Pipenv failing to install tensorflow-macos and tensorflow-metal during deployment

During deployment, pipenv is installing all but two of my dependencies: Tensorflow-macos and Tensorflow-metal. I’m also getting the following (seemingly) non-fatal message, which may be some sort of clue …

Ignoring appnope: markers ‘platform_system == “Darwin”’ don’t match your environment

I’m guessing the install errors are somehow due to a macOS (or specifically M1) related issue, but I don’t know whether there’s a straightforward / obvious solution that doesn’t involve me using Rosetta on everything from now until the end of time? :cry:

Hi @stephen-crone,

Thanks for posting!

Since tensorflow-macos and tensorflow-metal are built for MacOS, you likely need to use tensorflow in your app if you’re deploying it with Streamlit Cloud

Caroline :balloon:

Hi @Caroline. Thanks very much for your reply. Switching out Tensorflow-macos and Tensorflow-metal for ‘vanilla’ Tensorflow seems to have done the trick – the app deployed from GitHub without errors this time :slight_smile:.

2 Likes