You just finished looking at TensorFlow on NVIDIA Jetson TX1 Development Kit and are wondering, “Is there anyway that I could actually use TensorFlow on the Jetson TX1 without having to compile the whole thing myself?” The official JetsonHacks response? “No. Suck it up!”
This is an unofficial post, just a point of information that perhaps a clever person like you can use to your advantage.
Actually, I happen to have placed the Wheel file that we created in the previous article on Google Drive. No guarantees about it’s suitability (or even that it works), but it is the result of the process that we followed building TensorFlow in the previous article. Be aware that this is completely untested, you are on your own. I haven’t even tried it myself. Looky here:
We assume that JetPack 2.3.1 is used to flash the Jetson TX1. Install:
- L4T 24.2.1 an Ubuntu 16.04 64-bit variant (aarch64)
- CUDA 8.0
- cuDNN 5.1.5
Note that the library locations when installed by JetPack may not match a manual installation. TensorFlow will use CUDA and cuDNN in this build. The TensorFlow build depends on the location of the libraries installed by JetPack.
You will need to install the appropriate dependencies. If you look at the build dependencies:
You’ll see the usual suspects, I would guess that you need python-dev python-pip at a minimum, I’m not sure about Java, that may have just been needed for the build.
Then it’s the usual:
pip install tensorflow-0.11.0-py2-none-any.whl
and TensorFlow should be installed. You’ll have to figure it out on your own, but it should be somewhat simpler than having to build it yourself, if just for starters.
We’ll keep this little secret between ourselves.