Federated Data-Efficient Instruction Tuning for Large Language Models

Zhen Qin, Zhaomin Wu, Bingsheng He, Shuiguang Deng
10/14/2024
cs.LGcs.AIcs.CL

No official code found. Best implementation: zhenqincn/FedHDS (3 stars, Apache-2.0, last commit 2025-07-05).

Abstract

Instruction tuning is a crucial step in improving the responsiveness of pretrained large language models (LLMs) to human instructions. Federated learning (FL) helps to exploit the use of vast private instruction data from clients, becoming popular for LLM tuning by improving data diversity. Existing federated tuning simply consumes all local data, causing excessive computational overhead and overfitting to local data, while centralized data-efficient solutions are not suitable for FL due to privacy concerns. This work presents FedHDS, a federated data-efficient instruction tuning approach, which tunes LLMs with a representative subset of edge-side data. It reduces the data redundancy at both intra- and inter-client levels without sharing raw data. Experiments with various LLMs, datasets and partitions show that FedHDS improves Rouge-L on unseen tasks by an average of 10.72% over the SOTA full-data federated instruction tuning methods, while using less than 1.5% of the data samples, improving training efficiency by up to tens of times.

View on arXivView PDF

Code Implementations(1)

zhenqincn/FedHDSCommunity62%
Apache-2.0

Implementation of "Federated Data-Efficient Instruction Tuning for Large Language Models", which is accepted to ACL 2025 (Findings).

31PythonMay 25, 20251 years agoApache-2.0

Similar Papers

Data Quality Control in Federated Instruction-tuning of Large Language Models

90% match2024

Optimizing Cross-Client Domain Coverage for Federated Instruction Tuning of Large Language Models

89% match2024

Leveraging Unstructured Text Data for Federated Instruction Tuning of Large Language Models

89% match2024

Automated Federated Pipeline for Parameter-Efficient Fine-Tuning of Large Language Models

89% match2024

Federated Fine-Tuning of LLMs: Framework Comparison and Research Directions

88% match2025

FedBiOT: LLM Local Fine-tuning in Federated Learning without Full Model

88% match1 repo622024

Cite this paper

@article{qin2024federated,
  title  = {Federated Data-Efficient Instruction Tuning for Large Language Models},
  author = {Zhen Qin and Zhaomin Wu and Bingsheng He and Shuiguang Deng},
  year   = {2024},
  eprint = {2410.10926},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2410.10926v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2410.10926.svg)](https://codeofpaper.com/paper/2410.10926)

Discussion