Foundry Docker Image Now Available

Rosetta Commons now offers an official Foundry Docker image designed to make it dramatically easier to run cutting-edge biomolecular machine learning models without complex setup. Foundry is the central repository for a suite of ML models used in protein science—including RF3 (a next-generation biomolecular structure prediction network), ProteinMPNN inverse-folding models, and the newest RFdiffusion3 (RFD3) generative design model—built on a shared framework with AtomWorks for structure handling and inference.

The Foundry Docker image comes with a default tag that includes pre-installed model weights for these core models, letting users start running inference and design workflows right away inside a containerized environment with consistent dependencies. For situations where image size matters—such as edge deployments or limited storage—there’s also a “slim” tag that omits the model weights so you can pull just the framework and add weights later as needed. 

This Docker support builds on Rosetta Commons’ ongoing effort to lower barriers to advanced protein design and structure prediction by packaging complex, state-of-the-art tools into reproducible, sharable environments. 

Quick Start (Docker)

• Install Docker on your system (if you don’t already).
• Pull the Foundry image with pre-installed model weights:

docker pull rosettacommons/foundry:latest

• Run a container from the image:

docker run --rm -it rosettacommons/foundry:latest

• • This starts a container with the Foundry environment set up and ready to use.

• (Optional) Use the “slim” tag if you want a smaller image without weights:

docker pull rosettacommons/foundry:slim

• Once inside the container, you can run Foundry commands or inference workflows using tools included in the image.

Need help or run into trouble? The Rosetta Commons team and community (including @Hope Woods) are available to help.