source $1 export CUDA_VISIBLE_DEVICES=0
python3 -m paddle.distributed.launch ./finetune_mrc.py \
--train_file $train_file \ --dev_file $dev_file \ --max_steps $max_steps \ --lr $lr \ --from_pretrained $from_pretrained \ --save_dir checkpoints
