app.py

import pandas as pd
from datasets import load_metric
import os
import streamlit as st
from transformers import MarianMTModel, MarianTokenizer


@st.cache(allow_output_mutation=True)
def downloading_model():
    sentence_pair_df = pd.read_json("sentence_pair.json")
    metric = load_metric("sacrebleu")
    tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-id-en")
    original_model = MarianMTModel.from_pretrained(
        "Helsinki-NLP/opus-mt-id-en")
    finetuned_model = MarianMTModel.from_pretrained(
        "wolfrage89/annual_report_translation_id_en")

    return sentence_pair_df, metric, tokenizer, original_model, finetuned_model


def get_translation(model, tokenizer, text):
    translated_tokens = model.generate(
        **tokenizer([text], return_tensors='pt', max_length=104, truncation=True))[0]
    translated_sentence = tokenizer.decode(
        translated_tokens, skip_special_tokens=True)
    return translated_sentence


def get_bleu_score(translated_sentence, reference_sentence, metric):
    metric.add(prediction=translated_sentence, reference=[reference_sentence])
    return metric.compute()['score']


# initalization
if "bahasa_input" not in st.session_state:
    st.session_state["bahasa_input"] = ""
if "ideal_translation" not in st.session_state:
    st.session_state['ideal_translation'] = ""
if "original_translation" not in st.session_state:
    st.session_state['original_translation'] = ""
if "finetuned_translation" not in st.session_state:
    st.session_state['finetuned_translation'] = ""


sentence_pair_df, metric, tokenizer, original_model, finetuned_model = downloading_model()


st.sidebar.title("Bahasa to English Translation (Finance Domain)")
st.sidebar.markdown("---")
random_button = st.sidebar.button(
    "Random")
st.sidebar.write("Randomly generates a bahasa sentence")
st.sidebar.markdown("---")
translate_button = st.sidebar.button(
    "Translate", help="translate bahasa to english")
st.sidebar.write("Translate!")


if random_button:
    sample_data = sentence_pair_df.sample(1)
    st.session_state['bahasa_input'] = sample_data['bahasa'].item()
    st.session_state['ideal_translation'] = sample_data['english'].item()
    st.session_state['original_translation'] = ""
    st.session_state['finetuned_translation'] = ""


if translate_button:
    if len(st.session_state['bahasa_input']) > 0:
        st.session_state['original_translation'] = get_translation(
            original_model, tokenizer, st.session_state['bahasa_input'])
        st.session_state['finetuned_translation'] = get_translation(
            finetuned_model, tokenizer, st.session_state['bahasa_input'])
        original_bleu_score = get_bleu_score(
            st.session_state['original_translation'], st.session_state['ideal_translation'], metric)
        finetuned_bleu_score = get_bleu_score(
            st.session_state['finetuned_translation'], st.session_state['ideal_translation'], metric)
    else:
        st.session_state['original_translation'] = ""
        st.session_state['finetuned_translation'] = ""
        st.session_state['ideal_translation'] = ""
        original_bleu_score = 0
        finetuned_bleu_score = 0

with st.container():
    col_1, col_2 = st.columns(2)
    with col_1:
        st.session_state['bahasa_input'] = st.text_area(
            "Bahasa (Input text here)", value=st.session_state['bahasa_input'], height=200)

        st.text_area(
            "Pretrained model Translation (Helsinki_id_en)", value=st.session_state['original_translation'], height=200)
        if translate_button:
            st.write("Bleu score: ", original_bleu_score)

    with col_2:
        st.text_area("Ideal translation (Target)",
                     value=st.session_state['ideal_translation'], height=200)

        st.text_area("Finetuned translation (Finetuned on annual report)",
                     value=st.session_state['finetuned_translation'], height=200)
        if translate_button:
            st.write("Bleu Score: ", finetuned_bleu_score)