app.py

import os
import gradio as gr
from huggingface_hub import InferenceClient

# Initialize the Hugging Face Inference Client
# Make sure to add your HF_TOKEN in the Space Settings if it's a gated model
client = InferenceClient(
    model="meta-llama/Llama-3.3-70B-Instruct",
    token=os.getenv("HF_TOKEN")
)

def respond(message, chat_history, system_message, max_tokens, temperature, top_p):
    # Format the chat history for the conversational model
    messages = [{"role": "system", "content": system_message}]
    
    for val in chat_history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
            
    messages.append({"role": "user", "content": message})

    response = ""

    # Stream the response back from the Llama 3.3 model
    for msg in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = msg.choices[0].delta.content
        if token:
            response += token
            yield response

# Define a clean Gradio Chat Interface
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a helpful, smart AI assistant powered by Llama 3.3.", label="System Message"),
        gr.Slider(minimum=1, maximum=2048, 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"),
    ],
    title="Llama 3.3 70B Instruct - Agent Demo",
    description="A simple conversational agent interface leveraging Meta's Llama-3.3-70B-Instruct model.",
)

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