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

# --- Dùng thư mục cache riêng, tránh PermissionError ---
os.environ["TRANSFORMERS_CACHE"] = "./hf_cache"
os.environ["HF_HOME"] = "./hf_cache"

st.title("🤖 Fine-tuned Qwen3 Chatbot")

# --- Model paths ---
BASE_MODEL = "unsloth/Qwen3-4B-Instruct-2507"
FINE_TUNED = "phuphan1310/Fine-tuned-model-test"

device = "cuda" if torch.cuda.is_available() else "cpu"

@st.cache_resource(show_spinner=True)
def load_model():
    # ⚠️ Dùng tokenizer từ model gốc (Unsloth) vì tokenizer fine-tuned lỗi format
    tokenizer = AutoTokenizer.from_pretrained(
        BASE_MODEL,
        trust_remote_code=True
    )
    model = AutoModelForCausalLM.from_pretrained(
        FINE_TUNED,
        trust_remote_code=True,
        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
        device_map="auto"
    )
    return tokenizer, model

tokenizer, model = load_model()

def generate_response(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=200,
        temperature=0.7,
        top_p=0.9,
        do_sample=True
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

if "messages" not in st.session_state:
    st.session_state.messages = []

user_input = st.text_input("Enter your message:")
if user_input:
    st.session_state.messages.append({"role": "user", "content": user_input})
    response = generate_response(user_input)
    st.session_state.messages.append({"role": "assistant", "content": response})

for msg in st.session_state.messages:
    if msg["role"] == "user":
        st.markdown(f"**You:** {msg['content']}")
    else:
        st.markdown(f"**Bot:** {msg['content']}")