Distroless Dockerfiles (#13)

Add distroless dockerfiles links 

images/README.md

@@ -5,6 +5,7 @@ Run the below commands to create manylinux2014 build environment.
 ```
 docker build -t build/manylinux2014 -f Dockerfile.manylinux2014.cpu .
 docker run -it build/manylinux2014:latest /bin/bash
+```
 
 Run the below commands to build TF.  
 ```
@@ -14,3 +15,23 @@ cd tensorflow/
 ./configure
 bazel build -c opt --cxxopt="-D_GLIBCXX_USE_CXX11_ABI=0" --verbose_failures //tensorflow/tools/pip_package:build_pip_package
 ```
+
+## Distroless Images
+The [Distroless](https://github.com/GoogleContainerTools/distroless) base images developed by Google are meant to contain an application and its runtime dependencies only. Please visit their GitHub for more information.
+
+Distroless Dockerfiles for TensorFlow CPU (`Dockerfile.cpu`) and GPU (`Dockerfile.distroless`) are hosted in the [rivanna-docker repository](https://github.com/uvarc/rivanna-docker). Using the [MNIST benchmark](https://www.tensorflow.org/tutorials/quickstart/beginner), the containers were tested on UVA's HPC Rivanna cluster to have the same performance as the official ones.
+
+Image size comparison:
+| [Official](https://hub.docker.com/r/tensorflow/tensorflow/tags) | Size | [Distroless](https://hub.docker.com/r/uvarc/tensorflow/tags) | Size | Reduction |
+|---|---|---|---|---|
+| `2.3.0` | 582 MB | `2.3.0-cpu` | 227 MB | 61% |
+| `2.3.0-gpu` | 1.44 GB | `2.3.0-distroless` | 1.18 GB | 18% |
+| `nightly-gpu` | 2.27 GB | `2.4.0-distroless` | 1.47 GB | 35% |
+
+Pull command:
+```
+docker pull uvarc/tensorflow:<tag>
+```
+
+### Contact
+Research Computing at the University of Virginia (external) maintains Distroless Dockerfiles for TensorFlow in the [rivanna-docker repository](https://github.com/uvarc/rivanna-docker). Please reach out to hpc-support@virginia.edu for questions and feedback.