[BUG] Mac M1 environment unable to use zenml CLI

[liguodongiot]

How should I solve this problem?

zsh: illegal hardware instruction zenml

[htahir1]

Could you please add the full traceback as well as the ZenML version and MAC OS version?

[liguodongiot]

Thank you very much for your answer.

My installation steps：

pip  install  -i  https://pypi.doubanio.com/simple/  --trusted-host pypi.doubanio.com  zenml
pip install pytorch-lightning==1.1.8

# add this to ~/.zshrc
eval "$(_ZENML_COMPLETE=source_zsh zenml)"

zenml version

My environment：

mac os version: 11.1

zenml version: 0.3.8

conda version: 4.9.2

python version: 3.8

[htahir1]

Could try running this script:

from zenml.version import __version__
print(__version__)

And let us know what you get?

[liguodongiot]

Running this script is normal.

python

Python 3.8.5 (default, Sep  4 2020, 02:22:02) 
[Clang 10.0.0 ] :: Anaconda, Inc. on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from zenml.version import __version__
>>> print(__version__)
0.3.8

[htahir1]

As it seems to be working installation wise, I feel like the CLI problem might be due to a zsh configuration rather than a ZenML installation one. Were you able to run zenml init on the CLI? Could you try alternatively using simple bash than zsh?

[liguodongiot]

Run zenml init on the CLI, same error.

When I use bash, I still get an error：

[htahir1]

Thank you for the info. I feel like this an underlying problem with Tensorflow on Mac M1. See the following issues:

apple/tensorflow_macos#143
ludwig-ai/ludwig#1101
openeventdata/mordecai#97

Is it possible for you to try the fixes mentioned here apple/tensorflow_macos#143 and let us know if that works? Otherwise I'd ask to please wait until we get rid of the tensorflow dependency (circa ~1 month) for our big refactored release and it should then work!

[liguodongiot]

Hi, there is a version conflict in installing zenml using tensorflow_macos virtual environment。

---

[htahir1]

This will be solved when we get rid of the Tensorflow dependency. Another suggestion: Can you try using our Docker Image instead to run ZenML: docker pull zenmldocker/zenml

[liguodongiot]

Hi, I tried this way, but I still get an error:

[htahir1]

This is strange - it seems like a segmentation fault in the C++ core of tensorflow now. It seems like they have just stopped supporting Mac altogether: tensorflow/tensorflow#42387

Unfortunately at this point, I would suggest you wait for our updated release as it will get us away from the Tensorflow dependency. Sorry for the inconvenience here!

[htahir1]

Closing due to staleness, and hopefully solved in the new release.