support std log for ocr

python/examples/pipeline/ocr/benchmark.py

@@ -24,16 +24,19 @@ def parse_benchmark(filein, fileout):
     with open(fileout, "w") as fout:
         yaml.dump(res, fout, default_flow_style=False)
 
-def gen_yml(device):
+def gen_yml(device, gpu_id):
     fin = open("config.yml", "r")
     config = yaml.load(fin)
     fin.close()
     config["dag"]["tracer"] = {"interval_s": 10}
     if device == "gpu":
         config["op"]["det"]["local_service_conf"]["device_type"] = 1
-        config["op"]["det"]["local_service_conf"]["devices"] = "2"
+        config["op"]["det"]["local_service_conf"]["devices"] = gpu_id
         config["op"]["rec"]["local_service_conf"]["device_type"] = 1
-        config["op"]["rec"]["local_service_conf"]["devices"] = "2"        
+        config["op"]["rec"]["local_service_conf"]["devices"] = gpu_id
+    else:
+        config["op"]["rec"]["local_service_conf"]["device_type"] = 0     
+        config["op"]["det"]["local_service_conf"]["device_type"] = 0
     with open("config2.yml", "w") as fout: 
         yaml.dump(config, fout, default_flow_style=False)
 
@@ -85,7 +88,11 @@ if __name__ == "__main__":
         mode = sys.argv[2] # brpc/  local predictor
         thread = int(sys.argv[3])
         device = sys.argv[4]
-        gen_yml(device)
+        if device == "gpu":
+            gpu_id = sys.argv[5]
+            gen_yml(device, gpu_id)
+        else:
+            gen_yml(device, "0")
     elif sys.argv[1] == "run":
         mode = sys.argv[2] # http/ rpc
         thread = int(sys.argv[3])

python/examples/pipeline/ocr/benchmark.sh

 export FLAGS_profile_pipeline=1
 alias python3="python3.7"
 modelname="ocr"
+use_gpu=0
+gpu_id="0"
+benchmark_config_filename="benchmark_config.yaml"
+
 # HTTP
 ps -ef | grep web_service | awk '{print $2}' | xargs kill -9 
 sleep 3
-python3 benchmark.py yaml local_predictor 1 gpu 
+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 8 16
+for thread_num in 1
 do
   for batch_size in 1
   do
-    echo "----Bert thread num: $thread_num batch size: $batch_size mode:http ----" >>profile_log_$modelname
+    echo "#----OCR 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 web_service.py >web.log 2>&1 &
     sleep 3
-    nvidia-smi --id=2 --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=2 --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
+    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=$modelname}} END {print "MAX_GPU_MEMORY:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTILIZATION:", max}' gpu_utilization.log >> profile_log_$modelname
+    awk 'BEGIN {max = 0} {if(NR>1){if ($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
-# RPC
-ps -ef | grep web_service | awk '{print $2}' | xargs kill -9
-sleep 3
-python3 benchmark.py yaml local_predictor 1 gpu
-
-for thread_num in 1 8 16
-do
-  for batch_size in 1
-  do
-    echo "----Bert thread num: $thread_num batch size: $batch_size mode:rpc ----" >>profile_log_$modelname
-    rm -rf PipelineServingLogs
-    rm -rf cpu_utilization.py
-    python3 web_service.py >web.log 2>&1 &
-    sleep 3
-    nvidia-smi --id=2 --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
-    nvidia-smi --id=2 --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 rpc $thread_num $batch_size
-    python3 cpu_utilization.py >>profile_log_$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=$modelname}} END {print "MAX_GPU_MEMORY:", max}' gpu_use.log >> profile_log_$modelname
-    awk 'BEGIN {max = 0} {if(NR>1){if ($modelname > max) max=$modelname}} END {print "GPU_UTILIZATION:", max}' gpu_utilization.log >> profile_log_$modelname
-    #rm -rf gpu_use.log gpu_utilization.log
-    cat benchmark.log >> profile_log_$modelname
-  done
-done

python/examples/pipeline/ocr/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: "yingxiaoting"
+
+model_name: "ocr"
+model_type: "static"
+model_source: "PaddleOCR"
+model_url: ""
+
+batch_size: 1
+num_of_samples: 1000
+input_shape: "3,32,X"
+
+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/ocr/eval_top.sh

+sed -e "s/imagenet/${modelname}/g" benchmark_config.yaml.template >benchmark_config.yaml
+sh benchmark.sh
+mv std_benchmark.log std_benchmark.log.cpu
+sed -e "s/use_gpu=0/use_gpu=1/g" benchmark.sh>benchmark_gpu.sh
+sed -e "s/imagenet/${modelname}/g" -e "s/runtime_device: \"cpu\"/runtime_device: \"gpu\"/g" benchmark_config.yaml.template >benchmark_config.yaml
+sh benchmark_gpu.sh
+mv std_benchmark.log std_benchmark.log.gpu
+cp std_benchmark.log.cpu /paddle/Cls/tools/serving_log/${modelname}.log.cpu
+cp std_benchmark.log.gpu /paddle/Cls/tools/serving_log/${modelname}.log.gpu