@lixiri

Day#2

Completed the generation scripts and scaled up the training wohooo!
Temporarily using Kaggle for their TPUs so shoutout to them and their support team who solved my issue veeery fast. Loss curves doing pretty good although it has plateaued a bit but it haappennss we fix and continue!

Day#1

Built the decode/encode pipeline and prepared the training dataset. Training scripts are ready. After creating generation and inference pipeline we will continue with training. Shoutout to TPU Research Cloud for providing access to compute and support!