Tuning GPT2 on 305K lines of dialog from popular films

Below is the Hugging Face space for my version of GPT-2 I quickly tuned on movie dialog. The original dataset is from Cristian Danescu-Niculescu-Mizil at the University of Cornell and can be found in a Hugging Face dataset here.

This was done on the free version of Google Colab and took ~90 mins and is hosted in a free space, and uses the version of GPT-2 with 137M parameters.