Awesome NLP on AWS GPU

Tech:

Ubuntu
Nvidia Cuda
Python
Theano
TensorFlow
Keras
Scikit Learn
VowPal Wabbit
LDA2Vec
spaCy
and more

Create GPU instance

ubuntu 14.04 ami-7c927e11 from Canonical set up on GPU instance (HVM-SSD)

sudo apt-get update
sudo apt-get install aptitude wget python-numpy python-scipy python-dev python-pip python-nose g++  libatlas-base-dev gfortran libopenblas-dev git build-essential linux-image-extra-virtual  libboost-dev libboost libboost-program-options-dev libboost-python-dev   libboost-mpi-python-dev libboost-python libhd5-dev libhdf5-7 pkg-config zip g++ zlib1g-dev unzip swig

Nouveau workaround

echo -e "blacklist nouveau\nblacklist lbm-nouveau\noptions nouveau modeset=0\nalias nouveau off\nalias lbm-nouveau off\n" | sudo tee /etc/modprobe.d/blacklist-nouveau.conf
echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf
sudo update-initramfs -u
sudo reboot

CUDA

sudo apt-get install linux-headers-3.13.0-86-generic
sudo apt-get install -y build-essential python-pip python-dev git python-numpy swig python-dev  zip zlib1g-dev

sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb
sudo apt-get update
sudo apt-get install cuda
sudo reboot

cd /usr/local/cuda/samples/1_Utilities/deviceQuery
make
./deviceQuery
cd

CUDADNN

sudo dpkg -i libcudnn5_5.0.4-1+cuda7.5_amd64.deb
sudo dpkg -i libcudnn5-dev_5.0.4-1+cuda7.5_amd64.deb
sudo cp /usr/include/cudnn* /usr/local/cuda/include/
echo -e "\nexport PATH=/usr/local/cuda/bin:$PATH\n\nexport LD_LIBRARY_PATH=/usr/local/cuda/lib64" >> .bashrc

Theano

Reinstalls scipy since we have Fortran and LibAtlas

# sudo pip install --upgrade pip
sudo pip install --upgrade  git+git://github.com/Theano/Theano.git
sudo pip install --upgrade --no-cache  git+git://github.com/fchollet/keras.git

Specify to always use GPU

echo -e "\n[global]\nfloatX=float32\ndevice=gpu\n[mode]=FAST_RUN\n\n[nvcc]\nfastmath=True\n\n[cuda]\nroot=/usr/local/cuda" >> ~/.theanorc

Some warnings about Python version.

testing

cd
source .bashrc
git clone git://github.com/fchollet/keras.git
cd keras/examples/
python lstm_text_generation.py
cd

Should see

Using Theano backend.
Using gpu device 0: GRID K520 (CNMeM is disabled, cuDNN 5004)
corpus length: 600901
total chars: 59
nb sequences: 200287
Vectorization...
Build model...

And in top, something like

18765 ubuntu    20   0   63484  53724   3568 R  17.6  0.3   0:00.53 cudafe
19055 ubuntu    20   0   95964  59760   4728 R   9.3  0.4   0:00.28 cc1plus

Tensorflow

Via https://www.tensorflow.org/versions/r0.8/get_started/os_setup.html#installation-for-linux

sudo pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.8.0-cp27-none-linux_x86_64.whl

But, compiling from source may be necessary for levaraging GPU.

Prerequisites

Requires Java install, you may prefer Java7 for safety

sudo apt-add-repository ppa:webupd8team/java

Note: Personal Package Archives include unsupported packages and are untrusted by the primary Ubuntu branch. At the time of publication, the WebUpd8 Oracle Java PPA is just an installer (meaning it does not include any Oracle Java binaries, but will download and install them). Use this PPA at your own risk.

sudo apt-get update
sudo apt-get install oracle-java8-installer

Bazel

Material here mostly from the https://gist.github.com/erikbern/78ba519b97b440e10640 and https://github.com/tensorflow/models/tree/master/syntaxnet

sudo apt-get install
git clone https://github.com/bazelbuild/bazel.git
cd bazel
git checkout tags/0.2.2
./compile.sh
sudo cp output/bazel /usr/bin
sudo pip install -U protobuf==3.0.0b2
sudo pip instal asciitree

Compile Install

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64"
export CUDA_HOME=/usr/local/cuda

for syntax net

git clone --recursive https://github.com/tensorflow/models.git
cd models/syntaxnet/tensorflow
./configure
cd ..
bazel test syntaxnet/... util/utf8/...

full tensor flow

compiling was not working, but I was using the wrong version of Bezel

git clone --recurse-submodules https://github.com/tensorflow/tensorflow
cd tensorflow
TF_UNOFFICIAL_SETTING=1 ./configure

You will see

Please specify the location of python. [Default is /usr/bin/python]:
Do you wish to build TensorFlow with GPU support? [y/N] y
GPU support will be enabled for TensorFlow
Please specify which gcc nvcc should use as the host compiler. [Default is /usr/bin/gcc]:
Please specify the Cuda SDK version you want to use, e.g. 7.0. [Leave empty to use system default]: 7.5
...


bazel build -c opt --config=cuda //tensorflow/cc:tutorials_example_trainer
bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
sudo pip install /tmp/tensorflow_pkg/tensorflow-*-none-linux_x86_64.whl

Test

cd tensorflow/models/image/cifar10/
python cifar10_multi_gpu_train.py

Others

sudo aptitude install vowpal-wabbit
# pandas and h5 can take a while
sudo pip install --upgrade --no-cache ipython scikit-learn sklearn-pandas nose nltk h5py
sudo pip install --upgrade --no-cache git+https://github.com/piskvorky/gensim.git
sudo pip install --upgrade --no-cache spacy
sudo pip install --upgrade --no-cache pyvw
sudo pip install --upgrade --no-cache git+https://github.com/cemoody/lda2vec.git#egg=lda2vec

skip thoughts

https://github.com/ryankiros/skip-thoughts

git clone https://github.com/ryankiros/skip-thoughts

Some sources

http://markus.com/install-theano-on-aws/ http://tleyden.github.io/blog/2014/10/25/cuda-6-dot-5-on-aws-gpu-instance-running-ubuntu-14-dot-04/ http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using_cluster_computing.html

Hendler/cuda_aws_ubuntu_theano_tensorflow_nlp.md