Easier Cuda + Cudnn Install Process

jradz · December 6, 2019, 11:35pm

Hello,
I’m from the Microsoft Model Builder .NET team: ML.NET Model Builder | Machine learning in Visual Studio .

We’re looking to enable GPU Training but we’re finding many users have difficulty getting through the setup process. Currently the user has to go through these steps:

github.com

dotnet/machinelearning/blob/main/docs/api-reference/tensorflow-usage.md

## Using TensorFlow based APIs
In order to run any TensorFlow based ML.Net APIs you must first add a NuGet dependency
on the TensorFlow redist library. There are currently two versions you can use. One which is
compiled for GPU support, and one which has CPU support only.

### CPU only
CPU based TensorFlow is currently supported on:
* Linux
* MacOS
* Windows

To get TensorFlow working on the CPU only all that is to take a NuGet dependency on
SciSharp.TensorFlow.Redist v1.14.0

### GPU support
GPU based TensorFlow is currently supported on:
* Windows
* Linux
As of now TensorFlow does not support running on GPUs for MacOS, so we cannot support this currently.

This file has been truncated. show original

The installers that those docs reference seem to have some issues. They try to install a graphics driver that isn’t signed or has an invalid signature. So the user then has to un-install and re-install a different graphics driver. It’s an odd complication for many users to work around.

Is it possible for us to re-distribute Cuda installer and cuDNN to make the process easier for our users? Who could I reach out to for more information?

Cheers,
Jake