app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

model_id = "matteoangeloni/EduRaccoon"

# Carica tokenizer e modello
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto",
    trust_remote_code=True
)

pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
)

def respond(message, history):
    # System prompt: controlla lingua e ruolo
    system_prompt = (
        "You are EduRaccoon, an AI tutor. "
        "Always reply in the same language used by the user. "
        "If the question is educational (school, university, science, history, math, literature, etc.), "
        "answer clearly and in detail as a teacher would. "
        "If the question is not educational, answer briefly and naturally."
    )

    # Costruiamo il prompt finale
    prompt = system_prompt + "\nUser: " + message + "\nAI:"

    result = pipe(
        prompt,
        max_new_tokens=128,
        do_sample=True,
        temperature=0.7,
        top_p=0.9,
        pad_token_id=tokenizer.eos_token_id
    )

    output_text = result[0]["generated_text"]
    return output_text.split("AI:")[-1].strip()

# Interfaccia Gradio
chatbot = gr.ChatInterface(
    fn=respond,
    type="messages",
    title="EduRaccoon - Educational AI",
    description="Chatta con EduRaccoon: un tutor AI che risponde sempre nella tua lingua!"
)

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