app.py

from datasets import load_dataset
import pandas as pd
import numpy as np
import streamlit as st
from transformers import AutoTokenizer
import matplotlib.pyplot as plt

st.set_page_config(layout="wide")

with st.sidebar:
    subset = st.selectbox('subset', ('dev', 'devtest'))

with st.echo():
    tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
    flores = load_dataset("facebook/flores", "eng_Latn-ukr_Cyrl")
    dataset = flores[subset]
    eng_num_tokens = dataset.map(lambda x: {'num_tokens':len(tokenizer(x['sentence_eng_Latn'])['input_ids'])})['num_tokens']
    ukr_num_tokens = dataset.map(lambda x: {'num_tokens':len(tokenizer(x['sentence_ukr_Cyrl'])['input_ids'])})['num_tokens']

fig, (axl, axr) = plt.subplots(1, 2, figsize=(10,3))
axl.hist(eng_num_tokens)
axl.set_title(f'eng mistral tokens ({np.sum(eng_num_tokens)} total)')
axr.hist(ukr_num_tokens)
axr.set_title(f'ukr mistral tokens ({np.sum(ukr_num_tokens)} total)')
st.pyplot(fig)

st.dataframe(pd.DataFrame(dataset))