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": msg}
        for turn in history
        for i, msg in enumerate(turn)
        if msg
    ])
    messages.append({"role": "user", "content": user_message})
    return messages

def respond(message: str, history: List[Tuple[str, str]], system_message: str, max_tokens: int, temperature: float, top_p: float) -> str:
    messages = format_messages(history, system_message, message)
    response = ""
    
    for msg in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
        repetition_penalty=1.2,  # Add repetition penalty
        presence_penalty=0.5,    # Penalize presence of repeated tokens
        frequency_penalty=0.5,   # Penalize frequency of repeated tokens
    ):
        token = msg.choices[0].delta.content
        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=256, 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()