pip install -r requirements.txt
python preprocess.py -share_vocab -save_data vi_data.pkl
python train.py -data_pkl ./vi_data.pkl \
-embs_share_weight -proj_share_weight -label_smoothing \
-output_dir checkpoint -b 32 -warmup=8000 -epoch 100
python translate.py -data_pkl vi_data.pkl -model ./output/model.chkpt -output prediction.txt
Model | R1 | R2 | RL |
---|---|---|---|
SEGMENT | 35.95 | 10.95 | 28.33 |
KeyBERT(phobert-base) | 45.46 | 22.79 | 35.21 |
Transformer small | 51.61 | 30.79 | 44.11 |
Most of the code is referenced in here