Trying to choose a model/methodology (text generation)

Hello. I’m very new to NLP, though I’ve played around with Char-RNN and GPT-2 for text generation.

The task I’m working on is a text generation one, and I’m trying to choose a model that will work for it. I’m trying to take a corpus of about 5 MB of emails and generate new emails that are similar in tone and subject matter. I’d like to run the generation “online” so that a user could visit a webpage and see a new generated email.

Ideally, this is something that could run on CPU instead of GPU so I could use a low cost EC2 instance.

I’d also like to use a model that could be “fine tuned” (I believe is the expression) to really capture again the tone and subject matter of the emails.

My first attempt at this was using GPT-2 in a Google colab notebook where I generated 1000 examples and downloaded them, with the website being static and simply pulling one of the examples. However like I said, I’d really like to do this “online” so that the output is effectively endless.