run_all_inductor.sh

#!/bin/bash
# This script is used to print out all guard check logs with TorchInductor
SHELL_FOLDER=$(
    cd "$(dirname "$0")"
    pwd
)
source ${SHELL_FOLDER}/run_base.sh

cd $tb_path

profile_suffix=logs_profile_${mode}
if [ $env1 ] && [ $env1 != "pt_sep14" ]; then
    profile_suffix=logs_profile_${mode}_$env1_$(date +'%Y%m%d%H%M')
fi

enable_profile=${enable_profile:-0}

enable_amp=${enable_amp:-0}
if [ $enable_amp -eq 1 ]; then
    amp_placeholder="--amp"
else
    amp_placeholder=""
fi

max_iter=1
func_torchinductor() {
    if [ $enable_profile -eq 1 ]; then
        profile_placeholder="--profile --profile-detailed --profile-devices cpu,cuda --profile-folder ${work_path}/${profile_suffix}/${model}/"
    else
        profile_placeholder=""
    fi
    for ((i = 1; i <= $max_iter; i++)); do
        python run.py -d cuda $profile_placeholder $amp_placeholder -t $mode --metrics none $model --torchdynamo inductor >>$output 2>&1
    done
}

source $cuda_env1

echo $(date) >>$output
conda activate $env1
# for model in $all_models
for model in resnet18 resnet50; do
    echo "@Yueming Hao origin $model" >>$output
    func_torchinductor
done
echo $(date) >>$output
notify