Writing-Zero: Bridge the Gap Between Non-verifiable Tasks and Verifiable Rewards

TL;DR

This paper introduces Writing-Zero, a novel RLVR-based training framework that enhances language models' creative writing abilities by transforming subjective assessments into verifiable rewards, reducing reward hacking, and improving performance on writing tasks.

Contribution

It proposes a unified RLVR paradigm with a pairwise GenRM and BRPO algorithm, enabling robust, reference-free training for creative language tasks without supervised fine-tuning.

Findings

Writing-Zero improves writing quality and robustness against reward hacking.

The approach achieves competitive results on multiple writing benchmarks.

It demonstrates potential to unify various reward modeling methods under RLVR.

Abstract

Reinforcement learning with verifiable rewards (RLVR) has enabled large language models (LLMs) to achieve remarkable breakthroughs in reasoning tasks with objective ground-truth answers, such as mathematics and code generation. However, a significant gap remains for non-verifiable tasks, like creative writing and open-ended dialogue, where quality assessment is inherently subjective and lacks definitive references. Existing approaches for these domains often rely on scalar reward models trained with human preferences, which suffer from limited generalization and are prone to reward hacking, such as over-explanation and length bias. In this work, we propose a unified RLVR-based training paradigm that bridges the gap between non-verifiable tasks and verifiable rewards. We introduce a writing-principle-based pairwise Generative Reward Model (GenRM) and a novel Bootstrapped Relative Policy Optimization (BRPO) algorithm. The pairwise writing GenRM leverages self-principled critique to transform subjective assessments into reliable, verifiable rewards, while BRPO enables dynamic, reference-free pairwise comparison by leveraging a bootstrapped response as temporary reference from within group rollouts during RL training. Our approach empowers LLMs to develop robust writing capabilities without supervised fine-tuning, as demonstrated by Writing-Zero, which shows consistent improvement and strong resistance to reward hacking compared to scalar reward baselines. Furthermore, our method achieves competitive results on both in-house and open-source writing benchmarks. Our findings suggest the potential to unify rule-based, reference-based, and reference-free reward modeling under the RLVR framework, thus paving the way for a comprehensive and scalable RL training paradigm applicable across all language tasks.