When MONTAG the magazine ended in early 2019, I was beginning to be interested in LLMs and text
generation (before it was cool) and
created
MONTAG.XYZ as its immortal AI-powered spawn and published it
from
May 2019 - June 2020. I
wrote up the
project including how I did the text scraping to create the data set and how I used Max Woolf's
gpt-2-simple and Spell to train and
sample from the model in Towards
Data
Science,
"The
magazine
that writes itself" in June of 2019. Unfortunately Spell, the ML ops platform I wrote this guide
for,
is now defunct after being
bought by Reddit in 2022.