The article introduces Ai2's Open Coding Agents, which allow developers to train coding models on their private codebases with a new method that simplifies data generation and reduces costs. The recent release of SERA-14B enhances this capability, making it easier to adapt coding agents for specific needs. The approach focuses on generating synthetic training data that reflects developer workflows rather than relying solely on correct coding examples.

This article details a project where the author trains a smaller LLM to understand and generate diagrams in the Pintora language. The process includes dataset creation, two training phases, and evaluation of the model's accuracy in producing valid diagram syntax.