🔧 Add Apple Silicon (M1) Compatibility — Dependency & API Updates

[abmallick]

Summary
This PR adds support for Apple Silicon by introducing OS-specific TensorFlow dependencies. It also updates the code and dependency versions to work with newer versions of TensorFlow, NumPy, and related libraries.

Changes Made
✅ requirements.txt
Added OS-specific TensorFlow packages:

tensorflow==2.16.2 for Linux
tensorflow-macos==2.16.2 and tensorflow-metal==1.2.0 for macOS ARM64

Updated core libraries:

numpy==1.26.4

tf2onnx==1.16.1

onnx, onnxruntime, scikit-learn, and others to latest compatible versions

✅ models_utils.py
Removed deprecated import:

from keras.utils.layer_utils import count_params

Replaced with direct Keras API:

model.count_params(model.trainable_weights)
model.count_params(model.non_trainable_weights)

Tested On
macOS (Apple Silicon M1)
Linux (Ubuntu x64)

[fazenrazen]

This is awesome!!! wow does it work with the new ST Edge 2.1? And the newer 3.1 version of Model Services?

[abmallick]

@fazenrazen yes, just tested on Ubuntu as well. Just ensure you are using python 3.10.13.

[fazenrazen]

SUPER COOL thank you very much! Which boards were you deploying to. I am just curious! I have the N6 board and the Mac OS framework is just really great so I was wondering if that everything was possible from the mac computer.

[abmallick]

I am fairly new to embedded systems, N6 dev kit is my first one xD Got it yesterday and have been trying to get a NN running on it. Let us connect on Discord abmallick#7531

[fazenrazen]

I am trying to make a discord of all the people working on N6 stuff, i sent you a friend request!!

[lucabrini]

Works great on M4 Pro