ILRR: Inference-Time Steering Method for Masked Diffusion Language Models
This work addresses the underexplored challenge of controlling text generation in non-autoregressive models, offering a practical solution for applications requiring attribute-specific outputs, though it is incremental as it builds on existing diffusion model frameworks.
The paper tackles the problem of inference-time control for discrete diffusion language models by introducing ILRR, a learning-free framework that uses a reference sequence to steer generation, achieving attribute accuracy improvements of 10% to 60% over baselines with minimal computational overhead.
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.