Leveraging Local LLMs for Automated Changelog Generation (osc26)

Creating accurate changelog entries is often the final hurdle in the package submission process, where time pressure can lead to critical updates being overlooked. Since submission requests in the Open Build Service (OBS) contain the necessary technical context—including spec file diffs, file lists, and release notes—this task is an ideal candidate for automation using Large Language Models (LLMs). This talk explores the end-to-end workflow of fine-tuning a local LLM for this specialized task. We will cover how to extract data from OBS and the essential steps required to prepare it for training. Additionally, the talk discusses the fundamentals of fine-tuning, including model size selection and the optimization of tuning parameters. Licensed to the public under https://creativecommons.org/licenses/by-sa/4.0/ about this event: https://c3voc.de