Cannot install tensorflow-macos from conda miniforge?-CodePudding

As title, I've installed miniforge3 with pyenv and try using it to create a virtual environment with tensorflow-macos package, which can be found from the pypi website: search result. I have tried three different python versions 3.8.12, 3.9.10, 3.10.2 and have .condarc as follow:

channels: [apple, conda-forge, pypi]
auto_activate_base: false

the error all the same:

Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.

PackagesNotFoundError: The following packages are not available from current channels:

  - tensorflow-macos

Can anyone explain why the package tensorflow-macos can be found on pypi but I still cannot install it by conda with pypi channel added?

CodePudding user response：

Tensorflow it difficult even under native Linux operating systems (Ubuntu YECH!). I would suggest you dual-boot into a Linux distro of your choice and setup Docker for a simple solution... unless you REALLY want to endure pain.

Here is the documentation from NVidia: https://github.com/NVIDIA/nvidia-docker

Docker makes this stuff a lot easier!

CodePudding user response：

Since tensorflow-macos is platform specific it's no need to install it via conda miniforge3 channel for compatibility, just follow the guide from Apple's tensorflow-metal PluggableDevice guide to further utilize your M1(or AMD) GPU.

Notice that you have to use python 3.8.x to run the following command:

pip install tensorflow-macos tensorflow-metal