Apple M1 support

[SaschaHeyer]

---

Feature Request

Describe the problem the feature is intended to solve

TensorFlow is promoting Apples M1 Macs, would be great to have TFServing running on M1 Macs as well
https://blog.tensorflow.org/2020/11/accelerating-tensorflow-performance-on-mac.html

[netfs]

Once M1 is officially supported by TF, we can try and provide (docker/m1) builds for TF Serving.

Presently M1 builds in TF is driven by community supported builds.

[1vn]

This would be much appreciated. I almost got TF Serving running on m1 in minikube (using this image https://github.com/emacski/tensorflow-serving-arm) but hit this error on start:

2021-05-31 12:46:35.956430: F tensorflow/core/lib/monitoring/sampler.cc:42] Check failed: bucket_limits_[i] > bucket_limits_[i - 1] (0 vs. 10)

[coreation]

Once M1 is officially supported by TF, we can try and provide (docker/m1) builds for TF Serving.

Presently M1 builds in TF is driven by community supported builds.

Hi @netfs do you happen to know where a tensorflow serving community build can be found?

[lzuwei]

Is there an ETA for this to become available? This is blocking M1 Mac users from running tfs on docker for development.

[jzamalloa1]

Bumping along with TFX support for M1

[sdchc66]

Any updates or ETA on this?

[MarioNavarrete]

Any updates on this one?

[GautamSinghania]

Any updates on this one?

[fabienric]

Hi there,

I'm able to run Tensorflow Serving on my M1 by using this custom build: https://github.com/emacski/tensorflow-serving-arm

docker pull emacski/tensorflow-serving:latest-linux_arm64

docker run -t --rm -p 8501:8501 --mount type=bind,source=/tmp/model_name/,target=/models/model_name/ -e MODEL_NAME=model_name emacski/tensorflow-serving:latest-linux_arm64

(replace /tmp/model_name by your model's directory)

[magedhelmy1]

For those who are stuck, check out

Solution 1:

FROM emacski/tensorflow-serving:2.5.1
....your logic here...

Alternative (if you want to install TensorFlow in docker):

FROM --platform=linux/x86_64 python:3.9
RUN python -m pip install --upgrade pip
RUN pip install tensorflow==2.6.2

[gaikwadrahul8]

We are currently working on this issue, and will have an update in the fairly near future. In the meantime, some users have reported success with Rosetta. Other options include using a VM. We understand that neither of those is ideal.

At the moment we don't have ETA but you can check update about Mac M1 support on Tensorflow Forum here

Thank you!

[coreation]

Thanks for the update @gaikwadrahul8

[glynjackson]

Any update on this?

[glynjackson]

I want to thank you for this comment! I managed to get this working with the above on my M2 Mac for local development.

[benzitohhh]

Hey any update on this? Thanks!

[sdchc66]

Any update on this ?

[gcuder]

Any updates on this?

[datduonguva]

more than 3 years and TF serving is still not working. Google, please do better!

[ConfuzedCoder]

Is there any ETA for this?

[sd3ntato]

shoot it's two years now!

[robtare]

I was trying to find a solution for this and randomly stumbled on this report... does this really not work? (M3s are out now)

[axeltidemann]

Bitnami provides ARM64 images for TensorFlow Serving. I have been using them for some time for Apple M1 Macs, they work well. Unfortunately, there are some minor differences in the run arguments compared to the official TensorFlow Serving image, so it is not a direct drop-in replacement, but it is not hard to figure out.

[gcuder]

Update on this topic. The newest Docker release 4.35.0 (172550) for Mac introduces Docker VMM Beta, a replacement for the Apple Virtualisation Framework using Rosetta. Good news is that I can run the native TF Serving image now on my M2 Pro Macbook

See https://docs.docker.com/desktop/release-notes/#4350