app.py

import gradio as gr
from huggingface_hub import InferenceClient
import os

client = InferenceClient(
    model="mistralai/Mistral-7B-Instruct-v0.3",
    token=os.getenv('HF_TOKEN')
)

def chat_fn(message, system_message, history_str, max_tokens, temperature, top_p):
    # Convert history string (optional) to message list
    messages = [{"role": "system", "content": system_message}]
    
    if history_str:
        # Format: user1||assistant1\nuser2||assistant2
        for pair in history_str.split("\n"):
            if "||" in pair:
                user_msg, assistant_msg = pair.split("||", 1)
                messages.append({"role": "user", "content": user_msg})
                messages.append({"role": "assistant", "content": assistant_msg})
    
    messages.append({"role": "user", "content": message})

    # Get response from HF
    response = ""
    for chunk in client.chat_completion(
        messages=messages,
        stream=True,
        max_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    ):
        response += chunk.choices[0].delta.content or ""
    
    return response

demo = gr.Interface(
    fn=chat_fn,
    inputs=[
        gr.Textbox(lines=2, label="User Message"),
        gr.Textbox(value="You are a friendly Chatbot.", label="System Prompt"),
        gr.Textbox(lines=4, placeholder="user||bot\nuser2||bot2", label="Conversation History (optional)"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max 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"),
    ],
    outputs="text",
    allow_flagging="never",
    title="LLM Budaya",
    description="Chatbot menggunakan model HuggingFace Zephyr-7B"
)

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