app.py

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

st.set_page_config(page_title="Grammar Corrector", page_icon="📝")
st.title("📝 AI Grammar & Spell Corrector")

@st.cache_resource
def load_model():
    model = AutoModelForSeq2SeqLM.from_pretrained("vennify/t5-base-grammar-correction")
    tokenizer = AutoTokenizer.from_pretrained("vennify/t5-base-grammar-correction")
    return model, tokenizer

model, tokenizer = load_model()

def correct_grammar(text):
    input_text = "grammar: " + text
    inputs = tokenizer.encode(input_text, return_tensors="pt", max_length=512, truncation=True)
    with torch.no_grad():
        outputs = model.generate(inputs, max_length=512, num_beams=4, early_stopping=True)
    corrected = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return corrected

text = st.text_area("Enter your text:", height=200)

if st.button("Correct Text"):
    if text.strip():
        with st.spinner("Correcting..."):
            corrected = correct_grammar(text)
            st.subheader("✅ Corrected Text")
            st.write(corrected)
    else:
        st.warning("Please enter some text.")