import streamlit as st
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

# Load model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_name = "Ateeqq/Text-Rewriter-Paraphraser"

@st.cache_resource
def load_model():
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)
    return tokenizer, model

tokenizer, model = load_model()

# Rewrite function
def rewrite(text):
    input_ids = tokenizer(f"paraphraser: {text}", return_tensors="pt", truncation=True, max_length=1024).input_ids.to(device)
    output = model.generate(
        input_ids=input_ids,
        num_beams=5,
        no_repeat_ngram_size=3,
        temperature=0.9,
        max_length=1024,
        early_stopping=True,
        eos_token_id=tokenizer.eos_token_id
    )
    return tokenizer.decode(output[0], skip_special_tokens=True)

# UI
st.title("📝 Text Rewriter (Paraphraser)")
text_input = st.text_area("Enter text to rewrite:", height=300)
if st.button("Rewrite"):
    with st.spinner("Rewriting..."):
        result = rewrite(text_input)
        st.success("Done!")
        st.markdown("### 🔁 Rewritten Text")
        st.write(result)