What is gpu_0/data?

I build GLOW and now want to test in GPU mode. Recently I work with image-classifier tool but I failed to understand what is


I pass ONNX model file and PNG picture what role model-input-name plays?

Fill line is

./bin/image-classifier tests/images/imagenet/*.png -image-mode=0to1 -m=resnet50 -model-input-name=gpu_0/data?

I found fix using lldb debugger if I pass


instead it works in my case.

How to specify appropriate -model-input-name when using -input-image-list-file option of image-classifier tool?

If use gpu_0/data_0 I will get

name : OC2_DUMMY_0
Input : float<9 x 2048 x 1 x 1>
Dims : [1, 2048]
users : 1
Result : float<1 x 2048>

Reshape into a different size
For comparison `LHS Equal RHS` with:
LHS: 2048
RHS: 18432
Encountered an error, exiting.
location: ../lib/Importer/ONNXModelLoader.cpp:1265 message: Function verification failed.

Which is not good

Please see this comment here and the couple after it about the model input name. It has nothing to do with the input image list – it’s the name of the input in the protobuf representing the model itself.

Hello, How do I run glow on a GPU?

./bin.mnsit -backend=OpenCL doesnt seems to work. I have a nvidia gpu in my computer.

When you say it doesn’t work, does that mean it executes but doesn’t appear to execute on the GPU, or it doesn’t execute at all?

If it’s the former, OpenCL allows you to target many different devices on your machine. You may need to specify which device you want to target via --device=[integer>=0], e.g. --device=1.