I'm new to python and the onnxruntime. I have successfully managed to do inferencing. I'm currently trying to direct an inference instance to a specific gpu on my system via IObinding. I'm receiving the following error when attempting to create an OrtValue: AttributeError: module 'onnxruntime' has no attribute 'OrtValue'. I've not found a solution as of yet. Hoping someone can shed some light on the matter. Here is some information.
Environment
OS: Windows 10
Python version: 3.7.8
OnnxRuntime-gpu: 1.5.2 install via pip
Cuda version: 10.2
CudNN vesion: 8.03.33
Code in Error:
X_ortvalue = ort.OrtValue.ortvalue_from_numpy(ort_inputs, 'cuda', 0)
It looks like
OrtValue
is a very recent addition (https://github.com/microsoft/onnxruntime/commit/6f54113a1bcea4e643de844ee33e4914145915ef) and is not in any release yet.You can build onnxruntime from source or wait until the feature officially released.