HiFlow: Hierarchical Feedback-Driven Optimization for Constrained Long-Form Text Generation

Researchers behind arXiv paper 2603.04996v2 introduced LoopWeaver, a hierarchical workflow that uses two sequential offline preference-optimization stages to improve constrained long-form text generation in large language models. LoopWeaver combines constraint-aware planning, feasibility filtering and a shared deterministic reward to build preference pairs for plan-level and generation-level DPO, while inference stays feed-forward. Across four LongGenBench-style structured scenarios and multiple open-source backbones, the method improved both constraint accuracy and text quality, with software and data released at github.com/cgtaa-cmyk/LoopWeaver.

arXiv:2603.04996v2 Announce Type: replace Abstract: Large Language Models face a multi-objective optimization problem in constrained long-text generation, balancing global structure, local coherence, and constraint satisfaction. We propose LoopWeaver, a hierarchical workflow with two sequential offline preference-optimization stages. Constraint-aware planning produces structured plans, feasibility filtering removes invalid plans, and a shared deterministic reward constructs preference pairs for plan-level and generation-level DPO. At inference time, LoopWeaver remains feed-forward: it generates and revises a complete plan, filters the revised plan, and then generates segments conditioned on the current sub-plan and running narrative. We evaluate the method on four LongGenBench-style structured scenarios across multiple backbones. Results show improvements in both constraint accuracy and text quality across all four open-source backbones. Our software and data are publicly available at https://github.com/cgtaa-cmyk/LoopWeaver.