Deployment Guide

This section provides additional details for the deployment of the AlphaFold2-Multimer NIM.

View Available NIM Container Information

Container image tags can be seen with the command below, similar to other container images on NGC.

ngc registry image info nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0

Pull the Container Image From NGC

Pull the container image using one of the following commands:

Docker

docker pull nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0

NGC SDK

ngc registry image pull nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0

Run the Container

As described in the Quickstart Guide, you can run the following command to start the AlphaFold2-Multimer NIM:

export LOCAL_NIM_CACHE=~/.cache/nim
export NGC_CLI_API_KEY=<Your NGC CLI API Key>

docker run -it --rm --name alphafold2-multimer --runtime=nvidia \
    -p 8000:8000 \
    -e NGC_CLI_API_KEY \
    -v $LOCAL_NIM_CACHE:/opt/nim/.cache \
    nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0

The following list shows the commonly used options for starting the NIM container:

• docker run: This is the command to run a new container from a Docker image.

• -it: The -i flag allows access to STDIN for the container, and -t allocates a pseudo-TTY that interprets commands, for example Ctrl+C to stop the NIM container.

• --rm: This flag tells Docker to automatically remove the container when it exits. This property is useful for one-off runs or testing, as it prevents the container from being left behind.

• --name alphafold2-multimer: This flag gives the container the name “alphafold2-multimer”.

• --runtime=nvidia: This flag specifies the runtime to use for the container. In this case, it is set to “nvidia”, which is used for GPU acceleration.

• -e NGC_CLI_API_KEY: This passes the NGC_CLI_API_KEY environment variable (and the value set in the parent terminal) to the container. This is used for authentication with NVIDIA’s NGC (NVIDIA GPU Cloud) service, including downloading the model weights and databases if it is not present in the NIM Cache.

• -p 8000:8000: This flag maps port 8000 on the host machine to port 8000 in the container. This allows you to access the container’s services from the host machine.

• -v <source>:<dest>: Mounts the host directory ~/.cache/nim in the container as /opt/nim/.cache so that it can be used for storing and reusing downloaded models.

• -e NIM_IGNORE_MODEL_DOWNLOAD_FAIL=true: Ignores model download and detection errors, such as when mounting the model from a local cache to a NIM container hosted on an offline device that does not have access to NGC. Requires a locally cached and mounted model for expected behavior in the NIM. For more information or alternatives, refer to Configuring the NIM at Runtime.

• -e NIM_CACHE_PATH=<dest>: Changes the location where you mount your model in the filesystem of the NIM container. The default value is /opt/nim/.cache. Associated with -v <source>:<dest>, and for more information refer to Configuring the NIM at Runtime.

• -e NIM_HTTP_API_PORT=<port>: Changes the HTTP API port for sending queries to the NIM. The default value is 8000. Associated with -p <host_port>:<server_port>, and for more information refer to Configuring the NIM at Runtime.

• -e NIM_PARALLEL_MSA_RUNNERS=<int> and -e NIM_PARALLEL_THREADS_PER_MSA=<int>: Change the number of parallel runners and threads per runner for MSA. For more information, refer to Configuring the NIM at Runtime.

Set Optional Runtime Parameters

This sample uses 3 MSA runners and 12 threads per runner. The 64-core example on the runtime page uses 6 runners and 10 threads.

docker run -it --rm --name alphafold2-multimer --runtime=nvidia \
    -e CUDA_VISIBLE_DEVICES=0 \
    -e NGC_CLI_API_KEY \
    -e NIM_CACHE_PATH=/opt/nim/.cache \
    -e NIM_PARALLEL_MSA_RUNNERS=3 \
    -e NIM_PARALLEL_THREADS_PER_MSA=12 \
    -p 8000:8000 \
    -v $LOCAL_NIM_CACHE:/opt/nim/.cache \
    nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0

• -e CUDA_VISIBLE_DEVICES=0: This flag sets an environment variable CUDA_VISIBLE_DEVICES to the value 0. This variable controls which GPU devices are visible to the container. In this case, it is set to 0, which means the container will only use the first GPU (if available).

• nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0: This is the Docker image name and tag. The image is hosted on NVIDIA’s container registry (nvcr.io) and is named alphafold2-multimer. The tag 1.0.0 specifies a specific version of the image.

Check NIM Status

You can view running Docker containers on your system by using the following command:

docker ps

This will return an output that looks like the following if the NIM is running:

CONTAINER ID   IMAGE            COMMAND                  CREATED          STATUS          PORTS                                                                                                          NAMES
d114948j4f55   alphafold2-multimer   "/opt/nvidia/nvidia_…"   46 minutes ago   Up 46 minutes   6006/tcp, 0.0.0.0:8000->8000/tcp, :::8000->8000/tcp, 8888/tcp, 0.0.0.0:50051->50051/tcp, :::50051->50051/tcp   test1

The first column in the output is the Docker container ID, which can be useful for interacting with the container. The remaining fields describe the image the container is running, the command (in this case, the NIM server software), when the container was built, and its status. They also show how long it has been running, any open ports, and finally its name (given by the startup command).

If the NIM is not running, only the header will be returned.

Stop a Running Container

If for some reason your container has entered a state in which you cannot interact with it, or for any reason it must be killed, you may use the following command and the ID obtained from the docker ps command. Here is an example using the container ID from the previous example:

docker kill d114948j4f55

This will immediately terminate the running container.

Cache the Model Checkpoint

On initial startup, the container will download the AlphaFold2 model parameters and supporting databases from NGC. You can skip this download step on future runs by caching the model weights locally using a cache directory as in the example below.

# Create the cache directory on the host machine
export LOCAL_NIM_CACHE=~/.cache/nim
mkdir -p $LOCAL_NIM_CACHE

Run the container with the cache directory mounted as -v $LOCAL_NIM_CACHE:/opt/nim/.cache:

docker run -it --rm --name alphafold2-multimer --runtime=nvidia \
    -e CUDA_VISIBLE_DEVICES=0 \
    -e NGC_CLI_API_KEY \
    -v $LOCAL_NIM_CACHE:/opt/nim/.cache \
    -p 8000:8000 \
    nvcr.io/nim/deepmind/alphafold2-multimer:1.0.0