Edit model card

Megatron-BERT-large Swedish 110k

This BERT model was trained using the Megatron-LM library. The size of the model is a regular BERT-large with 340M parameters. The model was trained on about 70GB of data, consisting mostly of OSCAR and Swedish newspaper text curated by the National Library of Sweden.

Training was done for 110k training steps using a batch size of 8k; the number of training steps is set to 500k, meaning that this version is a checkpoint. The hyperparameters for training followed the setting for RoBERTa.

The model has three sister models trained on the same dataset:

and a later checkpoint

Acknowledgements

We gratefully acknowledge the HPC RIVR consortium (https://www.hpc-rivr.si) and EuroHPC JU (https://eurohpc-ju.europa.eu) for funding this research by providing computing resources of the HPC system Vega at the Institute of Information Science (https://www.izum.si).

Downloads last month
2
Hosted inference API

Unable to determine this model’s pipeline type. Check the docs .