This was trained as part of the paper RomanSetu: Efficiently unlocking multilingual capabilities of Large Language Models via Romanization. The codebase used to train and evaluate this model can be found at https://github.com/AI4Bharat/romansetu.