Installing R + Tensorflow on M1

installation.md

Install native ARM R: https://cloud.r-project.org/

Install RStudio (probably preview) which is not native as of today: https://www.rstudio.com/products/rstudio/download/preview/

This blog post is helpful for getting Tensorflow working, but it is older, before the ARM version of R and the RStudio version that supports it was available.

This other blog post has specific steps outlined, but some of it has also been superseded.

Looks to me like Apple's instructions are the best as of today: https://developer.apple.com/metal/tensorflow-plugin/

Here is what I think you need to do:

• Make sure you have Python 3.8 or Python 3.9
• Install Miniforge for ARM, using reticulate::install_miniconda() from within R
• Create conda environment for your TensorFlow install (I called it tf_env and have had better luck doing it from the command line)
• Install packages like numpy, pandas, etc for ARM (try it with conda install -c apple tensorflow-deps from Apple's instructions? is that good enough?)
• Install base Tensorflow (I have not tried the tensorflow-metal yet)

You can turn it on (in the terminal) with conda activate tf_env and turn it off with conda deactivate. I have this in my .Renviron:

RETICULATE_PYTHON = "~/miniforge3/envs/tf_env/bin/python"

@pmgarafola Can you please file an issue on https://github.com/rstudio/tensorflow
The double 'bin/bin' in /usr/local/bin/bin/python indicates something is very wrong with your python setup.

In most situations, this is all you need these days to get Tensorflow working in reticulate on an M1 Mac:

install.packages("remotes")
remotes::install_github(sprintf("rstudio/%s", c("reticulate", "tensorflow", "keras")))
reticulate::miniconda_uninstall() # start with a blank slate
reticulate::install_miniconda()
keras::install_keras()

I dont think they have an m1, shows x86 in trace —h

…

On May 1, 2022, at 6:31 AM, Tomasz Kalinowski ***@***.***> wrote: ***@***.*** commented on this gist. @pmgarafola Can you please file an issue on https://github.com/rstudio/tensorflow The double 'bin/bin' in /usr/local/bin/bin/python indicates something is very wrong with your python setup. In most situations, this is all you need these days to get Tensorflow working in reticulate on an M1 Mac: install.packages("remotes") remotes::install_github(sprintf("rstudio/%s", c("reticulate", "tensorflow", "keras"))) reticulate::miniconda_uninstall() # start with a blank slate reticulate::install_miniconda() keras::install_keras() — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.

oh, thank you @hcarstens, good catch. It looks like the "tensorflow-macos" pip package is installed, but it fails because R and miniconda are running under Rosetta. @pmgarafola, you need to install the arm64 build of R

ok, Installed the arm64 version of R. After repeating the steps above starting with Keras install still does not work. I wonder if this is a pathing issue or did I install the mac-os version of tensor-flow in the wrong place? I will file an issue per your request. Thank you for the help.

Hi @t-kalinowski - Thanks for providing the steps above. I ran them successfully in my mac m1. And, I am now running some models in RStudio (ebecht/infinityFlow#13).

In my Activity Monitor, it does show GPU % below and when I tried to run some functions in R-Studio, it shows below. Does that means it is working, even though the warning "Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support" is still there?

Thank you for your help!

Related: rstudio/tensorflow#555

If you review the issue on git hub cited below you will see that I was able to successfully install tensorflow on the mac using instructions provided in the issue thread, and the issue is closed.

Unable to install tensor-flow in R-Studio on m1 mac #530

Thanks @pmgarafola (rstudio/tensorflow#530)