Merge pull request #1757 from RainFrost1/benchmark_bug

fix static r50 N1C8 benchmark

test_tipc/static/ResNet50/N1C1/ResNet50_bs128_amp_fp16_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=amp_fp16
+run_mode=DP
+device_num=N1C1
+max_epochs=1
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C1/ResNet50_bs128_fp32_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=fp32
+run_mode=DP
+device_num=N1C1
+max_epochs=1
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item}  ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;
+# run profiling
+sleep 10;
+export PROFILING=true
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C1/ResNet50_bs128_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=pure_fp16
+run_mode=DP
+device_num=N1C1
+max_epochs=1
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C1/ResNet50_bs256_fp16_DP.sh → test_tipc/static/ResNet50/N1C1/ResNet50_bs256_amp_fp16_DP.sh

 model_item=ResNet50
 bs_item=256
-fp_item=fp16
+fp_item=amp_fp16
 run_mode=DP
 device_num=N1C1
 max_epochs=1

test_tipc/static/ResNet50/N1C1/ResNet50_bs256_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=256
+fp_item=pure_fp16
+run_mode=DP
+device_num=N1C1
+max_epochs=1
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C1/ResNet50_bs64_fp16_DP.sh → test_tipc/static/ResNet50/N1C1/ResNet50_bs64_amp_fp16_DP.sh

 model_item=ResNet50
 bs_item=64
-fp_item=fp16
+fp_item=amp_fp16
 run_mode=DP
 device_num=N1C1
 max_epochs=1

test_tipc/static/ResNet50/N1C1/ResNet50_bs64_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=64
+fp_item=pure_fp16
+run_mode=DP
+device_num=N1C1
+max_epochs=1
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C8/ResNet50_bs128_amp_fp16_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=amp_fp16
+run_mode=DP
+device_num=N1C8
+max_epochs=8
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C8/ResNet50_bs128_fp32_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=fp32
+run_mode=DP
+device_num=N1C8
+max_epochs=8
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C8/ResNet50_bs128_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=pure_fp16
+run_mode=DP
+device_num=N1C8
+max_epochs=8
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C8/ResNet50_bs256_fp16_DP.sh → test_tipc/static/ResNet50/N1C8/ResNet50_bs256_amp_fp16_DP.sh

 model_item=ResNet50
 bs_item=256
-fp_item=fp16
+fp_item=amp_fp16
 run_mode=DP
 device_num=N1C8
-max_epochs=1
+max_epochs=8
 num_workers=8
 
 # get data

test_tipc/static/ResNet50/N1C8/ResNet50_bs256_fp32_DP.sh

@@ -3,7 +3,7 @@ bs_item=256
 fp_item=fp32
 run_mode=DP
 device_num=N1C8
-max_epochs=1
+max_epochs=8
 num_workers=8
 
 # get data

test_tipc/static/ResNet50/N1C8/ResNet50_bs256_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=256
+fp_item=pure_fp16
+run_mode=DP
+device_num=N1C8
+max_epochs=8
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N1C8/ResNet50_bs64_fp16_DP.sh → test_tipc/static/ResNet50/N1C8/ResNet50_bs64_amp_fp16_DP.sh

 model_item=ResNet50
 bs_item=64
-fp_item=fp16
+fp_item=amp_fp16
 run_mode=DP
 device_num=N1C8
-max_epochs=1
+max_epochs=8
 num_workers=8
 
 # get data

test_tipc/static/ResNet50/N1C8/ResNet50_bs64_fp32_DP.sh

@@ -3,7 +3,7 @@ bs_item=64
 fp_item=fp32
 run_mode=DP
 device_num=N1C8
-max_epochs=1
+max_epochs=8
 num_workers=8
 
 # get data

test_tipc/static/ResNet50/N1C8/ResNet50_bs64_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=64
+fp_item=pure_fp16
+run_mode=DP
+device_num=N1C8
+max_epochs=8
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs128_amp_fp16_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=amp_fp16
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs128_fp32_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=fp32
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs128_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=128
+fp_item=pure_fp16
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs256_amp_fp16_DP.sh

+model_item=ResNet50
+bs_item=256
+fp_item=amp_fp16
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs256_fp32_DP.sh

+model_item=ResNet50
+bs_item=256
+fp_item=fp32
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs256_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=256
+fp_item=pure_fp16
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs64_amp_fp16_DP.sh

+model_item=ResNet50
+bs_item=64
+fp_item=amp_fp16
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs64_fp32_DP.sh

+model_item=ResNet50
+bs_item=64
+fp_item=fp32
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/N4C32/ResNet50_bs64_pure_fp16_DP.sh

+model_item=ResNet50
+bs_item=64
+fp_item=pure_fp16
+run_mode=DP
+device_num=N4C32
+max_epochs=32
+num_workers=8
+
+# get data
+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh
+# run
+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_mode} ${device_num} ${max_epochs} ${num_workers} 2>&1;

test_tipc/static/ResNet50/benchmark_common/run_benchmark.sh

@@ -34,8 +34,10 @@ function _train(){
 
     if [ ${fp_item} = "fp32" ]; then
         config_file="-c ppcls/configs/ImageNet/ResNet/ResNet50.yaml"
-    else
+    elif [ ${fp_item} = "amp_fp16" ]; then
         config_file="-c ppcls/configs/ImageNet/ResNet/ResNet50_amp_O1.yaml"
+    elif [ ${fp_item} = "pure_fp16" ]; then
+    	config_file="-c ppcls/configs/ImageNet/ResNet/ResNet50_amp_O2.yaml"
     fi
     if [ ${profiling} = "false" ]; then
 	profiling_config=""