IFDECORATOR: Wrapping Instruction Following Reinforcement Learning with Verifiable Rewards

TL;DR

IFDecorator enhances instruction-following reinforcement learning for large language models by improving training efficiency, preventing reward hacking, and ensuring better alignment with user intent through a novel framework.

Contribution

We introduce IFDecorator, a robust framework that wraps RLVR training with components to improve difficulty assessment, intent alignment, and reward hacking detection.

Findings

Achieved 87.43% accuracy on IFEval, surpassing larger models.

Significantly reduced reward hacking rates with trip wires.

Improved performance on FollowBench while maintaining general capabilities.

Abstract

Reinforcement Learning with Verifiable Rewards (RLVR) improves instruction following capabilities of large language models (LLMs), but suffers from training inefficiency due to inadequate difficulty assessment. Moreover, RLVR is prone to over-optimization, where LLMs exploit verification shortcuts without aligning to the actual intent of user instructions. We introduce Instruction Following Decorator (IFDecorator}, a framework that wraps RLVR training into a robust and sample-efficient pipeline. It consists of three components: (1) a cooperative-adversarial data flywheel that co-evolves instructions and hybrid verifications, generating progressively more challenging instruction-verification pairs; (2) IntentCheck, a bypass module enforcing intent alignment; and (3) trip wires, a diagnostic mechanism that detects reward hacking via trap instructions, which trigger and capture shortcut exploitation behaviors. Our Qwen2.5-32B-Instruct-IFDecorator achieves 87.43% accuracy on IFEval, outperforming larger proprietary models such as GPT-4o. Additionally, we demonstrate substantial improvements on FollowBench while preserving general capabilities. Our trip wires show significant reductions in reward hacking rates. We will release models, code, and data for future research.