Merge pull request #1255 from bjjwwang/develop

rm some redundant files in PaddleClas

python/examples/pipeline/PaddleClas/DarkNet53/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "DarkNet53"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/DarkNet53/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/DarkNet53/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/HRNet_W18_C/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "HRNet_W18_C"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/HRNet_W18_C_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/HRNet_W18_C/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/HRNet_W18_C/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/MobileNetV1/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "MobileNetV1"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/MobileNetV1/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/MobileNetV1/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/MobileNetV2/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "MobileNetV2"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/MobileNetV2/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/MobileNetV2/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-     ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/MobileNetV3_large_x1_0/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "MobileNetV3_large_x1_0"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x1_0_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/MobileNetV3_large_x1_0/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/MobileNetV3_large_x1_0/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/ResNeXt101_vd_64x4d/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "ResNeXt101_vd_64x4d"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_64x4d_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNeXt101_vd_64x4d/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNeXt101_vd_64x4d/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/ResNet50_vd/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "ResNet50_vd"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/ResNet50_vd_FPGM/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "ResNet50_vd_FPGM"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_FPGM_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd_FPGM/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd_FPGM/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/ResNet50_vd_KL/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "ResNet50_vd_KL"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_KL_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd_KL/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd_KL/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/ResNet50_vd_PACT/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "ResNet50_vd_PACT"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_PACT_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd_PACT/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ResNet50_vd_PACT/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done

python/examples/pipeline/PaddleClas/ShuffleNetV2_x1_0/benchmark_config.yaml

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "ShuffleNetV2_x1_0"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x1_0_pretrained.tar"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "gpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ShuffleNetV2_x1_0/benchmark_config.yaml.template

-
-cuda_version: "10.1"
-cudnn_version: "7.6"
-trt_version: "6.0"
-python_version: "3.7"
-gcc_version: "8.2"
-paddle_version: "2.0.1"
-
-cpu: "Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz X12"
-gpu: "T4"
-xpu: "None"
-api: ""
-owner: "cuicheng01"
-
-model_name: "imagenet"
-model_type: "static"
-model_source: "PaddleClas"
-model_url: "model_url_path"
-
-batch_size: 1
-num_of_samples: 1000
-input_shape: "3,224,224"
-
-runtime_device: "cpu"
-ir_optim: true
-enable_memory_optim: true
-enable_tensorrt: false
-precision: "fp32"
-enable_mkldnn: false
-cpu_math_library_num_threads: ""
-
-

python/examples/pipeline/PaddleClas/ShuffleNetV2_x1_0/benchmark_gpu.sh

-export FLAGS_profile_pipeline=1
-alias python3="python3.7"
-modelname="imagenet"
-use_gpu=1
-gpu_id="0"
-benchmark_config_filename="benchmark_config.yaml"
-
-# HTTP
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
-sleep 3
-if [ $use_gpu -eq 1 ]; then
-  python3 benchmark.py yaml local_predictor 1 gpu $gpu_id
-else
-  python3 benchmark.py yaml local_predictor 1 cpu
-fi
-rm -rf profile_log_$modelname
-for thread_num in 1
-do
-  for batch_size in 1
-  do
-    echo "#----imagenet thread num: $thread_num batch size: $batch_size mode:http use_gpu:$use_gpu----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 resnet50_web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=${gpu_id} --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=${gpu_id} --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
-    echo "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
-    python3 benchmark.py run http $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.profiler >>profile_log_$modelname
-    ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-    ps -ef | grep nvidia-smi | awk '{print $2}' | xargs kill -9
-    python3 benchmark.py dump benchmark.log benchmark.tmp
-    mv benchmark.tmp benchmark.log
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_MEM:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTIL:", max}' gpu_utilization.log >> profile_log_$modelname
-    cat benchmark.log >> profile_log_$modelname
-    python3 -m paddle_serving_server_gpu.parse_profile --benchmark_cfg $benchmark_config_filename --benchmark_log profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-  done
-done