GPU Support

[yuzawa-san]

I originally distributed GPU artifacts, but I found these were hard to test, quite large, and depended greatly on the target hardware and things like CUDA. For that reason, I encourage users to bring their own onnxruntime native library. Install it via conda or whatever and point this library at that. I have tried to support as many of the execution providers as the C-API exposes, but I recommend folks make sure they are compiling onnxruntime with whatever their need for their advanced cases.

If you would like me to package up the GPU artifacts again, please use this issue to discuss.