run_finetune.sh

export FLAGS_eager_delete_tensor_gb=0.0
export CUDA_VISIBLE_DEVICES=0

# The suggested hyper parameters for difference task
# squad: batch_size=8, weight_decay=0.01, num_epoch=3, max_seq_len=512, lr=3e-5
# squad2.0: batch_size=8, weight_decay=0.01, num_epoch=3, max_seq_len=512, lr=3e-5
# cmrc2018: batch_size=8, weight_decay=0, num_epoch=2, max_seq_len=512, lr=2.5e-5
# drcd: batch_size=8, weight_decay=0, num_epoch=2, max_seq_len=512, lr=2.5e-5

python -u reading_comprehension.py \
                   --batch_size=8 \
                   --use_gpu=True \
                   --checkpoint_dir="./ckpt_squad" \
                   --learning_rate=3e-5 \
                   --weight_decay=0.01 \
                   --warmup_proportion=0.1 \
                   --num_epoch=2 \
                   --max_seq_len=512 \
                   --use_data_parallel=False