Super Bowl Tweet Generator

The goal of this course project was to create a twitter thread generating model. Ideally our model will be able to model a twitter thread conversation between a Seahawks and a Patriots fan from Super Bowl XLIX using the data from different tweet datasets. To perform this tasks we finetune GPT2 with a combination of the scraped data and benchmark the performance of the finetuned model against zero shot performance of other popular more powerful models from OpenAI.

We also compared the outputs of all the models (finetuned GPT2, GPT3 and GPT3.5) along with an explanation for the results coupled with a discussion on which model could be more suited for which cases. The training scripts are available on github