Cross-build docker image

[liuqun]

Here's a feature branch that switch the base image to multi-architecture docker image:

• https://github.com/liuqun/rpi-docker-tensorflow/tree/crossbuild-armhf-stretch-py27

The docker contains:

• Docker base image multiarch/debian-debootstrap:armhf-stretch
• Python 2.7
• tensorflow-1.9.0-cp27-none-linux_armv7l.whl built by Google .
• numpy 1.14.5 built from original source package.
• Lots of pre-built python wheel packages from Raspbian's extra pip index repo https://www.piwheels.org/ .

It enables image cross-building from any 64-bit PC.

And I have uploaded another none-official docker image for raspberry pi 2/3 at:

• https://hub.docker.com/r/liuqun/rpi-docker-tensorflow

Welcome to test (I have only one Raspberry Pi 2 at hand. And I don't know whether my docker image works on other device or not.)

docker pull liuqun/rpi-docker-tensorflow:1.9.0

Known bugs:

1. Not compatible with numpy-1.15.0rc2 ~ numpy-1.15.0: numpy.dtype size changed, may indicate binary incompatibility. Expected 56, got 52
2. I tried to upgrade to python-3.5, but got warnings that in tensorflow-1.9.0 or earlier version module 'tensorflow.python.framework.fast_tensor_util' is compiled with python-3.4. The following warning messages would be fixed in tensorflow-1.10.x release. See issues Submit Raspberry Pi releases to piwheels.org tensorflow/tensorflow#18202 and Missing package: tensorflow piwheels/piwheels#41

Python 3.5.3 (default, Jan 19 2017, 14:11:04)
[GCC 6.3.0 20170118] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
/usr/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: compiletime version 3.4 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.5
/usr/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: builtins.type size changed, may indicate binary incompatibility. Expected 432, got 412

[antoniosap]

and for tensorflow 1.12 ?

[romilly]

This project was originally an experiment to explore the practicality of running Sam Abrahams' Tensorflow port in a docker container on the Pi. These days there is an up-to-date official Tensorflow wheel, and I no longer see value in running Tensorflow inside a container on the Pi. Therefore I do not plan to develop or support this project further. If anyone wants to do so there should be enough information here to enable then to build their own docker image.

…

On Wed, 5 Dec 2018 at 10:44, antoniosap ***@***.***> wrote: and for tensorflow 1.12 ? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#9 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAJHPt2yfS5p4mRaCL7PZb3kDs6SBzD4ks5u16OQgaJpZM4VfbaF> .

-- personal:@romillyc work:@rareblog skype:romilly.cocking web: http://blog.rareschool.com/

[romilly]

The project has risen from the dead, as it's now realistic to run TensorFlow on a Pi 4 with 4 GB of RAM. I do not intenb to create cross-builds for X86, as the current build takes a reasonable me on Rasbia/Pi 4.