Prompt Public Large Language Models to Synthesize Data for Private On-device Applications

TL;DR

This paper demonstrates how large language models can generate synthetic data resembling private user data, improving on-device language models trained with federated learning and differential privacy, leading to significant performance gains.

Contribution

It introduces a prompt-based data synthesis method using LLMs to enhance pre-training data quality for private, on-device language models trained with federated learning.

Findings

Synthetic data improves next word prediction accuracy by up to 22.8%.

Models trained on synthetic data outperform baseline models in real-world tests.

The approach achieves comparable or better results than traditional methods during privacy-preserving fine-tuning.

Abstract

Pre-training on public data is an effective method to improve the performance for federated learning (FL) with differential privacy (DP). This paper investigates how large language models (LLMs) trained on public data can improve the quality of pre-training data for the on-device language models trained with DP and FL. We carefully design LLM prompts to filter and transform existing public data, and generate new data to resemble the real user data distribution. The model pre-trained on our synthetic dataset achieves relative improvement of 19.0% and 22.8% in next word prediction accuracy compared to the baseline model pre-trained on a standard public dataset, when evaluated over the real user data in Gboard (Google Keyboard, a production mobile keyboard application). Furthermore, our method achieves evaluation accuracy better than or comparable to the baseline during the DP FL fine-tuning over millions of mobile devices, and our final model outperforms the baseline in production A/B testing. Our experiments demonstrate the strengths of LLMs in synthesizing data close to the private distribution even without accessing the private data, and also suggest future research directions to further reduce the distribution gap.