setup-aws-tensorflow.bash
installs the following things on the ec2 p2.xlarge
instance running Ubuntu 14.04:
- Required linux packages
- CUDA 8.0
- cuDNN v5.1
- Anaconda with Python 3.5
- TensorFlow 0.12rc
- Keras
Also configures the Jupyter Notebook for interactive use
Forked from: https://github.com/jwittenbach/aws-tensorflow-setup
- Install the Amazon Web Services CLI or use the AWS Dashboard online
- Setup your AWS credentials
- Create a security group with the following ports open to TCP/IP traffic: 22 (SSH), 80 (HTTP), 9999 (Jupyter Notebook)
- Launch a
p2.xlarge
instance running Ubuntu server - Log in to your instance via SSH
- Update
apt-get
and packages
sudo apt-get update && sudo apt-get -y upgrade
sudo apt-get install -y build-essential git python-pip
sudo pip install awscli
aws configure
aws s3 sync s3:// `your s3 path` $HOME/s3
- Clone this repo
git clone https://github.com/duducheng/aws-tensorflow-setup
- Run the installation script
./aws-tensorflow-setup/setup-aws-tensorflow.bash
or setup conda only
chmod +x ./aws-tensorflow-setup/setup_conda.sh
./aws-tensorflow-setup/setup_conda.sh
Note: When installing Anaconda3, it may cause some problems. If it doesn't work well, run the scipts from "# install anaconda" again.
- Launch a Jupyter Notebook server
jupyter Notebook
and access it by pointing your browser to [publicIP]:9999
You have two options for shutting down the instance:
- If you
stop
the instance, then you can laterstart
the instance back up again, and you will not need to go through the setup again. However, the disk image will be stored on AWS S3, which does cost a small fee. - If you
terminate
the instance, then you will need tolaunch
a branch new instance the next time, which will entail redoing the setup process.