-
Notifications
You must be signed in to change notification settings - Fork 26
Installation
Here we describe how to customize builds of TensorFlow via Collective Knowledge workflow framework
- Python 2.x:
$ sudo apt-get install python-dev python-pip python-setuptools python-opencv git $ sudo pip --upgrade install pip $ sudo pip install protobuf easydict joblib image $ sudo pip install ck
- Python 3.x:
$ sudo apt-get install python3-dev python3-pip python3-setuptools $ sudo pip3 --upgrade install pip $ sudo pip3 install wheel protobuf easydict joblib image $ sudo pip3 install ck
- Java:
$ sudo apt install openjdk-8-jdk-headless- NB:** Installation fails with `openjdk-9-jdk-headless` (cf. this.
- CUDA, cuDNN (GPU version only)
- Check if you have a CUDA-enabled GPU.
- Download and install cuDNN (requires registration).
You can install CK-TensorFlow repository as follows:
$ ck pull repo:ck-tensorflow
You should now be ready to install the CPU version of CK-TensorFlow (CK will automatically download and install Bazel and JDK if they are not already installed):
$ ck install package:lib-tensorflow-cpu
You can also install the CUDA version of TensorFlow (which when installed via CK can co-exist with the CPU version):
$ ck install package:lib-tensorflow-cuda
Finally, you can try to install the OpenCL version of TensorFlow (which requires ComputeCPP; unfortunately, at the time of this writing we were not able to run it):
$ ck install package:lib-tensorflow-opencl
- for ARM64-based platforms (should be connected to your host via adb):
$ ck install package:lib-tensorflow-cpu-make --target_os=android21-arm64 $ ck compile program:tensorflow-classification-cpu --target_os=android21-arm64 $ ck run program:tensorflow-classification-cpu --target_os=android21-arm64
- for ARM32-based platforms:
$ ck install package:lib-tensorflow-cpu-make --target_os=android21-arm-v7a $ ck compile program:tensorflow-classification-cpu --target_os=android21-arm-v7a --env.OPTFLAGS="-O2 -march=armv7-a -mfloat-abi=softfp -mfpu=neon" $ ck run program:tensorflow-classification-cpu --target_os=android21-arm-v7a
Note, that you can build and run TensorFlow on older Android 4.2+ devices - just substitute "android21-arm-v7a" with "android19-arm-v7a" above.
Above method is used to prepare CK-based crowd-scenarios for our Android app to crowdsource deep learning optimization!
You can install tensorflow for devices with constrained resources using Makefile via CK as follows (particularly if GCC fails with a memory corruption error):
$ ck install package:lib-tensorflow-cpu-make --env.OPTFLAGS="-O2"
or
$ ck install package:lib-tensorflow-cpu-make --env.OPTFLAGS="-O1"
You can then compile and run C++ classification example as follows:
$ ck compile program:tensorflow-classification-cpu $ ck run program:tensorflow-classification-cpu
TensorFlow installation may occasionally fail due to failing to download some dependencies from GitHub. Restart package installation several times until Bazel downloads all necessary files.
Sometimes, after installation, TensorFlow crashes with undefined "syntax". It is usually related to outdated default protobuf (you need version >=3.0.0a4). To fix this problem upgrade protobuf via
$ sudo pip install protobuf --upgrade or $ sudo pip3 install protobuf --upgrade
It may also fail with the following message "can't combine user with prefix, exec_prefix/home, or install_(plat)base". The following fix may help:
$ sudo pip install --upgrade pip"
If you are installing TF on Raspberry Pi and encountered build errors in which the compiler is being killed, it is likely that the compiler is running out of memory (especially if you are on Raspberry Pi 1, 2 or Zero, which have less than 1GB of RAM), this can often be rectified by increasing the swapfile size on the Pi by editing the file /etc/dphys-swapfile and changing the line CONF_SWAPSIZE=100 to CONF_SWAPSIZE=1024, then running:
sudo vim /etc/dphys-swapfile
sudo /etc/init.d/dphys-swapfile stop
sudo /etc/init.d/dphys-swapfile start