Performance / differences from published paper?

[jlindsey15]

Hi! I was wondering how the performance of this model compares to the published version and what architectural / algorithmic differences, if any, there are between them. Thanks!