export FLAGS_eager_delete_tensor_gb=0.0 export CUDA_VISIBLE_DEVICES=0 CKPT_DIR="./ckpt_chnsenticorp"
python -u predict.py \ --checkpoint_dir=$CKPT_DIR \ --max_seq_len=128 \ --use_gpu=True \ --batch_size=24
