TensorFlow on Jetson TX2

To Andrey1984.
Sorry for this. My previous comment has the mistake.
Please try to install the numpy.

➜  ~ sudo pip install numpy
The directory '/home/nvidia/.cache/pip/http' or its parent directory is not owned by the current user and the cache has been disabled. Please check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
The directory '/home/nvidia/.cache/pip' or its parent directory is not owned by the current user and caching wheels has been disabled. check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
Collecting numpy
  Downloading numpy-1.13.1.zip (5.0MB)
    100% |████████████████████████████████| 5.0MB 134kB/s
Installing collected packages: numpy
  Running setup.py install for numpy ... \
done
Successfully installed numpy-1.13.1

after few minute, it may work.

Or try to uninstall all package installed by pip.

pip freeze | xargs sudo pip uninstall -y

Then reinstall the tensorflow again.

sudo pip install tensorflow-1.3.0rc0-cp27-cp27mu-linux_aarch64.whl

Thank you for the update!

>>> import tensorflow as tf
>>> print(tf.__version__)
1.3.0-rc0
>>>

Does it compile "bazel build tensorflow/examples/image_retraining:retrain " at your side?
It used to work with previous version.

can anyone share the bazel 0.5.2file?
i tried to compile the bazel and failed

I believe bazel 5.3 causes issues;

https://github.com/bazelbuild/bazel/releases/download/0.5.2/bazel-0.5.2-dist.zip

will likely not to cause issues, as far as I remember;

Quote:

Start the compilation process by issuing ./compile.sh
Copy the build to your system bin folder sudo cp output/bazel /usr/local/bin

Reference: https://syed-ahmed.gitbooks.io/nvidia-jetson-tx2-recipes/content/first-question.html

some updates

tensorflow 1.3.0 whl file for python3.5
https://drive.google.com/open?id=0Bz4bYWmfjaxeVWlPclVUelJ2LTQ

bazel 0.5.2 file if you want to build one for yourself
https://drive.google.com/open?id=0Bz4bYWmfjaxeSTFMOGotS3k4b1E

Does inception “retrain” works with Tensorflow 1.3 ?

In my case it resulted with errors and I downgraded to 1.2.1

Hello.

It seems my TX2 (R27.1) did not come with Cuda 8.0 and Cudnn 5.1 installed. Is this normal? As a result I’m not able to install tensorflow. Is there a way to install cuda and cudnn manually or only through jetpack? I have noticed that on the Nvidia webpage Jetpack 3.0 (R27.1) was removed, so I’m unable to download it.

Well, it should be fine, not sure what problem you had but may be caused by the rc version of tensorflow

the easiest way to install cuda and cudnn is through the jetpack. download jetpack 3.1 and install cuda 8 and cudnn 6

you may thick check-boxes at Jetpack , running it from the Host OS to select/ deselect components which are to be installed

My TX2 came with R27.1 (which would correspond to Jetpack 3.0), does that matter or can I still install cudnn 6 from Jetpack 3.1? Also, I do not currently have a workstation with Ubuntu 14.04 or a network switch. Is there any other way to install cuda and cudnn?

I think if you want to use Jetpack 3.0 - you rather should use the corresponding versions.
But if I were you - I would try to install new jetpack with updated versions.
However, In my case the upgrade resulted in some “monitor-vga-adapter” issue .
I do not think if you install cudnn from 3.1 to the 3.0 it would get coherent at once.

Problem is that the R27.1 Jetpack 3.0 Link on the Nvidia website is broken. Hence I cannot download it. Also if I upgrade with Jetpack 3.1 is this like flashing my device? I’m asking g since I have a lot of data and compiled programs on the TX2 that I don’t really wanna loose.

y may download 3.0 there:
https://drive.google.com/file/d/0B86kh-1GJW29TERCek50R184Unc/view?usp=sharing

Thanks a lot @Andrey1984. I’m trying it out right now.

Has anyone had success building Tensorflow 1.3.0 with CPU support only from source?

Hi,

Some users have built tensorflow 1.3.0 with GPU support successfully.
https://devtalk.nvidia.com/default/topic/1000717/jetson-tx2/tensorflow-on-jetson-tx2

If you want a CPU only version, this tutorial can give you some hint.
Remember to turn-off GPU when running ./configure.

Do you wish to build TensorFlow with CUDA support? [y/N] N

How did you build the tensorflow 1.3 on TX2? I keep running into different problems.
I was following the tutorial [url]https://syed-ahmed.gitbooks.io/nvidia-jetson-tx2-recipes/content/first-question.html[/url], and was building for python 3.5.