Differentially Private Federated Low Rank Adaptation beyond Fixed-Matrix

Large language models LLMs typically require fine-tuning for domain-specific tasks, and LoRA offers a computationally efficient approach by training low-rank adapters. LoRA is also communication-efficient for federated LLMs when multiple users collaboratively fine-tune a global LLM model without...