# STAIF: A Stage-wise Optimization for Complex Instruction Following

arXiv: https://arxiv.org/abs/2607.22649 · HTML: https://mlsanity.com/paper/2607.22649/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 18 references that are in the map.
- First posted on arXiv 26 Jun 2026.

## Builds on (most cited first, 10 of 10)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society](https://mlsanity.com/paper/2303.17760/) (2023)
- [VerIF: Verification Engineering for Reinforcement Learning in Instruction Following](https://mlsanity.com/paper/2506.09942/) (2025)
- [IOPO: Empowering LLMs with Complex Instruction Following via Input-Output Preference Optimization](https://mlsanity.com/paper/2411.06208/) (2024)
- [The Impossibility of Fair LLMs](https://mlsanity.com/paper/2406.03198/) (2024)
- [MuSC: Improving Complex Instruction Following with Multi-granularity Self-Contrastive Training](https://mlsanity.com/paper/2502.11541/) (2025)

## Abstract (arXiv)

Following complex instructions with multiple explicit constraints remains a fundamental challenge for large language models (LLMs). Existing alignment methods, such as DPO, optimize holistic reward signals that often underemphasize strict satisfaction of individual constraints, particularly under out-of-distribution or multi-constraint settings. In this paper, we propose STAIF, a stage-wise optimization framework that decouples the alignment of subjective (soft) constraints from the optimization of objectively verifiable (hard) constraints. Stage 1 applies preference optimization with multiple negative samples to sharpen sensitivity to soft constraints, while Stage 2 applies Reinforcement Learning with Verifiable Rewards (RLVR) to enforce strict compliance with hard constraints. To support this method, we construct STAINSTRUCT, a high-quality bilingual (English, Chinese) dataset of approximately 31,000 complex multi-constraint instructions. Extensive analyses validate the design of STAIF and show state-of-the-art performance on representative benchmarks against strong baselines, as well as genuine generalization.