when i want use a model for multi threading, use threading with lock threading in python. When predict from thread another, my memory leak with ram gpu from 2GB -> full
I see error leak memory in mxnet merged from 7/2019 but don’t change in that situation
thank you!
Mxnet does not support threading as far as I know. The cpp backend already takes care of parallelization. If you need threading for model serving take a look at the mxnet model server on github.
thank you very much! i running, my system still active with threading python, but error leak memory.
i will survey serving mxnet in github