hmByT5 - Preliminary Language Models

Preliminary Historic Multilingual and Monolingual ByT5 Models. Following languages are currently covered:

More details can be found in our GitHub repository.

In this experiment we sample 4B bytes (~4GB of text) from each corpora (and upsample Swedish and Finnish) and train for another epoch (3 epochs in total).

Pretraining

We use the official JAX/FLAX example in Hugging Face Transformers to pretrain a ByT5 model on a single v3-8 TPU. Details about the training can be found here.

Evaluation on Downstream Tasks (NER)

We evaluated the hmByT5 model on downstream tasks:

Model English AjMC German AjMC French AjMC Finnish NewsEye Swedish NewsEye Dutch ICDAR French ICDAR Avg.
hmbyt5-preliminary/byt5-small-multilingual-4g-3e 83.49 ± 0.99 87.38 ± 0.53 84.30 ± 0.51

Acknowledgements

Research supported with Cloud TPUs from Google's TPU Research Cloud (TRC). Many Thanks for providing access to the TPUs ❤️