#!/usr/bin/env bash . ./tests/lib.sh i=0 # test kfserver for hparams in \ "--model TransformerSeq2Seq --pretrained_model sshleifer/bart-tiny-random" ; do # train genienlp train \ $SHARED_TRAIN_HPARAMS \ --train_tasks almond \ --train_batch_tokens 100 \ --val_batch_size 100 \ --train_iterations 2 \ --save $workdir/model_$i \ --data $SRCDIR/dataset/ \ $hparams # generate a long sequence long_sequence='' set +x for j in {1..2000}; do long_sequence="${long_sequence} XXX" done set -x # test cuda errors input="{\"id\": \"test\", \"context\": \"${long_sequence}\", \"question\": \"translate to thingtalk\", \"answer\": \"YYY\"}" set +e echo ${input} | genienlp server --path $workdir/model_$i --stdin exit_code=$? set -e if [ $exit_code != 100 ] ; then echo "Cuda error not caught!" exit 1 fi rm -rf $workdir/model_$i i=$((i+1)) done rm -fr $workdir rm -rf $SRCDIR/torch-shm-file-*