frankhinek/keras repository overview

⁠Using the TensorFlow Container

This repository contains two Dockerfile versions for Keras with the TensorFlow backend.

The keras-tf-cpu⁠ version builds an image that contains Keras and the TensorFlow CPU version. It is based on the frankhinek/tensorflow:cpu container.

The keras-tf-gpu⁠ version builds an image that contains Keras and TensorFlow with support for NVIDIA GPUs enabled by the CUDA tool kit and accelerated by cuDNN. It is based on the frankhinek/tensorflow:gpu container.

Both images include Miniconda, Python 3.5, a collection of common data science packages, and the C++ based protobuf library that is 10x-50x faster than the Python-only implementation.

⁠Which containers exist?

Currently I maintain two Keras with TensorFlow Docker container images:

• frankhinek/keras:tf-cpu - Keras 2.1.2, TensorFlow 1.4.1, and Python 3.5
• frankhinek/keras:tf-gpu - Keras 2.1.2, TensorFlow 1.4.1, Python 3.5, CUDA 9.0, and cuDNN 7

These containers are published to Docker Hub⁠.

⁠Running the CPU-only container

Run the container using

$ docker run -it frankhinek/keras:tf-cpu

⁠Running the GPU (CUDA) container

Install nvidia-docker⁠ and run

$ nvidia-docker run -it frankhinek/keras:tf-gpu

⁠Rebuilding the containers

Just pick the Dockerfile associated with this container, and run:

$ docker build --pull -t $USER/keras:suffix -f Dockerfile .