Try the demo.
I used download-urls.py to quickly download the HTML from poetryfoundation.org based on the urls in romantic-urls.txt.
Then I used Parse Poetry.ipynb to parse the HTML and extract the title, author, and poem. There are some glitches here with newlines being rendered in some places they shouldn't, and not being rendered in places where they should. This notebook saves a bunch of text files to output/ that include metadata as the first few lines.
Then I used Generate GPT-2.ipynb to generate poems based on random chunks from the poems and the seed words. This notebook saves files to poems.json and generated.json. To run this notebook, first get GPT-2 running, and drop the notebook in the gpt-2/src/ directory.
Both Python notebooks import from utils which I have separately pushed here.
Finally, I load generated.json and poems.json with JavaScript in index.html and display the results.