python run_qa.py \
  --model_name_or_path bert-large-uncased-whole-word-masking  \
  --dataset_name squad_v2 \
  --do_train \
  --do_eval \
  --save_steps 2500 \
  --eval_steps 2500 \
  --evaluation_strategy steps \
  --per_device_train_batch_size 12 \
  --learning_rate 3e-5 \
  --num_train_epochs 2 \
  --max_seq_length 384 \
  --doc_stride 128 \
  --output_dir /data_2to/devel_data/nn_pruning/output/teacher/squadv2-large-wwm \
  --version_2_with_negative 1