Proofread: Fixes All Errors with One Tap

TL;DR

Proofread is a new Gboard feature powered by a server-side LLM that enables users to fix all errors in their text with a single tap, using a carefully tuned model and efficient deployment techniques.

Contribution

This paper introduces Proofread, a novel sentence and paragraph correction feature in Gboard, including the complete system design, data pipeline, model tuning, and deployment strategies.

Findings

Achieved 85.56% good correction ratio with the tuned PaLM2-XS model.

Successfully deployed the feature on Pixel 8 devices with reduced latency.

Demonstrated the effectiveness of sequential tuning and reward design in model quality.

Abstract

The impressive capabilities in Large Language Models (LLMs) provide a powerful approach to reimagine users' typing experience. This paper demonstrates Proofread, a novel Gboard feature powered by a server-side LLM in Gboard, enabling seamless sentence-level and paragraph-level corrections with a single tap. We describe the complete system in this paper, from data generation, metrics design to model tuning and deployment. To obtain models with sufficient quality, we implement a careful data synthetic pipeline tailored to online use cases, design multifaceted metrics, employ a two-stage tuning approach to acquire the dedicated LLM for the feature: the Supervised Fine Tuning (SFT) for foundational quality, followed by the Reinforcement Learning (RL) tuning approach for targeted refinement. Specifically, we find sequential tuning on Rewrite and proofread tasks yields the best quality in SFT stage, and propose global and direct rewards in the RL tuning stage to seek further improvement. Extensive experiments on a human-labeled golden set showed our tuned PaLM2-XS model achieved 85.56\% good ratio. We launched the feature to Pixel 8 devices by serving the model on TPU v5 in Google Cloud, with thousands of daily active users. Serving latency was significantly reduced by quantization, bucket inference, text segmentation, and speculative decoding. Our demo could be seen in \href{https://youtu.be/4ZdcuiwFU7I}{Youtube}.