Improving Low-Resource Question Answering with Cross-Lingual Data Augmentation Strategies

This repo is forked form indonlu-repo with several adjusment and addition. This repo is implementation of "Improving Low-Resource Question Answering with Cross-Lingual Data Augmentation Strategies" paper (accepted on ICOICT 2022) (paper).

Requirement

Check on requirment_file

Reproduce Step

• Clone This Repo
• Run Training Script

CUDA_VISIBLE_DEVICES=6 \
python3 main.py \
      --n_epochs=25 \
      --train_batch_size=8 \
      --model_checkpoint=xlm-roberta-base \
      --step_size=1 \
      --gamma=0.9 \
      --device=cuda \
      --experiment_name=xlm-roberta-base-2step-indo-dataset-e3 \
      --lr=1e-5 \
      --early_stop=12 \
      --dataset=qa-factoid-itb \
      --lower \
      --num_layers=12 \
      --max_norm=10 \
      --seed=42 \
      --data_type=original \
      --force

• Or you can test your own model with eval_only

CUDA_VISIBLE_DEVICES=6 \
python3 main.py \
      --n_epochs=25 \
      --train_batch_size=8 \
      --model_checkpoint=./save/qa-factoid-itb/xlm-roberta-base-english-only-dataset-e3/xlm-roberta-pretrained \
      --step_size=1 \
      --gamma=0.9 \
      --device=cuda \
      --experiment_name=xlm-roberta-base-2step-indo-dataset-e3 \
      --lr=1e-5 \
      --early_stop=12 \
      --dataset=qa-factoid-itb \
      --lower \
      --num_layers=12 \
      --max_norm=10 \
      --seed=42 \
      --data_type=original \
      --eval_only \
      --force

Help

You can submit a GitHub issue for asking a question or help. Or you can contact me directly at [email protected] as well