x_run.sh

#!/bin/bash
set -e
set -x

export TASK=SQUAD
export SQUAD_VERSION=2.0
export MODEL_NAME=base
export SQUAD_DATA=/home/ubuntu/SQuAD_data
export NUM_GPUS=4

export BS=8
export ACCUMULATE=1
GBS=$(($BS * $ACCUMULATE))
export LR=3e-05
export MSL=512
export WD=0.01
export OUTPUT_DIR=albert_out/${TASK}${SQUAD_VERSION}_${MODEL_NAME}_${GBS}_${LR}_${WD}

pip3 install numpy
set +x

python3 -m run_squad \
    --model_name=google_albert_${MODEL_NAME}_v2 \
    --do_train \
    --do_eval \
    --data_dir=${SQUAD_DATA} \
    --output_dir=${OUTPUT_DIR} \
    --gpus=0,1,2,3 \
    --num_accumulate=${ACCUMULATE} \
    --version=${SQUAD_VERSION} \
    --batch_size=${BS} \
    --lr=${LR} \
    --wd=${WD} \
    --max_seq_length=${MSL} \
    --epochs=3 \
    --warmup_ratio=0.1 \