As Jeremy said, Final two lessons of this course will be on Swift for Tensorflow. So I wrote this blog post as a Guide to install Swift-TF on your ubuntu local machine or VM
https://vibhuthasak.github.io/posts/swift-tensorflow-guide/
Thanks, I will be reading this later. Have definitely been having trouble with this myself.
Hey. Thanks for the blogpost. It would be great if you could post this in the harebrain category also. People who are interested in learning swift checkout that regularly.
Did you run into any issues with the kernel not being able to find your CUDA install? Initially, I copied the cudnn lib files to my swift-toolchain dir (as in my Medium post); however, on Jeremy’s suggestion, I set the LD_LIBRARY_PATH
env var (as below) and that offered a simpler solution.
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
Also, in your post, you offer a solution to verifying your install if your PATH
env var isn’t set, but you don’t give the instruction as to how to set PATH
such that nvcc
will work from any dir. I suggest adding something like this to your post:
export PATH=/usr/local/cuda/bin:$PATH
You could even suggest people add it to their .bash_profile
or .zshrc
.