dv-labse / README.md
monsoon-nlp's picture
training with tokenizer on
7027ae8
|
raw
history blame
991 Bytes
---
language: dv
---
# dv-labse
This is an experiment in cross-lingual transfer learning, to insert Dhivehi word and
word-piece tokens into Google's LaBSE model.
- Original model weights: https://huggingface.co/setu4993/LaBSE
- Original model announcement: https://ai.googleblog.com/2020/08/language-agnostic-bert-sentence.html
This currently outperforms dv-wave and dv-MuRIL (a similar transfer learning model) on
the Maldivian News Classification task https://github.com/Sofwath/DhivehiDatasets
- mBERT: 52%
- dv-wave (ELECTRA): 89%
- dv-muril: 90.7%
- dv-labse: 91.3-91.5% (may continue training)
## Training
- Start with LaBSE (similar to mBERT) with no Thaana vocabulary
- Based on PanLex dictionaries, attach 1,100 Dhivehi words to Sinhalese or English embeddings
- Add remaining words and word-pieces from dv-wave's vocabulary to vocab.txt
- Continue BERT pretraining on Dhivehi text
CoLab notebook:
https://colab.research.google.com/drive/1CUn44M2fb4Qbat2pAvjYqsPvWLt1Novi