## 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