diff --git a/TensorFlow/built-in/cv/Video_enhancement/EDVR_ID0056_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/Video_enhancement/EDVR_ID0056_for_TensorFlow/test/train_performance_1p.sh index c64f63e125329b1d51d0779527eb6f5d62210114..a5abb3573c576e882c67560ca3fa3dc1f025e795 100644 --- a/TensorFlow/built-in/cv/Video_enhancement/EDVR_ID0056_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/Video_enhancement/EDVR_ID0056_for_TensorFlow/test/train_performance_1p.sh @@ -28,7 +28,7 @@ batch_size=4 train_steps=800 #学习率 learning_rate= - +ffts='None' #维测参数,precision_mode需要模型审视修改 precision_mode="allow_mix_precision" #维持参数,以下不需要修改 @@ -70,6 +70,8 @@ do mkdir -p ${data_dump_path} elif [[ $para == --data_dump_step* ]];then data_dump_step=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --profiling* ]];then profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling @@ -99,6 +101,10 @@ sed -i "s|data/|${data_path}/|g" $cur_path/../ascendvsr/config/defaults.py sed -i "s|cfg.solver.print_interval = 20|cfg.solver.print_interval = 100|g" $cur_path/../ascendvsr/config/defaults.py sed -i "s|2000|${train_steps}|g" $cur_path/../configs/edvr.yaml +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) cd $cur_path/../ @@ -150,7 +156,11 @@ echo "E2E Training Duration sec : $e2e_time" #训练用例信息,不需要修改 BatchSize=${batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据 #吞吐量,不需要修改 @@ -173,4 +183,4 @@ echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}. echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log -echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log \ No newline at end of file +echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log diff --git a/TensorFlow/built-in/cv/detection/SSD-Resnet34_ID0048_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/detection/SSD-Resnet34_ID0048_for_TensorFlow/test/train_performance_1p.sh index 2e601e6edf3d9fdc8fcf369f0de2c846af2919b4..5a9d324db6e368150784973d889389fa72bb177b 100644 --- a/TensorFlow/built-in/cv/detection/SSD-Resnet34_ID0048_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/detection/SSD-Resnet34_ID0048_for_TensorFlow/test/train_performance_1p.sh @@ -14,7 +14,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 export ASCEND_GLOBAL_LOG_LEVEL_ETP=3 @@ -75,6 +75,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` mv $install_path/fwkacllib/data/rl/Ascend910/custom $install_path/fwkacllib/data/rl/Ascend910/custom_bak @@ -94,7 +96,9 @@ if [[ $data_path == "" ]];then echo "[Error] para \"data_path\" must be confing" exit 1 fi - +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) cd $cur_path/../ @@ -151,7 +155,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据 #吞吐量,不需要修改 ActualFPS=${FPS} diff --git a/TensorFlow/built-in/cv/detection/YoloV3_ID0076_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/detection/YoloV3_ID0076_for_TensorFlow/test/train_performance_1p.sh index 52e0ef983e7766af01dd4200bcd8c2d449419cce..604af32e6d1e3969b18d06db8ca72890739cd1ec 100644 --- a/TensorFlow/built-in/cv/detection/YoloV3_ID0076_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/detection/YoloV3_ID0076_for_TensorFlow/test/train_performance_1p.sh @@ -12,7 +12,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 export ASCEND_GLOBAL_LOG_LEVEL=3 @@ -72,6 +72,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` export autotune=$autotune @@ -96,6 +98,10 @@ fi cp -r $data_path/data/* ${cur_path}/../data/ #sed -i "s/total_epoches = 200/total_epoches = 1/g" ${cur_path}/../args_single.py +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) @@ -156,7 +162,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据 grep "fps" $cur_path/output/${ASCEND_DEVICE_ID}/train_${ASCEND_DEVICE_ID}.log > FPS.log sed -i '1d' FPS.log diff --git a/TensorFlow/built-in/cv/image_classification/DenseNet121_ID0067_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/image_classification/DenseNet121_ID0067_for_TensorFlow/test/train_performance_1p.sh index 61b484ee11868dd17cdef3f3e5cfa825ba6fa7ea..0c272f09e6c15aea742dc9dbacbde198beb5f14b 100644 --- a/TensorFlow/built-in/cv/image_classification/DenseNet121_ID0067_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/image_classification/DenseNet121_ID0067_for_TensorFlow/test/train_performance_1p.sh @@ -12,7 +12,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 export ASCEND_GLOBAL_LOG_LEVEL_ETP=3 @@ -74,6 +74,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --data_path* ]];then data_path=`echo ${para#*=}` fi @@ -84,6 +86,10 @@ if [[ $data_path == "" ]];then echo "[Error] para \"data_path\" must be confing" exit 1 fi +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) @@ -146,7 +152,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 ActualFPS=${FPS} diff --git a/TensorFlow/built-in/cv/image_classification/InceptionV4_ID0002_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/image_classification/InceptionV4_ID0002_for_TensorFlow/test/train_performance_1p.sh index 0e0488de2a3f7f9849ab7280c9b9d1dc2a0f6c5a..a62cd5c25ea7cec48c1358ce82901552c30e90fc 100644 --- a/TensorFlow/built-in/cv/image_classification/InceptionV4_ID0002_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/image_classification/InceptionV4_ID0002_for_TensorFlow/test/train_performance_1p.sh @@ -12,7 +12,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 export ASCEND_GLOBAL_LOG_LEVEL_ETP=3 @@ -73,6 +73,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` mv $install_path/fwkacllib/data/rl/Ascend910/custom $install_path/fwkacllib/data/rl/Ascend910/custom_bak @@ -93,6 +95,10 @@ if [[ $data_path == "" ]];then exit 1 fi + +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) cd $cur_path/../ @@ -151,6 +157,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据 #吞吐量,不需要修改 @@ -173,4 +184,4 @@ echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}. echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log -echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log \ No newline at end of file +echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log diff --git a/TensorFlow/built-in/cv/image_classification/MobileNetV2_ID0074_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/image_classification/MobileNetV2_ID0074_for_TensorFlow/test/train_performance_1p.sh index 3c9d27f2da5623f27d5b02b5d6d653aaf76552cb..59b8dfdc1704dc2fb7cdd4969f375258148dfaea 100644 --- a/TensorFlow/built-in/cv/image_classification/MobileNetV2_ID0074_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/image_classification/MobileNetV2_ID0074_for_TensorFlow/test/train_performance_1p.sh @@ -15,7 +15,7 @@ export HCCL_CONNECT_TIMEOUT=600 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 export ASCEND_GLOBAL_LOG_LEVEL_ETP=3 @@ -77,6 +77,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` export autotune=$autotune @@ -98,7 +100,9 @@ if [[ $data_path == "" ]];then exit 1 fi - +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) @@ -176,6 +180,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据 #吞吐量,不需要修改 diff --git a/TensorFlow/built-in/cv/image_classification/ResNet101_ID0063_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/image_classification/ResNet101_ID0063_for_TensorFlow/test/train_performance_1p.sh index 9e816f6d149330ef6d75f91f555cba7648899f29..259868fe1d8cacc8be7d38d4ed7f31cdaad22821 100644 --- a/TensorFlow/built-in/cv/image_classification/ResNet101_ID0063_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/image_classification/ResNet101_ID0063_for_TensorFlow/test/train_performance_1p.sh @@ -3,6 +3,7 @@ cur_path=`pwd` # 数据集路径,保持为空,不需要修改 data_path="" +ffts='None' #/autotest/CI_daily/ModelZoo_Resnet101_TF_Atlas/data/resnet50/imagenet_TF #集合通信参数,不需要修改 @@ -73,6 +74,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` mv $install_path/fwkacllib/data/rl/Ascend910/custom $install_path/fwkacllib/data/rl/Ascend910/custom_bak @@ -93,6 +96,10 @@ if [[ $data_path == "" ]];then exit 1 fi +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) cd $cur_path/../ @@ -153,6 +160,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据 #吞吐量,不需要修改 diff --git a/TensorFlow/built-in/cv/image_classification/ResNet50_ID0058_for_TensorFlow/test/train_performance_bs256_1p.sh b/TensorFlow/built-in/cv/image_classification/ResNet50_ID0058_for_TensorFlow/test/train_performance_bs256_1p.sh index 48d4a07342cb7cff45a86f0ee9a33530ba56d196..48c689f74015938507571b04776ed2a376652ade 100644 --- a/TensorFlow/built-in/cv/image_classification/ResNet50_ID0058_for_TensorFlow/test/train_performance_bs256_1p.sh +++ b/TensorFlow/built-in/cv/image_classification/ResNet50_ID0058_for_TensorFlow/test/train_performance_bs256_1p.sh @@ -12,7 +12,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 export ASCEND_GLOBAL_LOG_LEVEL_ETP=3 @@ -77,6 +77,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` mv $install_path/fwkacllib/data/rl/Ascend910/custom $install_path/fwkacllib/data/rl/Ascend910/custom_bak @@ -106,6 +108,10 @@ sed -i "50s|PATH_TO_BE_CONFIGURED|${data_path}|g" $cur_path/../src/configs/res5 sed -i "107s|PATH_TO_BE_CONFIGURED|${cur_path}/output/0/d\_solution/ckpt0|g" $cur_path/../src/configs/res50_256bs_1p.py cp data_loader.py $cur_path/../src/data_loader/resnet50/ + +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) cd $cur_path/../ @@ -166,9 +172,14 @@ if [[ ${fp32} == "--fp32" ]];then CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'fp32'_'perf' elif [[ ${hf32} == "--hf32" ]];then CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'hf32'_'perf' +elif [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' else CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' fi + + + ##获取性能数据 #吞吐量,不需要修改 ActualFPS=${FPS} diff --git a/TensorFlow/built-in/cv/image_classification/Resnet50v1.5_ID1721_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/image_classification/Resnet50v1.5_ID1721_for_TensorFlow/test/train_performance_1p.sh index f42f0844adc13c1ea3d6b643f5199630cad3f53f..d7eaecf97f6a5e9c53737247cd598c913cb10a5a 100644 --- a/TensorFlow/built-in/cv/image_classification/Resnet50v1.5_ID1721_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/image_classification/Resnet50v1.5_ID1721_for_TensorFlow/test/train_performance_1p.sh @@ -12,7 +12,7 @@ export PYTHONPATH=${cur_path}/../:$PYTHONPATH # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #设置默认日志级别,不需要修改 #export ASCEND_GLOBAL_LOG_LEVEL_ETP=1 @@ -65,6 +65,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --autotune* ]];then autotune=`echo ${para#*=}` mv $install_path/fwkacllib/data/rl/Ascend910/custom $install_path/fwkacllib/data/rl/Ascend910/custom_bak @@ -87,7 +89,9 @@ if [[ $data_path == "" ]];then exit 1 fi - +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) @@ -143,7 +147,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi #获取性能数据,不需要修改 #吞吐量 ActualFPS=${FPS} diff --git a/TensorFlow/built-in/cv/image_synthesis/SRGAN_ID1881_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/cv/image_synthesis/SRGAN_ID1881_for_TensorFlow/test/train_performance_1p.sh index c1f19bd34764ad52651e31b107abc96286299a90..fa33bd3c13097207d79cbb4ea552a38f8f81a75f 100644 --- a/TensorFlow/built-in/cv/image_synthesis/SRGAN_ID1881_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/cv/image_synthesis/SRGAN_ID1881_for_TensorFlow/test/train_performance_1p.sh @@ -21,7 +21,7 @@ learning_rate=1e-5 #参数配置 data_path="" - +ffts='None' if [[ $1 == --help || $1 == --h ]];then echo "usage:./train_performance_1p.sh" exit 1 @@ -31,6 +31,8 @@ for para in $* do if [[ $para == --data_path* ]];then data_path=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=} fi if [[ $para == --conda_name* ]];then @@ -72,6 +74,9 @@ else fi wait +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi start=$(date +%s) nohup python3 main.py > $cur_path/test/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log 2>&1 & wait @@ -120,7 +125,11 @@ echo "Final Performance images/sec : $FPS" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 ActualFPS=${FPS} diff --git a/TensorFlow/built-in/nlp/Albert_ID0632_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/nlp/Albert_ID0632_for_TensorFlow/test/train_performance_1p.sh index 6208416c36ac4b18a2ebb6ecb71d94a00fd022f6..89b55f87e6b83cb805335474bfd649f7d388f8fd 100644 --- a/TensorFlow/built-in/nlp/Albert_ID0632_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/nlp/Albert_ID0632_for_TensorFlow/test/train_performance_1p.sh @@ -22,7 +22,7 @@ learning_rate=0.00001375 #参数配置 data_path="" - +ffts='None' if [[ $1 == --help || $1 == --h ]];then echo "usage:./train_performance_1p.sh" exit 1 @@ -32,6 +32,8 @@ for para in $* do if [[ $para == --data_path* ]];then data_path=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` fi done @@ -53,6 +55,10 @@ else fi wait +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + start=$(date +%s) nohup python3 -m albert.run_pretraining \ --input_file=${data_path}/data/tf_news_2016_zh_raw_news2016zh_1.tfrecord \ @@ -99,8 +105,11 @@ echo "Final Train Accuracy : ${train_accuracy}" #训练用例信息,不需要修改 BatchSize=${batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 ActualFPS=${FPS} @@ -123,4 +132,4 @@ echo "ActualFPS = ${ActualFPS}" >> $cur_path/test/output/$ASCEND_DEVICE_ID/${Cas echo "TrainingTime = ${TrainingTime}" >> $cur_path/test/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "ActualLoss = ${ActualLoss}" >> $cur_path/test/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/test/output/$ASCEND_DEVICE_ID/${CaseName}.log -echo "TrainAccuracy = ${train_accuracy}" >> $cur_path/test/output/$ASCEND_DEVICE_ID/${CaseName}.log \ No newline at end of file +echo "TrainAccuracy = ${train_accuracy}" >> $cur_path/test/output/$ASCEND_DEVICE_ID/${CaseName}.log diff --git a/TensorFlow/built-in/nlp/Bert-NER_ID0797_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/nlp/Bert-NER_ID0797_for_TensorFlow/test/train_performance_1p.sh index f27d80c0b91d499b79ef5cd975cbb430294689ef..f5d09b6216e8599ea68daaac35d1397f82e7b600 100644 --- a/TensorFlow/built-in/nlp/Bert-NER_ID0797_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/nlp/Bert-NER_ID0797_for_TensorFlow/test/train_performance_1p.sh @@ -39,7 +39,7 @@ data_dump_flag=False data_dump_step="10" profiling=False autotune=False - +ffts='None' # 帮助信息,不需要修改 if [[ $1 == --help || $1 == -h ]];then echo"usage:./train_performance_1P.sh " @@ -71,6 +71,8 @@ do mkdir -p ${data_dump_path} elif [[ $para == --data_dump_step* ]];then data_dump_step=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --profiling* ]];then profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling @@ -92,6 +94,10 @@ sed -i "s|#num_train_steps = 100|num_train_steps = 25|g" $cur_path/../BERT_NER.p mkdir -p $cur_path/../output cp -r $data_path/result_dir $cur_path/../output/ +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) @@ -161,7 +167,11 @@ echo "E2E Training Duration sec : $e2e_time" #训练用例信息,不需要修改 BatchSize=${batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 diff --git a/TensorFlow/built-in/nlp/BertGoogle_Series_for_TensorFlow/test/train_ID0495_Bert-Squad_performance_1p.sh b/TensorFlow/built-in/nlp/BertGoogle_Series_for_TensorFlow/test/train_ID0495_Bert-Squad_performance_1p.sh index 0d67fba344620ab05afedff404e4acbe7f2869e6..d8bcb692c7eaf953a930e29c2fa58e1f7f3103f1 100644 --- a/TensorFlow/built-in/nlp/BertGoogle_Series_for_TensorFlow/test/train_ID0495_Bert-Squad_performance_1p.sh +++ b/TensorFlow/built-in/nlp/BertGoogle_Series_for_TensorFlow/test/train_ID0495_Bert-Squad_performance_1p.sh @@ -11,7 +11,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #基础参数 需要模型审视修改 #网络名称,同目录名称 Network="Bertsquad_ID0495_for_TensorFlow" @@ -60,6 +60,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --data_path* ]];then data_path=`echo ${para#*=}` fi @@ -78,6 +80,9 @@ init_checkpoint=${data_path}/model/bert_model.ckpt train_file=${data_path}/dataset/train-v1.1_small.json predict_file=${data_path}/dataset/dev-v1.1.json +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) @@ -126,7 +131,11 @@ echo "E2E training Duration sec: $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi #获取性能数据 step_per_sec=`grep "global_step/sec:" $cur_path/output/${ASCEND_DEVICE_ID}/train_${ASCEND_DEVICE_ID}.log | awk -F 'global_step/sec:' '{print $2}'|awk 'END {print $1}'` diff --git a/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID0060_BertBase_performance_1p.sh b/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID0060_BertBase_performance_1p.sh index 7aad180ab0d2b3b4d3f6f62bbf87e9739a6c2766..b1fdedf89c2c815a7592f4c4dc12e9e601d5fe1b 100644 --- a/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID0060_BertBase_performance_1p.sh +++ b/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID0060_BertBase_performance_1p.sh @@ -10,7 +10,7 @@ RANK_ID_START=0 # 数据集路径,保持为空,不需要修改 data_path="" - +ffts='None' #基础参数,需要模型审视修改 #网络名称,同目录名称 Network="Bert-base_ID0060_for_TensorFlow" @@ -68,6 +68,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --data_path* ]];then data_path=`echo ${para#*=}` fi @@ -79,6 +81,10 @@ if [[ $data_path == "" ]];then exit 1 fi + +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi #训练开始时间,不需要修改 start_time=$(date +%s) #进入训练脚本目录,需要模型审视修改 @@ -149,7 +155,11 @@ echo "E2E Training Duration sec : $e2e_time" BatchSize=${batch_size} DeviceType=`uname -m` CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中,需要根据模型审视 grep "tensorflow:loss =" $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log | awk -F "loss = " '{print $2}' | awk -F "," '{print $1}' >> $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt diff --git a/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID3068_BertLarge-512_performance_1p_lamb_phase2.sh b/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID3068_BertLarge-512_performance_1p_lamb_phase2.sh index 0b4b153fdf2437db87debcf9b7f4b50d76df5b6f..3474ab1076c3deca262d2c0cc781829326f78cf5 100644 --- a/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID3068_BertLarge-512_performance_1p_lamb_phase2.sh +++ b/TensorFlow/built-in/nlp/BertNV_Series_for_TensorFlow/test/train_ID3068_BertLarge-512_performance_1p_lamb_phase2.sh @@ -22,7 +22,7 @@ batch_size=24 train_steps=200 #学习率 learning_rate= - +ffts='None' #维测参数,precision_mode需要模型审视修改 #precision_mode="allow_mix_precision" #维持参数,以下不需要修改 @@ -64,6 +64,8 @@ do mkdir -p ${data_dump_path} elif [[ $para == --data_dump_step* ]];then data_dump_step=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --profiling* ]];then profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling @@ -81,6 +83,10 @@ if [[ $data_path == "" ]];then exit 1 fi +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) #进入训练脚本目录,需要模型审视修改 @@ -152,8 +158,11 @@ echo "E2E Training Duration sec : $e2e_time" #训练用例信息,不需要修改 BatchSize=${batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi #从train_$ASCEND_DEVICE_ID.log提取Loss到train_${CaseName}_loss.txt中,需要根据模型审视 grep "tensorflow:loss =" $cur_path/output/$ASCEND_DEVICE_ID/train_$ASCEND_DEVICE_ID.log | awk -F "loss = " '{print $2}' | awk -F "," '{print $1}' >> $cur_path/output/$ASCEND_DEVICE_ID/train_${CaseName}_loss.txt @@ -170,4 +179,4 @@ echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}. echo "ActualFPS = ${ActualFPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "TrainingTime = ${TrainingTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log -echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log \ No newline at end of file +echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log diff --git a/TensorFlow/built-in/nlp/Roberta_ID2366_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/built-in/nlp/Roberta_ID2366_for_TensorFlow/test/train_performance_1p.sh index 76a87a1d89bfb44a4f78c3f0642b259183d7f93c..5b1241b02c5346a52fadcca09117faa959f8912f 100644 --- a/TensorFlow/built-in/nlp/Roberta_ID2366_for_TensorFlow/test/train_performance_1p.sh +++ b/TensorFlow/built-in/nlp/Roberta_ID2366_for_TensorFlow/test/train_performance_1p.sh @@ -15,7 +15,7 @@ Network="Roberta_ID2366_for_TensorFlow" batch_size=32 num_train_epochs=1 - +ffts='None' # 帮助信息,不需要修改 if [[ $1 == --help || $1 == -h ]];then echo"usage:./train_performance_1p.sh " @@ -50,6 +50,8 @@ do profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling mkdir -p ${profiling_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --data_path* ]];then data_path=`echo ${para#*=}` echo "${data_path}" @@ -70,6 +72,10 @@ fi cd $cur_path/../ #参数修改 sed -i "s|max_steps=num_train_steps|max_steps=100|g" run_classifier.py +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) for((RANK_ID=$RANK_ID_START;RANK_ID<$((RANK_SIZE+RANK_ID_START));RANK_ID++)); @@ -121,6 +127,11 @@ echo "E2E Training Duration sec : ${e2e_time}" BatchSize=${batch_size} DeviceType=`uname -m` CaseName="${Network}_bs${BatchSize}_${RANK_SIZE}p_perf" +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 ActualFPS=${FPS} diff --git a/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_finetune.sh b/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_finetune.sh index 5552d4802689ea63e0ca9ee0bd02429be75597a0..3dc7c0d69a2eb03328cf29a756fb3587b3404279 100644 --- a/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_finetune.sh +++ b/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_finetune.sh @@ -9,7 +9,7 @@ data_path="" Network="albert_xlarge_zh_ID2355_for_TensorFlow" export JOB_ID=10087 RANK_SIZE=1 - +ffts='None' #npu param task_name=lcqmc_pair do_train=true @@ -53,6 +53,8 @@ do do_train=`echo ${para#*=}` elif [[ $para == --do_eval* ]];then do_eval=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --max_seq_length* ]];then max_seq_length=`echo ${para#*=}` elif [[ $para == --train_batch_size* ]];then @@ -82,6 +84,10 @@ fi cp -r $data_path/albert_config $cur_path/ +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #执行训练 cd $cur_path sed -i "s|max_steps=num_train_steps|max_steps=100|g" run_classifier.py @@ -118,7 +124,11 @@ echo "Final Performance images/sec : $FPS" #训练用例信息,不需要修改 BatchSize=${train_batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 diff --git a/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_pretrain.sh b/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_pretrain.sh index 6b6ac7dad88a9829be8b404e847a7533d6d8286b..0e5db676984e2642f971392bf9f61ab2035c7f2c 100644 --- a/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_pretrain.sh +++ b/TensorFlow/built-in/nlp/albert_xlarge_zh_ID2348_for_TensorFlow/test/train_performance_1p_pretrain.sh @@ -6,7 +6,7 @@ cur_path=`pwd`/../ Network="albert_xlarge_zh_ID2348_for_TensorFlow" RankSize=1 export RANK_SIZE=1 - +ffts='None' #npu param do_train=true max_seq_length=512 @@ -51,6 +51,8 @@ do do_train=`echo ${para#*=}` elif [[ $para == --do_eval* ]];then do_eval=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --max_seq_length* ]];then max_seq_length=`echo ${para#*=}` elif [[ $para == --train_batch_size* ]];then @@ -75,6 +77,10 @@ else mkdir -p $cur_path/test/output/$ASCEND_DEVICE_ID fi +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + cd $cur_path start=$(date +%s) nohup python3 -u run_pretraining.py \ @@ -112,7 +118,11 @@ echo "E2E Training Duration sec : $e2e_time" #训练用例信息,不需要修改 BatchSize=${train_batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RankSize}'p'_'perf' +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 diff --git a/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3057_FwFM_performance_1p.sh b/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3057_FwFM_performance_1p.sh index f192262101cc2881d480bfb3f5ac3a28ff13a438..374c1f88e009e186afeacb9d4ef1de08f12ea83b 100644 --- a/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3057_FwFM_performance_1p.sh +++ b/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3057_FwFM_performance_1p.sh @@ -24,7 +24,7 @@ batch_size=128 train_steps= #学习率 learning_rate= - +ffts='None' #维测参数,precision_mode需要模型审视修改 precision_mode="allow_fp32_to_fp16" #维持参数,以下不需要修改 @@ -62,6 +62,8 @@ do data_dump_flag=`echo ${para#*=}` data_dump_path=${cur_path}/output/data_dump mkdir -p ${data_dump_path} + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --data_dump_step* ]];then data_dump_step=`echo ${para#*=}` elif [[ $para == --profiling* ]];then @@ -87,6 +89,10 @@ cd $cur_path/../examples sed -i "s|epochs=10|epochs=5|g" run_fwfm.py +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + for((RANK_ID=$RANK_ID_START;RANK_ID<$((RANK_SIZE+RANK_ID_START));RANK_ID++)); do #设置环境变量,不需要修改 @@ -137,8 +143,11 @@ echo "E2E Training Duration sec : $e2e_time" #训练用例信息,不需要修改 BatchSize=${batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量 diff --git a/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3058_MMoE_performance_1p.sh b/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3058_MMoE_performance_1p.sh index e2b5df89968d6a4fcd8e31b106503e1eab1f1baf..03ce6c03b3a32eab6b16c902800b0d7a427a5d05 100644 --- a/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3058_MMoE_performance_1p.sh +++ b/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3058_MMoE_performance_1p.sh @@ -24,7 +24,7 @@ batch_size=128 train_steps= #学习率 learning_rate= - +ffts='None' #维测参数,precision_mode需要模型审视修改 precision_mode="allow_fp32_to_fp16" #维持参数,以下不需要修改 @@ -64,6 +64,8 @@ do mkdir -p ${data_dump_path} elif [[ $para == --data_dump_step* ]];then data_dump_step=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --profiling* ]];then profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling @@ -79,6 +81,10 @@ if [[ $data_path == "" ]];then exit 1 fi +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) @@ -135,9 +141,11 @@ BatchSize=${batch_size} DeviceType=`uname -m` if [[ $precision_mode == "must_keep_origin_dtype" ]];then - CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'fp32'_'perf' + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'fp32'_'perf' +elif [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' else - CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' fi ##获取性能数据,不需要修改 diff --git a/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3062_DeepFM_performance_1p.sh b/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3062_DeepFM_performance_1p.sh index 53fdf2d0970dd0e4d3362181cda056b79aca5c44..ae2acfd649e047ab685b1bde1b2c6829b15973ec 100644 --- a/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3062_DeepFM_performance_1p.sh +++ b/TensorFlow/built-in/recommendation/DeepCTR_Series_for_TensorFlow/test/train_ID3062_DeepFM_performance_1p.sh @@ -24,7 +24,7 @@ batch_size=128 train_steps= #学习率 learning_rate= - +ffts='None' #维测参数,precision_mode需要模型审视修改 precision_mode="allow_fp32_to_fp16" #维持参数,以下不需要修改 @@ -65,6 +65,8 @@ do mkdir -p ${data_dump_path} elif [[ $para == --data_dump_step* ]];then data_dump_step=`echo ${para#*=}` + elif [[ $para == --ffts* ]];then + ffts=`echo ${para#*=}` elif [[ $para == --profiling* ]];then profiling=`echo ${para#*=}` profiling_dump_path=${cur_path}/output/profiling @@ -80,6 +82,10 @@ if [[ $data_path == "" ]];then exit 1 fi +if [[ ${ffts} == "--ffts" ]];then + export ASCEND_ENHANCE_ENABLE=1 +fi + #训练开始时间,不需要修改 start_time=$(date +%s) @@ -134,8 +140,11 @@ echo "E2E Training Duration sec : $e2e_time" #训练用例信息,不需要修改 BatchSize=${batch_size} DeviceType=`uname -m` -CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' - +if [[ ${ffts} == "--ffts" ]];then + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf'_'ffts' +else + CaseName=${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'perf' +fi ##获取性能数据,不需要修改 #吞吐量