search

mindspore-ai / mindspore

MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios.
https://gitee.com/mindspore/mindspore
Apache License 2.0
4.24k stars 701 forks source link

having issues on run mindspore lite benchmark tools On hi3516 GPU environment #174

Open vgg4resnet opened 2 years ago

vgg4resnet commented 2 years ago

I have converted the model on ubuntu ,and I am able to run the model on Hisi3516 cpu mode ,but when running in a GPU mode ,I am meeting this problems :

./benchmark --modelFile=../../../mindspore-lite-1.5.0-linux-x64/tools/converter/converter/mob.ms --device=GPU ModelPath = ../../../mindspore-lite-1.5.0-linux-x64/tools/converter/converter/mob.ms InDataPath = InDataType = bin LoopCount = 10 DeviceType = GPU AccuracyThreshold = 0.5 WarmUpLoopCount = 3 NumThreads = 2 Fp16Priority = 0 EnableParallel = 0 calibDataPath = cpuBindMode = HIGHER_CPU CalibDataType = FLOAT start reading model file ERROR [mindspore/lite/src/inner_context.cc:206] IsValid] GPU is not supported. ERROR [mindspore/lite/src/inner_context.cc:114] Init] Context is not valid ERROR [mindspore/lite/src/lite_session.cc:710] Init] Init Context failed ERROR [mindspore/lite/src/lite_session.cc:1045] CreateSession] init session failed ERROR [/benchmark.cc:490] RunBenchmark] CreateSession failed while running CreateSession failed while running ERROR [/run_benchmark.cc:63] RunBenchmark] Run Benchmark mob.ms Failed : -1

cpu mode works well

./benchmark --modelFile=../../../mindspore-lite-1.5.0-linux-x64/tools/converter/converter/mob.ms ModelPath = ../../../mindspore-lite-1.5.0-linux-x64/tools/converter/converter/mob.ms InDataPath = InDataType = bin LoopCount = 10 DeviceType = CPU AccuracyThreshold = 0.5 WarmUpLoopCount = 3 NumThreads = 2 Fp16Priority = 0 EnableParallel = 0 calibDataPath = cpuBindMode = HIGHER_CPU CalibDataType = FLOAT start reading model file PrepareTime = 696 ms Running warm up loops... Running benchmark loops... Model = mob.ms, NumThreads = 2, MinRunTime = 735.919006 ms, MaxRuntime = 736.770020 ms, AvgRunTime = 736.351990 ms Run Benchmark mob.ms Success.

./converter_lite --fmk=CAFFE --modelFile=./avger47Another14_mark_nnie_20210901193846.prototxt --weightFile=./new2ours_mobilefacenet_p0.25.caffemodel --configFile=./converter.cfg --outputFile=mob Mapper Version 1.2.3.2_B010 (NNIE_1.2) 2106240b5ebc1d begin net parsing.... end net parsing begin prev optimizing.... end prev optimizing.... begin net quantalizing(CPU).... end quantalizing begin optimizing.... end optimizing begin NNIE[0] mem allocation.... end NNIE[0] memory allocating begin NNIE[0] instruction generating.... end NNIE[0] instruction generating begin lbs binary code generating.... end lbs binary code generating CONVERT RESULT SUCCESS:0

jh0905 commented 2 years ago

Hi, we're sorry to tell you that mindspore currently only supports running hi3516 on CPU mode. GPU Mode is not supported for now. And you can follow the latest news on https://www.mindspore.cn/lite/docs/zh-CN/master/use/nnie.html