(
    base_url="https://api-inference.huggingface.co/v1/",
    api_key=ACCESS_TOKEN,
)

# Retry logic with tenacity for handling API rate limits
@tenacity.retry(wait=tenacity.wait_exponential(multiplier=1, min=4, max=10), stop=tenacity.stop_after_attempt(5))
async def respond(
    message,
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    try:
        # Only use the system message and the current message for the response
        messages = [{"role": "system", "content": system_message},
                    {"role": "user", "content": message}]

        response = ""
        # Properly stream chat completions using dot notation
        stream = client.chat.completions.create(
            model="NousResearch/Hermes-3-Llama-3.1-8B",
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
            messages=messages,
        )

        # Stream response and concatenate tokens
        for chunk in stream:
            if hasattr(chunk.choices[0].delta, 'content'):
                token = chunk.choices[0].delta.content
                response += token

        return response

    except APIError as e:
        # Handle both string and dict types of error bodies
        error_details = e.body
        if isinstance(error_details, dict):
            error_type = error_details.get("type", "Unknown")
            error_code = error_details.get("code", "Unknown")
            error_param = error_details.get("param", "Unknown")
            error_message = error_details.get("message", "An error occurred.")
            error_str = f"{error_type}: {error_message} (code: {error_code}, param: {error_param})"
        else:
            error_str = f"Error: {error_details}"

        print(f"APIError: {error_str}")
        return error_str

    except Exception as e:
        print(f"Exception: {e}")
        return "Error occurred. Please try again."


# Async Gradio function to handle user input and response generation without history
async def generate_response(message, system_message, max_tokens, temperature, top_p):
    response = await respond(message, system_message, max_tokens, temperature, top_p)
    return response


def launch_app():
    try:
        demo = gr.Blocks()
        with demo:
            gr.Markdown("# Chatbot")
            message = gr.Textbox(label="Message")
            system_message = gr.Textbox(label="System message")
            max_tokens = gr.Slider(minimum=1, maximum=2048, value=2048, step=1, label="Max new tokens")
            temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
            top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-P")
            response = gr.Text(label="Response")

            # Use the async version of generate_response without history
            gr.Button("Generate Response").click(
                generate_response,
                inputs=[message, system_message, max_tokens, temperature, top_p],
                outputs=[response],
                show_progress=False,
            )
        demo.launch(show_error=True)
    except KeyError as e:
        print(f"Error: {e}")
        print("Please try again.")

if __name__ == "__main__":
    launch_app()