Home

Introduction

Tensor/IO is declarative, On-Device Machine Learning for iOS, Android, and React Native, with support for packaging models, deploying them, prediction, training, and federated learning.

TensorFlow Compatibility

If you are using the TensorFlow backend with Tensor/IO and require a specific build of TensorFlow, use this table to lock Tensor/IO to a version that targets the desired build of TensorFlow.

Tensor/IO	TensorFlow Build	Compatible with Models Built in TF
v1.0	v1.13	v1.13 - v1.15
v1.1	v1.15	v1.13 - v2.0
v1.2	v2.0	v1.13 - v2.2

For example, use the following in your Podfile to lock yourself to the TF 1.13 backend:

pod 'TensorIO', '~> 1.0.0'

That version specifier ensures you will update to the latest Tensor/IO 1.0.x builds, which use TF 1.13, but never to Tensor/IO 1.1 builds, which use TF 1.15.

To lock yourself to the 1.15 backend, use:

pod 'TensorIO', '~> 1.1.0'

And so on.

Patch updates with a major.minor release will continue to target that version of TensorFlow.

Our v2.0 on-device build of TensorFlow will be the last major.minor update to it, as all support for building the full version of TensorFlow for iOS has been removed from TF v2.1 on.

Roadmap

See Project Roadmap.

Submitting an Update

See Releasing an Update.

~

Last Updated 6/5/27