Straightfoward to tune, but given the dataset size it would require a substantial amount of compute, more than what a Colab can provide without timing out. The

Straightfoward to tune, but given the dataset size it would require a substantial amount of compute, more than what a Colab can provide without timing out. The comments by the creators imply they used some sort of SaaS for both training and deployment.