Optimizing Cross-Client Domain Coverage for Federated Instruction Tuning of Large Language Models
This work provides an incremental but effective solution for enhancing domain-specific LLM performance in federated learning settings where private data is distributed and limited.
The paper tackles the challenge of identifying key performance drivers in federated domain-specific instruction tuning for large language models, finding that cross-client domain coverage is more critical than data heterogeneity. It introduces FedDCA, an algorithm that maximizes this coverage through client selection and augmentation, achieving performance gains of up to 29.19% and coverage improvements of 4.82%-21.36% over baselines.
Federated domain-specific instruction tuning (FedDIT) for large language models (LLMs) aims to enhance performance in specialized domains using distributed private and limited data, yet identifying key performance drivers and optimal augmentation strategies remains challenging. We empirically establish that cross-client domain coverage, rather than data heterogeneity, is the pivotal factor. We then introduce FedDCA, an algorithm that explicitly maximizes this coverage through diversity-oriented client center selection and retrieval-based augmentation, constructing diverse, non-redundant cross-client instruction sets. Extensive experiments across multiple domains demonstrate FedDCA's superiority over eleven baselines, achieving performance gains of up to 29.19\% and domain coverage improvements of 4.82\%-21.36\%. FedDCA maintains its effectiveness in diverse and challenging scenarios, including data selection, held-out settings where task-specific public data is scarce and various data heterogeneity, with manageable privacy risks. This work clarifies critical FedDIT dynamics and presents FedDCA as an effective, privacy-preserving, and scalable solution for advancing domain-specific LLM tuning.