import streamlit as st
from transformers import AutoModelForSeq2SeqLM, T5ForConditionalGeneration, NllbTokenizer, T5Tokenizer

# Initialize models and tokenizers
translation_model_name = 'sarahai/nllb-uzbek-cyrillic-to-russian'
translation_model = AutoModelForSeq2SeqLM.from_pretrained(translation_model_name)
translation_tokenizer = NllbTokenizer.from_pretrained(translation_model_name)

summarization_model_name = 'sarahai/ruT5-base-summarizer'
summarization_model = T5ForConditionalGeneration.from_pretrained(summarization_model_name)
summarization_tokenizer = T5Tokenizer.from_pretrained(summarization_model_name)

def split_into_chunks(text, tokenizer, max_length=150):
    # Tokenize the text and get ids
    tokens = tokenizer.tokenize(text)
    # Initialize chunks
    chunks = []
    current_chunk = []
    current_length = 0
    for token in tokens:
        current_chunk.append(token)
        current_length += 1
        if current_length >= max_length:
            chunks.append(tokenizer.convert_tokens_to_string(current_chunk))
            current_chunk = []
            current_length = 0
    # Add the last chunk if it's not empty
    if current_chunk:
        chunks.append(tokenizer.convert_tokens_to_string(current_chunk))
    return chunks

def translate(text, model, tokenizer, src_lang='uzb_Cyrl', tgt_lang='rus_Cyrl'):
    tokenizer.src_lang = src_lang
    tokenizer.tgt_lang = tgt_lang
    chunks = split_into_chunks(text, tokenizer)
    translated_chunks = []
    for chunk in chunks:
        inputs = tokenizer(chunk, return_tensors='pt', padding=True, truncation=True, max_length=128)
        outputs = model.generate(inputs['input_ids'], forced_bos_token_id=tokenizer.lang_code_to_id[tgt_lang])
        translated_chunks.append(tokenizer.decode(outputs[0], skip_special_tokens=True))
    return ' '.join(translated_chunks)

def summarize(text, model, tokenizer, max_length=250):
    input_ids = tokenizer.encode("summarize: " + text, return_tensors="pt", max_length=2048, truncation=True)
    summary_ids = model.generate(input_ids, max_length=max_length, length_penalty=2.0, num_beams=4, early_stopping=True)
    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summary

# Streamlit UI
st.title("Перевод с узбекского на русский и суммаризация")
text = st.text_area("Введите текст на узбекском:", height=200)

if st.button("Перевести и суммаризировать"):
    if text:
        with st.spinner('Переводим...'):
            translated_text = translate(text, translation_model, translation_tokenizer)
            st.text_area("Переведенный текст (на русском):", value=translated_text, height=200)
        
        with st.spinner('Суммаризируем...'):
            summary_text = summarize(translated_text, summarization_model, summarization_tokenizer, max_length=250)
            st.text_area("Суммаризация (на русском):", value=summary_text, height=100)
    else:
        st.warning("Пожалуйста, введите текст на узбекском языке для перевода.")