run_base.sh

#!/bin/bash
# This is a base script for all other scripts. It contains some common configs and checks.

var_date=$(date +'%Y%m%d%H%M')

# ====config begin======
# Required configs
# mode: train or eval
mode=${mode:-train}
work_path=${work_path:-/home/yhao/d}
conda_dir=${conda_dir:-/home/yhao/d/conda}
# env1 is the default environment
env1=${env1:-pt_jan02}

if [ ! -d $work_path ]; then
    echo "work_path not exist"
    exit 1
fi
# torchbench path
tb_path=${tb_path:-${work_path}/benchmark}
torchexpert_path=${torchexpert_path:-${work_path}/TorchExpert}
torchexpert=${torchexpert_path}/torchexpert.py
cur_filename=$(basename $0)
prefix_filename=${cur_filename%.*}
log_path=${log_path:-${work_path}/logs}
output=${log_path}/${prefix_filename}_${mode}_${var_date}.log

# env2 is used to speedup comparion scripts like run_all_speedup*
env2=${env2:-pt_jan02_all}
cuda_env1=${cuda_env1:-/home/yhao/setenvs/set11.6-cudnn8.3.3.sh}
# cuda_env2 is used to speedup comparion scripts like run_all_speedup_cuda
cuda_env2=${cuda_env2:-/home/yhao/setenvs/set11.6-cudnn8.5.0.sh}
max_iter=${max_iter:-10}
# pytorch path
pytorch_path=${pytorch_path:-${work_path}/pytorch}

# run.py arguments
metrics_gpu_backend=${metrics_gpu_backend:-default}
enable_jit=${enable_jit:-0}
if [ $enable_jit -eq 1 ]; then
    jit_placeholder="-m jit"
else
    jit_placeholder=""
fi
enable_amp=${enable_amp:-1}
if [ $enable_amp -eq 1 ]; then
    amp_placeholder="--amp"
else
    amp_placeholder=""
fi
enable_inductor=${enable_inductor:-0}
if [ $enable_inductor -eq 1 ]; then
    inductor_placeholder="--torchdynamo inductor"
else
    inductor_placeholder=""
fi


# ====config end======

echo $output
source ${conda_dir}/bin/activate
if [ $? -ne 0 ]; then
    echo "can not activate conda"
    exit 1
fi

if [ ! -d ${work_path}/logs ]; then
    mkdir ${work_path}/logs
fi

echo "" >$output
echo $(date) >>$output
echo "torchexpert: $torchexpert" >>$output
echo "work_path: $work_path" >>$output
echo "tb_path: $tb_path" >>$output
echo "output_csv_file: $output" >>$output
echo "mode: $mode" >>$output
echo "conda_dir: $conda_dir" >>$output
echo "conda envs:" >>$output
echo "env1" $env1 >>$output
echo "env2" $env2 >>$output
if [ $enable_jit -eq 1 ]; then
    echo "enable_jit: True" >>$output
fi
echo "metrics_gpu_backend: $metrics_gpu_backend" >>$output

# use pushover to notify the end of the script, need extra environment variables PUSHOVER_API PUSHOVER_USER_KEY
notify() {
    hostname=$(cat /proc/sys/kernel/hostname)
    curl -s \
        --form-string "token=${PUSHOVER_API}" \
        --form-string "user=${PUSHOVER_USER_KEY}" \
        --form-string "message=${prefix_filename} on ${hostname} done! Output file is ${output}. " \
        https://api.pushover.net/1/messages.json
}

# list all folder names under torchbenchmark/models
all_models=$(ls -d ${tb_path}/torchbenchmark/models/*/ | xargs -n 1 basename)
all_models=$(echo $all_models | tr '\n' ' ')

# all_models="detectron2_fasterrcnn_r_101_dc5 drq hf_GPT2_large mobilenet_v3_large resnet152 timm_nfnet BERT_pytorch detectron2_fasterrcnn_r_101_fpn fambench_xlmr hf_Longformer moco resnet18 timm_regnet Background_Matting detectron2_fasterrcnn_r_50_c4 fastNLP_Bert hf_Reformer nvidia_deeprecommender resnet50 timm_resnest DALLE2_pytorch detectron2_fasterrcnn_r_50_dc5 functorch_dp_cifar10 hf_T5 opacus_cifar10 resnet50_quantized_qat timm_vision_transformer LearningToPaint detectron2_fasterrcnn_r_50_fpn functorch_maml_omniglot hf_T5_base phlippe_resnet resnext50_32x4d timm_vision_transformer_large Super_SloMo detectron2_fcos_r_50_fpn hf_Albert hf_T5_large pyhpc_equation_of_state shufflenet_v2_x1_0 timm_vovnet alexnet detectron2_maskrcnn hf_Bart lennard_jones pyhpc_isoneutral_mixing soft_actor_critic tts_angular attention_is_all_you_need_pytorch detectron2_maskrcnn_r_101_c4 hf_Bert maml pyhpc_turbulent_kinetic_energy speech_transformer vgg16 dcgan detectron2_maskrcnn_r_101_fpn hf_Bert_large maml_omniglot pytorch_CycleGAN_and_pix2pix squeezenet1_1 vision_maskrcnn demucs detectron2_maskrcnn_r_50_c4 hf_BigBird mnasnet1_0 pytorch_stargan tacotron2 yolov3 densenet121 detectron2_maskrcnn_r_50_fpn hf_DistilBert mobilenet_v2 pytorch_struct timm_efficientdet detectron2_fasterrcnn_r_101_c4 dlrm hf_GPT2 mobilenet_v2_quantized_qat pytorch_unet timm_efficientnet"

check_conda_env_exist() {
    if [[ $(conda env list | grep -c "${conda_env}") -eq 0 ]]; then
        echo "Conda environment ${conda_env} does not exist."
        exit 1
    fi
}

check_conda_env_exist $env1