Speakers
Description
NOMAD Actions provide a framework for executing complex or resource-intensive workflows within NOMAD. They are ideal for tasks that require special resources like GPUs, have long execution times, or need to interact with external systems, offering a more powerful alternative to standard normalizers.
Built as plugin entry points, Actions use Temporal.io to orchestrate workflows made up of reliable, self-contained steps. Temporal handles the hard distributed computing problems, such as fault tolerance, retries, and scaling, so that the plugin authors can focus on the core scientific logic. The resulting workflows can be triggered and inspected from the NOMAD GUI.
We illustrate Actions with examples from automated data processing and advanced ML workflows. These example Actions are already deployed in lab environments, supporting ML-assisted characterization, generative AI model sampling, and large-scale data transformation.