Fixed Issue #552

GETTING_STARTED.md

@@ -156,29 +156,29 @@ To use the Docker container as an interactive virtual environment, you can run a
 
 ### Using Singularity/Apptainer instead of Docker
 
-Since many compute clusters don't allow the usage of Docker due to securtiy concerns and instead encourage the use of [Singularity/Apptainer](https://github.com/apptainer/apptainer) (formerly Singularity, now called Apptainer), we also provide instructions on how to build an Apptainer container based on the here provided Dockerfile.
-
-To convert the Dockerfile into an Apptainer definition file, we will use [spython](https://github.com/singularityhub/singularity-cli):
+Since many compute clusters don't allow the usage of Docker due to securtiy concerns and instead encourage the use of [Singularity/Apptainer](https://github.com/apptainer/apptainer) (formerly Singularity, now called Apptainer), we also provide an Apptainer recipe (located at `docker/Singularity.def`) that can be used to build an image by running
 
 ```bash
-pip3 install spython
-cd algorithmic-efficiency/docker
-spython recipe Dockerfile &> Singularity.def
+singularity build --fakeroot <singularity_image_name>.sif Singularity.def
 ```
 
-Now we can build the Apptainer image by running
-
+Note that this can take several minutes. Then, to start a shell session with GPU support (by using the `--nv` flag), we can run
 ```bash
-singularity build --fakeroot <singularity_image_name>.sif Singularity.def
+singularity shell --bind $HOME/data:/data,$HOME/experiment_runs:/experiment_runs \
+    --nv <singularity_image_name>.sif
 ```
 
-To start a shell session with GPU support (by using the `--nv` flag), we can run
+Note the `--bind` flag which, similarly to Docker, allows to bind specific paths on the host system and the container, as explained [here](https://docs.sylabs.io/guides/3.7/user-guide/bind_paths_and_mounts.html).
+
+Also note that we generated `Singularity.def` automatically from the `Dockerfile` using [spython](https://github.com/singularityhub/singularity-cli), as follows:
 
 ```bash
-singularity shell --nv <singularity_image_name>.sif 
+pip3 install spython
+cd algorithmic-efficiency/docker
+python scripts/singularity_converter.py -i Dockerfile -o Singularity.def
 ```
 
-Similarly to Docker, Apptainer allows you to bind specific paths on the host system and the container by specifying the `--bind` flag, as explained [here](https://docs.sylabs.io/guides/3.7/user-guide/bind_paths_and_mounts.html).
+Users that wish to customize their images are invited to check and modify the `Singularity.def` recipe and the `singularity_converter.py` script.
 
 ## Download the Data
 

README.md

@@ -48,14 +48,6 @@ pip3 install -e '.[jax_gpu]' -f 'https://storage.googleapis.com/jax-releases/jax
 pip3 install -e '.[full]'
 ```
 
-*TL;DR to install the PyTorch version for GPU run:*
-
-```bash
-pip3 install -e '.[jax_cpu]'
-pip3 install -e '.[pytorch_gpu]' -f 'https://download.pytorch.org/whl/torch_stable.html'
-pip3 install -e '.[full]'
-```
-
 ## Getting Started
 
 For detailed instructions on developing and scoring your own algorithm in the benchmark see the [Getting Started](/GETTING_STARTED.md) document.

docker/Singularity.def

@@ -0,0 +1,76 @@
+Bootstrap: docker
+From: nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04
+Stage: spython-base
+
+%post
+# Dockerfile for AlgoPerf environment.
+# To build Docker image with only Jax GPU installed:
+# docker build -t <image_name> --build-arg framework=jax
+# To build Docker image with Pytorch GPU installed:
+# docker build -t <image_name> --build-arg framework=pytorch
+
+# To build Docker image
+
+# Installing machine packages
+echo "Setting up machine"
+apt-get update
+apt-get install -y curl tar
+DEBIAN_FRONTEND=noninteractive apt-get install -y git python3 pip wget ffmpeg
+apt-get install libtcmalloc-minimal4
+apt-get install unzip
+apt-get install pigz
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4
+
+# Install GCP tools
+echo "Setting up gsutil"
+curl -O https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-cli-413.0.0-linux-x86_64.tar.gz
+tar -xf google-cloud-cli-413.0.0-linux-x86_64.tar.gz
+yes | ./google-cloud-sdk/install.sh
+
+# Directory setup for input and output
+echo "Setting up directories for data and experiment_runs"
+mkdir -p data/
+mkdir -p experiment_runs/
+
+# Install Algorithmic efficiency repo
+echo "Setting up algorithmic_efficiency repo"
+branch="main"
+framework="both"
+git_url=https://github.com/mlcommons/algorithmic-efficiency.git
+git clone $git_url && cd /algorithmic-efficiency
+cd /algorithmic-efficiency && git checkout $branch
+
+cd /algorithmic-efficiency && pip install -e '.[full]'
+
+if [ "$framework" = "jax" ] ; then \
+echo "Installing Jax GPU" \
+&& cd /algorithmic-efficiency \
+&& pip install -e '.[jax_gpu]' -f 'https://storage.googleapis.com/jax-releases/jax_cuda_releases.html' \
+&& pip install -e '.[pytorch_cpu]' -f 'https://download.pytorch.org/whl/torch_stable.html'; \
+elif [ "$framework" = "pytorch" ] ; then \
+echo "Installing Pytorch GPU" \
+&& cd /algorithmic-efficiency \
+&& pip install -e '.[jax_cpu]' \
+&& pip install -e '.[pytorch_gpu]' -f 'https://download.pytorch.org/whl/torch_stable.html'; \
+elif [ "$framework" = "both" ] ; then \
+echo "Installing Jax GPU and Pytorch GPU" \
+&& cd /algorithmic-efficiency \
+&& pip install -e '.[jax_gpu]' -f 'https://storage.googleapis.com/jax-releases/jax_cuda_releases.html' \
+&& pip install -e '.[pytorch_gpu]' -f 'https://download.pytorch.org/whl/torch_stable.html'; \
+else \
+echo "Invalid build-arg $framework: framework should be either jax, pytorch or both." >&2 \
+&& exit 1 ; \
+fi
+
+cd /algorithmic-efficiency && pip install -e '.[wandb]'
+
+cd /algorithmic-efficiency && git fetch origin
+cd /algorithmic-efficiency && git pull
+
+# Todo: remove this, this is temporary for developing
+chmod a+x /algorithmic-efficiency/docker/scripts/startup.sh
+
+%runscript
+exec bash /algorithmic-efficiency/docker/scripts/startup.sh "$@"
+%startscript
+exec bash /algorithmic-efficiency/docker/scripts/startup.sh "$@"

docker/scripts/singularity_converter.py

@@ -0,0 +1,48 @@
+"""
+This script is a modification of the
+``spython recipe Dockerfile &> Singularity.def`` command, implemented here:
+github.com/singularityhub/singularity-cli/blob/master/spython/client/recipe.py
+
+It converts the Docker recipe to Singularity, but suppressing any %files
+command. Usage example:
+
+python singularity_converter.py -i Dockerfile -o Singularity.def
+"""
+
+import argparse
+
+from spython.main.parse.parsers import get_parser
+from spython.main.parse.writers import get_writer
+
+# globals
+ENTRY_POINT = "/bin/bash"  # seems to be a good default
+FORCE = False  # seems to be a good default
+#
+parser = argparse.ArgumentParser(description="Custom Singularity converter")
+parser.add_argument(
+    "-i", "--input", type=str, help="Docker input path", default="Dockerfile")
+parser.add_argument(
+    "-o",
+    "--output",
+    type=str,
+    help="Singularity output path",
+    default="Singularity.def",
+)
+args = parser.parse_args()
+INPUT_DOCKERFILE_PATH = args.input
+OUTPUT_SINGULARITY_PATH = args.output
+
+# create Docker parser and Singularity writer
+parser = get_parser("docker")
+writer = get_writer("singularity")
+
+# parse Dockerfile into Singularity and suppress %files commands
+recipeParser = parser(INPUT_DOCKERFILE_PATH)
+recipeWriter = writer(recipeParser.recipe)
+(key,) = recipeParser.recipe.keys()
+recipeWriter.recipe[key].files = []
+
+# convert to string and save to output file
+result = recipeWriter.convert(runscript=ENTRY_POINT, force=FORCE)
+with open(OUTPUT_SINGULARITY_PATH, "w") as f:
+  f.write(result)