All the executions in MLRun are based on Serverless functions. The functions allow specifying code and all the operational aspects (image, required packages, cpu/mem/gpu resources, storage, environment, etc.). The different function runtimes take care of automatically transforming the code and spec to fully managed and elastic services over Kubernetes, which saves significant operational overhead, addresses scalability and reduces infrastructure costs.
Function objects are all inclusive (code, spec, API, and metadata definitions), which allows placing them in a shared and versioned function market place. This means that different members of the team can produce or consume functions. Each function is versioned and stored in the MLRun database with a unique hash code, and gets a new hash code upon changes.
In this section