Bridging Auxiliary Constraints to Resolve Instruction Following in Large Reasoning Models

2026-06-02

AI Summary

This work addresses the challenge large reasoning models face in effectively coordinating multiple instructions, particularly when confronted with conflicting constraints. The authors propose the Constraint Relation Graph Completion (CRGC) framework, which formalizes instructions as a structured constraint knowledge graph. By explicitly modeling inter-constraint relationships and generating bridging constraints, CRGC guides the model to accurately identify and reconcile competing instruction requirements. Integrating knowledge graph construction, constraint relation modeling, and prompt engineering, the method reduces constraint violations by 39% compared to standard prompting across three mainstream instruction-following benchmarks, while preserving baseline reasoning capabilities. This approach substantially enhances performance on complex instruction-following tasks.

Abstract

Large Reasoning Models (LRMs) have demonstrated impressive capabilities in many tasks, yet they struggle with reliably following multiple instructions, either by failing to satisfy individual constraints or by struggling to balance competing constraints simultaneously. We formalize this challenge as the Constraint Adherence Problem (CAP). This paper introduces a novel framework that addresses CAP by representing instructions as a structured knowledge graph of constraints. Our approach, Constraint Relationship Graph Completion (CRGC), explicitly models relationships between constraints, identifies adherence challenges, and discovers ``bridge constraints'' that help the model better focus on and reconcile requirements. Bridge constraints act as auxiliary instructions that make primary constraints more salient and compatible. Unlike existing approaches that enhance instruction following through general training methods, CRGC specifically improves constraint satisfaction by leveraging the model's own knowledge to create better pathways for generation. Experiments across three popular instruction following datasets demonstrate that our approach reduces constraint violations by 39% compared to standard prompting while maintaining reasoning abilities of large reasoning models.

Problem

• Instruction Following
• Constraint Adherence
• Large Reasoning Models
• Multi-constraint Satisfaction

Innovation

• Constraint Relationship Graph Completion
• Bridge Constraints
• Instruction Following
• Large Reasoning Models
• Constraint Adherence Problem