problem with onnxruntime-gpu

[plomtb]

Hi, I'm new to shiv. I want to create and distribute an application based on the windows onnxruntime-gpu. Is that possible?
I tried
shiv onnxruntime-gpu onnx numpy -o app.pyz --site-packages .tox\dev\Lib\site-packages. It can run my code, but there are problems using the gpu:

[E:onnxruntime:Default, provider_bridge_ort.cc:940 onnxruntime::ProviderLibrary::Get] LoadLibrary failed with error 126 "Module not found." when trying to load "C:\Users\user\.shiv\app4_aab9e7b430e1e6fc362601dd6923e56b8e956dcc76d880edcb2c463e72b4fcc6\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

Is it possible to resolve that?
Thanks

[lorencarvalho]

Hi @plomtb,

I definitely want to fully support Windows, however I must admit I haven't used a Windows machine in many years and I don't have ready access to one for testing/debugging (plus, I'm not very well versed).

Does the dll file that the error mentions actually exist? I'm unfortunately at a loss as to why this error would be occurring.

[illtellyoulater]

@plomtb I'm getting the same error you reported... Did you find anything about onnxruntime_providers_cuda.dll ?

[plomtb]

Hi, sorry that I forgot to answer.
I think, in my case, the problem was missing dependencies. I wanted to build an environment, that includes everything. Since pip is not able to install cudnn,that did not work. So it was not really a shiv issue (for me). For my usecase, conda-pack seems more appropriate and I did not really try it further.