Abstract: Federated learning has emerged as a foundational technique in the training of contemporary large language models (LLMs) and other privacy-sensitive machine learning applications. As a ...