INSTALL-ML-SW-TOOLS
This tutorial is about how to install ML SW Tools as Caffe and Keras for Deep Learning on a Ubuntu 16.04 Linux PC equipped with one NVIDIA GPU at least, using CUDA 8.0 -or also CUDA 9.0- cuDNN 7.0.5 and NCCL1 NVIDIA libraries. Note also that Caffe was originally delivered with Python 2.7 and CUDA 8.0, this is why I prefer tu use CUDA 8.0, being compatible also with Keras/TensorFlow in the same Python2.7 virtual environment.
This tutorial also tells you about how to install the DeePhi DNNDK 2.0.8 beta release to quantize a CNN on XILINX Zynq SoC and MPSoC FPGAs.
Finally, this tutorial tells you how to install those tools on an Ubuntu 16.04 AMI p2.xlarge EC2 on AWS.
1.0 Getting Started
You can find very detailed instructions about how installing Ubuntu 16.04, Caffe, Python 2.7 and its virtual environments, all the GPU libraries in this PDF document Installing_ML_SW_DellPrecision5820TowerPC, which collects all the experience I have done on different machines (Laptop, Desktop and AWS).
Note that Section 11 of such document explains how to install the DeePhi tools on the Host and Target with many more details than what available in the official DNNDK User Guide 1327.
Note also that Section 12 tells you how setting correctly a p2.xlarge EC2 on a Ubuntu 16.04 AMI on the AWS.
I recommend you to skip Sections 8 and 9 and try to run directly Section 10, which contains a sophisticated script to install automatically Caffe, TensorFlow, Keras, provided you have already correcly installed the NVIDIA libraries according to Sections 6 and 7.
Assuming you have copied the install_caffe_scripts.tar to your $HOME directory, all what you need to do is to run the following commands from your Ubuntu Linux PC:
cd ~
tar –xvf install_caffe_scripts.tar
mkdir $HOME/caffe_tools
source $HOME/scripts/caffe/install_caffe.sh
2.0 References
-
Here is the official online Caffe tutorial.
-
Here are some very good posts - among many others - available from PyImageSearch: