This repository contains an implementation of GPT-2 model and it's smaller
variants and it's training on a random philosophy article. To install simply
run pip install -e . from the cloned repo. In order to keep the progress
updates in the notebook when it's being run for a long time despite
re-connections, modify the jupyter_notebook_config.py file by adding
c.ServerApp.rate_limit_window = float("inf")
To make sure that during training bokeh updates the plot as expected I had to use jupyterlab instead of jupyter.