ILRR: Inference-Time Steering Method for Masked Diffusion Language Models

2026-01-29

AI Summary

Existing discrete diffusion language models lack efficient inference-time control mechanisms, making it difficult to flexibly steer high-level semantic attributes of generated text. This work proposes ILRR—an inference-time guidance framework that requires no additional training—enabling precise semantic control by dynamically aligning the internal activations of the denoising sequence with those of a single reference sequence. The core innovations include iterative latent representation refinement and a spatial modulation guidance strategy, which together allow efficient adjustment of guidance strength with only a single forward pass. Experiments on LLaDA and MDLM demonstrate that ILRR improves attribute control accuracy by 10%–60% with minimal computational overhead while preserving generation quality.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Discrete Diffusion Language Models (DLMs) offer a promising non-autoregressive alternative for text generation, yet effective mechanisms for inference-time control remain relatively underexplored. Existing approaches include sampling-level guidance procedures or trajectory optimization mechanisms. In this work, we introduce Iterative Latent Representation Refinement (ILRR), a learning-free framework for steering DLMs using a single reference sequence. ILRR guides generation by dynamically aligning the internal activations of the generated sequence with those of a given reference throughout the denoising process. This approach captures and transfers high-level semantic properties, with a tunable steering scale enabling flexible control over attributes such as sentiment. We further introduce Spatially Modulated Steering, an extension that enables steering long texts using shorter references by regulating guidance intensity across the sequence. Empirically, we demonstrate that ILRR achieves effective attribute steering on LLaDA and MDLM architectures with a minor computational overhead, requiring only one additional parallel forward pass per denoising step. Under the same compute budget, ILRR improves attribute accuracy over comparable baselines by 10$\%$ to 60$\%$ points, while maintaining high generation quality.

Problem

• Discrete Diffusion Language Models
• inference-time control
• attribute steering
• non-autoregressive generation
• semantic guidance

Innovation

• Iterative Latent Representation Refinement
• Inference-Time Steering
• Discrete Diffusion Language Models
• Spatially Modulated Steering
• Non-Autoregressive Generation