provide basics

A second finetuing of the flan-T5-large model on the downsampled DeepMind LingAlg 1D dataset, this time with a GPU batch size of 256 as opposed to 32 used before

initial commit