How to install tensorflow 2.0 on Mac or Linux?

Question

TensorFlow 2.0 is released recently. There are no clear steps on how to install it. Any help with the steps for installation is appreciated.

Kronfeld

did you check the official page of Install TensorFlow with pip? – Agribusiness

Update: the steps for installation and testing is tensorflow 2 is found here towardsdatascience.com/… – Kronfeld

conda install -c anaconda tensorflow installs. But the GPU version depends on NVIDIA drivers. I have a AMD card and that isn't supported. – Currin

Answer 1

Four other developers have the same problem. The last comment is June 3rd. I eventually sent emails to all the email addresses of the contributors I could find listed at https://github.com/apple/tensorflow_macos, but none of them replied. The repository is now read-only, and archived, allowing no more comments. https://github.com/apple is Apple's official github page.

The world's first trillion-dollar company is unable to support one of its claims about its new computer - that it can be used to do machine learning in Python using Tensorflow. Contrast this with my experience of the open-source programming language Elixir. Every question I have asked at https://elixirforum.com/ got answered within 24 hours by the creator of the language, Jose Valim.

After five months, I did eventually get Tensorflow working as advertised on my Mac M1.

Thanks https://towardsdatascience.com/installing-tensorflow-on-the-m1-mac-410bb36b776

I needed to download from https://github.com/conda-forge/miniforge/releases and run Miniforge3-MacOSX-arm64.sh several times to get it to install a lot of software in /Users/rod/miniforge3

See also https://medium.com/gft-engineering/macbook-m1-tensorflow-on-jupyter-notebooks-6171e1f48060 who made the following useful comment:

"After publishing the article, someone reached me with this error when installing TensorFlow: ERROR: tensorflow_macos-0.1a3-cp38-cp38-macosx_11_0_arm64.whl is not a supported wheel on this platform. This happened because Anaconda was installed. As I describe above, Anaconda is not ARM compatible, so the most direct way to solve this is to uninstall Anaconda and install Miniforge instead."

$ brew uninstall anaconda

This made all my other python programs not work, so I had to redirect a series of soft links to point to /Users/rod/minforge3/bin.

However, using Tensorflow still involves using a version of Anaconda - this one:

/Users/rod/miniforge3/bin/conda

And you do conda activate apple_tensorflow before running a Tensorflow program.

The program won't run like this:

/Users/rod/miniforge3/bin/python my_tensorflow_program.py

Nor will it from from the command line if you put

#!/Users/rod/miniforge3/bin/python

at the top.

But if you put

#!python

at the top, it figures it out.

$ conda activate apple_tensorflow
$ which python
  /Users/rod/miniforge3/envs/apple_tensorflow/bin/python

This is Python 3.9.6

I also got help from https://medium.com/codex/installing-tensorflow-on-m1-macs-958767a7a4b3

At this point, I have to admit part of the delay was my bad. I didn't read the instructions clearly. You have to do the following, in this order, in folder miniforge3/ -

conda env create environment.yml
conda activate apple_tensorflow
pip install --upgrade --force --no-dependencies https://github.com/apple/tensorflow_macos/releases/download/v0.1alpha3/tensorflow_macos-0.1a3-cp38-cp38-macosx_11_0_arm64.whl https://github.com/apple/tensorflow_macos/releases/download/v0.1alpha3/tensorflow_addons_macos-0.1a3-cp38-cp38-macosx_11_0_arm64.whl```

conda install notebook -y
conda install matplotlib -y
conda install pandas -y
conda install scikit-learn -y
jupyter notebook

PS. I got shut out of developer.apple.com for a month. I can't remember why. I just got back in, on October 15th 2021. As expected, there are no answers to my questions there. https://developer.apple.com/forums/thread/686926

If I were the CEO of a commercial company which had bought 50 Apple M1s to do ML, I'd send them back, and buy PC's with NVidia GPUs, and put Linux on them. As it is, I'm just doing AI for fun. And what fun it is, now I've finally got it working, no thanks to Apple.

Presbyopia

developer.apple.com/metal/tensorflow-plugin gets you a bit further - it doesn't fail til you get to "python -m pip install tensorflow-macos". --> "ERROR: No matching distribution found for tensorflow-macos". Way to go, guys. – Presbyopia

Perhaps the Apple Developer Forum will be able to help. developer.apple.com/forums/thread/686926 - fingers crossed. – Presbyopia

So I tried developer.apple.com/forums/thread/686926. The answer I got said I was using Python 2. I pointed out that I'm using Python 3.8.5. No more answers yet. – Presbyopia

I also got stuck at python -m pip install tensorflow-macos. In my case, things start go wrong at "Building wheels for collected packages: grpcio", then there are too many errors and the process exits with "Failed to build grpcio". – Doughnut

I finally got it working, thanks to towardsdatascience.com, no thanks to Apple: conda env create environment.yml conda activate apple_tensorflow pip install --upgrade --force --no-dependencies github.com/apple/tensorflow_macos/releases/download/v0.1alpha3/… github.com/apple/tensorflow_macos/releases/download/v0.1alpha3/…``` conda install notebook -y conda install matplotlib -y conda install pandas -y conda install scikit-learn -y jupyter notebook – Presbyopia

It's now March 20th 2022, and someone just posted a helpful comment at developer.apple.com. It says this tutorial works: caffeinedev.medium.com/… – Presbyopia

Thank you so much. This worked and it was the only one that worked! – Bargain

Answer 2

For anyone still struggling with installing Tensorflow>=2.0 on MacOS, I was able to do the following:

pip install tensorflow-macos

Make sure you have the latest Numpy version though, otherwise you might run into other problems:

pip install numpy --upgrade

Guan

This helped me a lot. A lot of outdated instructions out there. – Ame

Answer 3

Thanks for this. I tried following the latest instructions.

Following the instructions at https://github.com/apple/tensorflow_macos In the Terminal on a new Mac M1:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/apple/tensorflow_macos/master/scripts/download__install.sh)"

Confirm y/N? y Installing and upgrading base packages.

Then it says

TensorFlow and TensorFlow Addons with ML Compute for macOS 11.0 successfully installed. To begin, activate the virtual environment:

. "/private/var/folders/6c/56kflzvn7vzcm7vx4kpnw0d00000gn/T/tmp.TNKnwmCZ/tensorflow_macos/tensorflow_macos_venv/bin/activate"

Below /private/var/folders/6c/56kflzvn7vzcm7vx4kpnw0d00000gn/T, the next folder, tmp.TNKnwmCZ, does not exist and /private is owned by root, so executing anything inside it, even if it does exist, requires ‘sudo’

Trying the other way of installing Tensorflow given at https://github.com/apple/tensorflow_macos

curl -fLO https://github.com/apple/tensorflow_macos/releases/download/v0.1alpha2/tensorflow_macos-${VERSION}.tar.gz
tar xvzf tensorflow_macos-${VERSION}.tar
cd tensorflow_macos
./install_venv.sh --prompt

It's not clear what VERSION is supposed to be, and https://github.com/apple/tensorflow_macos/releases/download/ does not exist, so the first of the above commands fails, whatever you set VERSION to

Trying another way: git clone https://github.com/apple/tensorflow_macos

cd tensorflow_macos/scripts

./download_and_install.sh 

It tells you to run

. "/private/var/folders/6c/56kflzvn7vzcm7vx4kpnw0d00000gn/T/tmp.xVyjLM93/tensorflow_macos/activate"

and again, these folders only exist down to T, and they are all owned by root

---

So I tried the method recommended above - conda install tensorflow

I found the conda executable in /opt/homebrew/anaconda3/bin

conda install tensorflow

It failed because

• tensorflow -> python[version='2.7.|3.7.|3.6.|3.5.']

Your python: python=3.8

Apple insists that their tensorflow depends on Python 3.8 https://github.com/apple/tensorflow_macos

Presbyopia

It finally worked for me when I updated pip to latest version (rn 22.0.3) – Propinquity

Answer 4

In case you are struggling to install Tensorflow on a mac M1 computer, I highly recommend taking a look at this. It works like a charm: https://www.mrdbourke.com/setup-apple-m1-pro-and-m1-max-for-machine-learning-and-data-science/

Songster

Answer 5

Emmmm, actually is quite easy, though the tensorflow 2.0 is still in experimental stage. It is better to install the preview version in a virtual environment with anaconda, now the following is a Linux example:

$ conda create --name tensorflow_2_0
$ conda activate tensorflow_2_0
$ pip install tf-nightly-2.0-preview # tf-nightly-gpu-2.0-preview for GPU version

Now let's have a try:

$ ipython # or ipython3/python/python3(at least one of which will work)
>>> import tensorflow as tf
>>> print(tf.__version__)
2.0.0-dev20190129

Done.

----------------2019.10.01 Update---------------
Today, the official tensorflow 2.0.0 is published. You can find how to install it here.
Simply, still activate your conda virtual environment and upgrade your pip first with:

pip install --upgrade pip

Then upgrade your tensorflow directly to 2.0.0:

pip install --upgrade tensorflow  # tensorflow-gpu for gpu verison

Finally:

>>> import tensorflow as tf
>>> tf.__version__
'2.0.0'

Tepee

Answer 6

TENSORFLOW 2.0 release version is out!

Since 01/10/2019 I'm not talking beta but the release version.

Using Anaconda (Recommended)

Since 01/11/2019 Anaconda is supporting the Tensorflow 2.0.0.

Option 1: For what the easiest way is just:

conda install tensorflow or conda install tensorflow-gpu

For the gpu mode, anaconda will take care of all the CUDA everything you need to install for the tensorflow gpu mode to work so I strongly recommend using this method.

Option 2 (virtual env): It is strongly recommended to use an environment on where to install tensorflow, for which you need the following command that will create an environment first and then install tensorflow within:

• CPU: conda create -n <your_env_name> tensorflow
• GPU: conda create -n <your_env_name> tensorflow-gpu

Change <your_env_name> by a meaningful name like tf-2

To use tensorflow run first conda activate <your_env_name>

Using pip

Using pip the tensorflow official instructions are quite complete.

Just install tensorflow using pip like:

# Current stable release for CPU-only
pip install tensorflow

I yet recommend before doing everything to install tensorflow in a new environment so the 3 steps would be (with anaconda):

conda create --n <our_env_name> pip
conda activate <your_env_name>
pip install tensorflow

Now for the GPU version it's harder with pip, I recommend you this link

Normanormal

Answer 7

brew install tensorflow

Even after above command, tensorflow is not installed, Then install via pip:

pip install tensorflow-macos

Halfbound