Extension of XLM-R with Siamese network and distance-based loss.
source activate pytorch_p36
pip install transformers
- [optional] Install apex
- Prep data:
python dataprep.py --parallel-data-dir data/parallel --output-dir data/processed
python pretrain_distributed.py -i data/processed/train.tsv -o models --batch-size 16 --gpus 8
python train.py