ILRR: Inference-Time Steering Method for Masked Diffusion Language Models
cs.CL, cs.AI, cs.LG
Submitted: 2026-01-29
Updated: 2026-09-20
Comments: Accepted to AACL-IJCNLP 2026 Main
License: http://creativecommons.org/licenses/by/4.0/
The gist: Discrete Diffusion Language Models (DLMs) offer a promising non-autoregressive alternative for text generation, yet effective mechanisms for inference-time control remain relatively underexplored.
Terminology
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 or trajectory optimization mechanisms. In this work, we study the paradigm of reference-based latent steering for DLMs. We introduce Iterative Latent Representation Refinement (ILRR), an efficient framework for steering DLMs using a reference text as a high-level semantic blueprint. ILRR extracts and injects reference-derived semantic signals into the evolving activations of the generated sequence, enabling tunable transfer of coarse properties such as sentiment. We further introduce Spatially Modulated Steering, an extension that enables long-form generation to be guided by shorter references by regulating intensity across the sequence. Empirically, we demonstrate that ILRR achieves effective control on LLaDA and MDLM architectures with low computational overhead, requiring only one additional parallel forward pass per denoising step. Under comparable compute budgets, ILRR improves attribute accuracy over baselines by 10% to 60% points. Our results suggest that the iterative, global denoising process makes DLMs a natural substrate for effective sequence-wide activation-level control.
Sources
- Plug and Play Language Models: A Simple Approach to Controlled Text Generation
- SSD-LM: Semi-autoregressive Simplex-based Diffusion Language Model for Text Generation and Modular Control
- GeDi: Generative Discriminator Guided Sequence Generation
- Refusal in Language Models Is Mediated by a Single Direction
- DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Experts
- Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution
- SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations
- ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
- Inference-Time Scaling of Diffusion Language Models via Trajectory Refinement
- Steering Llama 2 via Contrastive Activation Addition
- Steering Language Models With Activation Engineering
- Simple and Effective Masked Diffusion Language Models
- Simple Guidance Mechanisms for Discrete Diffusion Models
- Implicit Search via Discrete Diffusion: A Study on Chess
- Deep Unsupervised Learning using Nonequilibrium Thermodynamics
- Generative Modeling by Estimating Gradients of the Data Distribution
- Dream 7B: Diffusion Large Language Models
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering