Decoupling loading from inference models on multiprocessing

Hi all,

I’m working on a scheduler to allocate image detection inference on either the GPU or CPU. For this, I previously load the model to an object and then start processes for CPU and GPU respectively. When I try to do the inference on the GPU (with cuda), I get a cuda runtime error (801): pytorch\torch/csrc/generic/StorageSharing.cpp:245

I’m not sure why this happens when I decouple the loading from the inference on the GPU. Do you have any idea, how can I solve this?

My setup is:
Python 3.6
Jetson TX1
Pytorch 1.10 - I’m using multiprocessing for a distributed environment.

I’d really appreciate your help!


You can find error information from our document below:
Error 801 indicates cudaErrorNotSupported.

cudaErrorNotSupported = 801
This error indicates the attempted operation is not supported on the current system or device.

Please note that CUDA IPC doesn’t support the Jetson platform.
So you might face a similar issue as below topic:


1 Like

Thanks for the reply! I will consider a different approach.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.