Skip to content

Latest commit

 

History

History
59 lines (45 loc) · 11.8 KB

README.md

File metadata and controls

59 lines (45 loc) · 11.8 KB

TensorFlow wheels for Raspberry Pi

output image

Find your operating system and TensorFlow version in the table below. Follow the instructions in the provided guide.


Roadmap.

Operating system TF 2.15.0 TF 2.10.0 TF 2.9.1 TF 2.8.0 TF 2.7.0 TF 2.6.0 TF 2.5.1 TF 2.5.0 TF 2.4.1 TF 2.4.0 TF 2.3.1 TF 2.3.0 TF 2.2.0 TF 2.1.0 TF 1.15.2
Raspberry Pi 32-bit Buster Wheel
C API
Guide
GitHub GitHub
Raspberry Pi 64-bit Buster Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Raspberry Pi 64-bit Bullseye Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Raspberry Pi 64-bit Bookworm C API
Raspberry Pi Ubuntu 18.04 Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Raspberry Pi Ubuntu 20.04 Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide
Jetson Nano JetPack 4.6 Wheel
C API
Guide
Wheel
C API
Guide
Wheel
C API
Guide

output image Find TensorFlow with other frameworks and deep-learning examples on our SD-image

Buster 32-bit OS

As the massive TensorFlow evolves, building it on a simple 32-bit machine is getting more and more difficult. Many tricks and workarounds are now required to compile bazel and TensorFlow. That's why the last version of TensorFlow for a 32-bit OS, is the 2.2.0 release.

Buster 64-bit OS

TensorFlow 2.7 and higher relies on libclang 9.0.1. There is no distribution available for Debian 10. That's why there is only a TensorFlow 2.7+ installation for Debian 11, Bullseye. You could probably install libclang 9.0.1 on your Buster RPi from scratch so that you can then install TensorFlow. Be aware, the clang build takes huge resources, over 5 GB. It's better to switch to Bullseye and have TensorFlow up and running in half an hour.

Ubuntu 18.04

TensorFlow 2.5.0 depends on h5py version 3.1.0. Unfortunately, the h5py version 3.1.0 cannot be easily installed on Ubuntu 18.04, or to be more precise, on an aarch64 with Python 3.6. See #1760. That's why we don't have a wheel for Ubuntu 18.04. Use TensorFlow 2.4.1 or switch to Ubuntu 20.04.

Jetson Nano

TensorFlow 2.5, 2.6 and 2.7 depend on CUDA 11.0 and cuDNN version 8.0.4, both not yet available for the Jetson Nano. A workaround is cumbersome and probably not very reliable. Better to wait for the new announced JetPack to be released with the required versions of CUDA and cuDNN. Continue to use TensorFlow 2.4.1 for now.


tensorflow-io-gcs

TensorFlow 2.8 and higher depends on the tensorflow-io-gcs file system. However, there is no aarch64 distribution. You have to install it on forehand manually with the following commands.

$ git clone https://github.com/Qengineering/Tensorflow-io.git
$ cd Tensorflow-io
$ sudo -H pip3 install tensorflow_io_gcs_filesystem-0.23.1-cp39-cp39-linux_aarch64.whl
$ cd ~

TensorFlow 2.10.0

You can run into protobuffer incompatibility problems when running TensorFlow 2.10.0. as can seen on the screendum below.

output image

You have to lower the protobuf version to 3.19.0 with the command:

$ sudo -H pip3 install --upgrade protobuf==3.19.0

You may need to remove to old protobuf 4.21.2 manually before 3.19.0 becomes the default protobuf version.
Once installed, TensorFlow will work flawless.

output image


paypal