BERT Medium for Luxembourgish
Created from a dataset with 1M Luxembourgish sentences from Wikipedia. Corpus has approx. 16M words.
The 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
- Downloads last month
- 2