app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import os
import spaces

MODEL_NAME = "swiss-ai/Apertus-8B-Instruct-2509"
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load model and tokenizer
HF_TOKEN = os.getenv("HF_TOKEN")
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=HF_TOKEN)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, token=HF_TOKEN).to(device)

@spaces.GPU
def predict(message, history):
    messages = []
    
    # Add history to messages
    for user_msg, assistant_msg in history:
        messages.append({"role": "user", "content": user_msg})
        messages.append({"role": "assistant", "content": assistant_msg})
    
    # Add current message
    messages.append({"role": "user", "content": message})
    
    # Apply chat template
    text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
    )
    
    # Tokenize inputs
    model_inputs = tokenizer([text], return_tensors="pt", add_special_tokens=False).to(model.device)
    
    # Generate response
    generated_ids = model.generate(**model_inputs, max_new_tokens=1000)
    output_ids = generated_ids[0][len(model_inputs.input_ids[0]):]
    response = tokenizer.decode(output_ids, skip_special_tokens=True)
    
    return response

# Create ChatInterface
chatbot = gr.ChatInterface(
    predict,
    title="Apertus-8B Chatbot",
    description="Chat with the Apertus-8B-Instruct model. Enter your message and get a response.",
    examples=[
        "Explain quantum computing in simple terms",
        "How do I make a sandwich?",
        "What is the capital of France?",
        "Write a short poem about the ocean"
    ]
)

# Launch the app
chatbot.launch()