| ## BERT Medium for Luxembourgish | |
| Created from a dataset with 1M Luxembourgish sentences from Wikipedia. Corpus has approx. 16M words. | |
| MLM objective was trained. | |
| The BERT model has parameters `L=8` and `H=512`. | |
| Vocabulary has 70K word pieces. | |
| Final loss scores, after 3 epochs: | |
| Final train loss: 4.230 | |
| Final train perplexity: 68.726 | |
| Final validation loss: 4.074 | |
| Final validation perplexity: 58.765 | |