TensorFlow Metal 1.1.0 and TensorFlow > 2.15

What issues are you having and what have you tried to resolve them?

@ianlokh tensorflow-metal is now at version 1.2.0 (https://pypi.org/project/tensorflow-metal/).

I was pleased to discover that today when updating some dependencies, as I was nearly certain the development had stopped.

That being said, I'm running it on M1, and haven't tried on M3.

@jzombie123 do you have insights to share that might help @ianlokh?

Otherwise thank you for pointing out version 1.2.0.

My expectation is that using M1 or M3 won't differ much except in terms of execution times.

Hello, thanks to your question I came to the forum looking for a solution for a (local) This verification I did on a M1Pro::

You can use the following lines as a guide:

conda create -n <Environment Name> python 3.11.11 or 3.12.9 
conda activate <Environment name> 
pip install tensorflow==2.17 or 2.18 tensorflow-metal 
conda install <other package name>

It will probably automatically install tensorflow-metal 1.2 to check versions and installation status:

python --version
python -c “import tensorflow as tf; import keras ; print(tf.__version__); print(keras.__version__)”
 python -c “import tensorflow as tf; print(tf.config.list_physical_devices(‘CPU’))”
 python -c “import tensorflow as tf; print(tf.config.list_physical_devices(‘GPU’))”

In the tensorflow documentation from version 2.16 onwards it is recommended to use

pip install tensorflow 

and not other methods.

If you use conda or poetry an error occurs when using TF-2.17-2.18 with TF-Metal1.2 with python3.11 or 3.12 that actually prevents loading TF in the new versions

I confirm that you still need to install tensorflow-metal version1.1 or the new version1.2 (for TF and GPU usage in Apple Silicon)

Indeed it is stable TF2.18, Keras2.8 and Python 3.12 that I have been testing recently.

You should see a similar result:

Hi everyone, to round back on this, I was able to run TF 2.18 with Metal 1.2.0 successfully on M3.