app.py

import streamlit as st
from transformers import MarianMTModel, MarianTokenizer, T5ForConditionalGeneration, T5Tokenizer

# Load the MarianMT model for Urdu to English translation
src_lang = 'ur'
tgt_lang = 'en'
model_name = f'Helsinki-NLP/opus-mt-{src_lang}-{tgt_lang}'
model = MarianMTModel.from_pretrained(model_name)
tokenizer = MarianTokenizer.from_pretrained(model_name)

# Load T5 model for key phrase extraction
key_phrase_model = T5ForConditionalGeneration.from_pretrained("t5-small")
key_phrase_tokenizer = T5Tokenizer.from_pretrained("t5-small")

def translate_text(input_text):
    # Translate Urdu text to English
    translated = tokenizer(input_text, return_tensors="pt", padding=True, truncation=True)
    translated_output = model.generate(**translated)
    translated_text = tokenizer.decode(translated_output[0], skip_special_tokens=True)
    return translated_text

def extract_key_phrases(text):
    # Extract key phrases from English text using T5
    input_text = f"extract key phrases: {text}"
    inputs = key_phrase_tokenizer(input_text, return_tensors="pt", padding=True, truncation=True)
    outputs = key_phrase_model.generate(**inputs)
    key_phrases = key_phrase_tokenizer.decode(outputs[0], skip_special_tokens=True)
    return key_phrases

# Streamlit UI
st.title("Urdu Text Translator and Key Phrase Extractor")

# Input Urdu text
input_urdu_text = st.text_area("Enter Urdu Text", "", height=200)

if input_urdu_text:
    # Translate the Urdu text to English
    st.subheader("Translated English Text")
    translated_text = translate_text(input_urdu_text)
    st.text_area("Translated Text (English)", translated_text, height=150, disabled=True)

    # Extract key phrases from the translated English text
    st.subheader("Extracted Key Phrases")
    key_phrases = extract_key_phrases(translated_text)
    st.write(key_phrases)

# Add a footer
st.markdown("---")
st.markdown("Developed by [Your Name]")