import streamlit as st
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import os

# Imposta la directory di cache locale
os.environ["TRANSFORMERS_CACHE"] = "./hf_cache"

# Titolo dell'app
st.title("🤖 Chatbot DeepSeek con Transformers + Streamlit")

# Carica modello e tokenizer
@st.cache_resource
def load_model():
    model_name = "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
    return tokenizer, model

tokenizer, model = load_model()

# Inizializza la sessione
if "chat_history" not in st.session_state:
    st.session_state.chat_history = []

# Input utente
user_input = st.text_input("Scrivi il tuo messaggio:")

# Generazione risposta
if user_input:
    st.session_state.chat_history.append(("🧑", user_input))

    inputs = tokenizer(user_input, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=256, do_sample=True, temperature=0.7)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    st.session_state.chat_history.append(("🤖", response))

# Mostra la conversazione
for speaker, msg in st.session_state.chat_history:
    st.markdown(f"**{speaker}**: {msg}")