Streamlit treats timedelta column as strings

I have a sample code below that creates a timedelta column.
While it is sortabe in Python, streamlit treats the column as string column, giving invalid sorting results.

#! /usr/bin/env python3
import time
import pandas as pd
import datetime
import streamlit as st


def main():
    'Main function'

    # Create a Dataframe with random timestamps in last 6 years
    now = time.time()
    times = [
        now - 1,            # 1 second ago
        now - 2,            # 2 seconds ago
        now - 5,            # 5 seconds ago
        now - 60,           # 1 minute ago
        now - 60 * 2,       # 2 minutes ago
        now - 60 * 5,       # 5 minutes ago
        now - 60 * 60,      # 1 hour ago
        now - 60 * 60 * 2,  # 2 hours ago
        now - 60 * 60 * 5,  # 5 hours ago
    ]

    df = pd.DataFrame({'Delta': times})
    df['Delta'] = datetime.datetime.now().astimezone() - pd.to_datetime(
        df['Delta'], unit='s', utc=True
    )
    st.set_page_config(layout='wide')
    st.dataframe(df, use_container_width=True)
    df.sort_values(by='Delta', inplace=True, ascending=False)
    st.dataframe(df, use_container_width=True)

main()

To run

streamlit run --server.runOnSave True --server.port 8501 ./tmp.py 

Before sorting (Presorted DF in both ways)

After sorting via UI
image

I tried exploring other column config types available, but none works.

Is there a way to format the values as it shows in streamlit app while retaining the underlying numeric values for sorting.

Krishna