|
--- |
|
language: |
|
- sv |
|
|
|
--- |
|
|
|
# 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: |
|
- [🤗 BERT Swedish](https://huggingface.co/KBLab/bert-base-swedish-cased-new) |
|
- [Megatron-BERT-base-600k](https://huggingface.co/KBLab/megatron-bert-base-swedish-cased-600k) |
|
- [Megatron-BERT-base-125k](https://huggingface.co/KBLab/megatron-bert-base-swedish-cased-125k) |
|
|
|
and a later checkpoint |
|
- [Megatron-BERT-large-165k](https://huggingface.co/KBLab/megatron-bert-large-swedish-cased-165k) |
|
|
|
## 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). |
