roberta-base-als

This project pretrains a roberta-base on the Alemannic (als) data subset of the OSCAR corpus in JAX/Flax.

We will be using the masked-language modeling loss for pretraining.