@nicolaspi Can you check https://github.com/tensorflow/addons/runs/7999702173?check_suite_focus=true#step:6:5205

@bhack I am on it.

@bhack Please check #2751. It should fix it.

@bhack Please check #2751. It should fix it.

Thanks, let's see the new CI run..

🟢 Green-light here, waiting for the MacOS delayed release:
https://pypi.org/project/tensorflow-macos/#history

🟢 Green-light here, waiting for the MacOS delayed release:

Looks like 2.10.0 was just released yesterday: https://pypi.org/project/tensorflow-macos

I don't the story about arm and non arm MacOS diversification we had.
Can we align all to MacOS 12? Al the new tensorflow-macos wheels are 12 only.

I don't the story about arm and non arm MacOS diversification we had. Can we align all to MacOS 12? Al the new tensorflow-macos wheels are 12 only.

The tensorflow wheels are still supporting 10_14:
https://pypi.org/project/tensorflow/2.10.0/#files

This policy is confusing me a bit.
What are the differences between TF MacOS and Tensorflow-macos for X64? Do they have the same features?

As tensorflow-macos is MacOS 12 only for TF 2.10

This policy is confusing me a bit. What are the differences between TF MacOS and Tensorflow-macos for X64? Do they have the same features?

As tensorflow-macos is MacOS 12 only for TF 2.10

Same. tensorflow-macos just started adding x86 wheels on TF2.9. I'm not sure what optimization they provide for those chips or why you would install that package instead of TF. For now I'm just going to keep our macos_x86_64 built with macOS 10 since TF is still providing those wheels and I don't have justification for telling users to force an OS upgrade and switch to tensorflow-macos for non-ARM chipsets. They also don't build py37 wheels

Hopefully they align on what's supported.

For now I'm just going to keep our macos_x86_64 built with macOS 10 since TF is still providing those wheels and I don't have justification for telling users to force an OS upgrade and switch to tensorflow-macos for non-ARM chipsets.

Sounds great.
Just for reference since I am not sure how long GitHub will still support 10.15 macOS images: it is possible to build backwards compatible wheels on the latest macOS by setting --copt=-mmacosx-version-min=10.14 --linkopt=-mmacosx-version-min=10.14 as bazel flags or via the MACOSX_DEPLOYMENT_TARGET env variable.

It would be nice if @jhavukainen or @kulinseth could clarify on this.

For now I'm just going to keep our macos_x86_64 built with macOS 10 since TF is still providing those wheels and I don't have justification for telling users to force an OS upgrade and switch to tensorflow-macos for non-ARM chipsets.

Sounds great. Just for reference since I am not sure how long GitHub will still support 10.15 macOS images: it is possible to build backwards compatible wheels on the latest macOS by setting --copt=-mmacosx-version-min=10.14 --linkopt=-mmacosx-version-min=10.14 as bazel flags or via the MACOSX_DEPLOYMENT_TARGET env variable.

Ack, it appears to be imminent:
actions/runner-images#5583

Will upgrade the build OS, since the compilation settings are still here:
https://github.com/tensorflow/addons/blob/master/.github/workflows/make_wheel_macOS_x86.sh#L19-L20