# Controlled LLM Decoding via Discrete Auto-regressive Biasing

arXiv: https://arxiv.org/abs/2502.03685 · HTML: https://mlsanity.com/paper/2502.03685/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 6 Feb 2025.

## Built on by (most cited first, 5 of 5)

- [Guardians and Offenders: A Survey on Harmful Content Generation and Safety Mitigation of LLM](https://mlsanity.com/paper/2508.05775/) (2025)
- [Entropy-Aligned Decoding of LMs for Better Writing and Reasoning](https://mlsanity.com/paper/2601.01714/) (2026)
- [Decoding in Geometry: Alleviating Embedding-Space Crowding for Complex Reasoning](https://mlsanity.com/paper/2601.22536/) (2026)
- [DRIV-EX: Counterfactual Explanations for Driving LLMs](https://mlsanity.com/paper/2603.00696/) (2026)
- [Gradient-Guided Reward Optimization for Inference-time Alignment](https://mlsanity.com/paper/2606.09635/) (2026)

## Abstract (arXiv)

Controlled text generation allows for enforcing user-defined constraints on large language model outputs, an increasingly important field as LLMs become more prevalent in everyday life. One common approach uses energy-based decoding, which defines a target distribution through an energy function that combines multiple constraints into a weighted average. However, these methods often struggle to balance fluency with constraint satisfaction, even with extensive tuning of the energy function's coefficients. In this paper, we identify that this suboptimal balance arises from sampling in continuous space rather than the natural discrete space of text tokens. To address this, we propose Discrete Auto-regressive Biasing, a controlled decoding algorithm that leverages gradients while operating entirely in the discrete text domain. Specifically, we introduce a new formulation for controlled text generation by defining a joint distribution over the generated sequence and an auxiliary bias sequence. To efficiently sample from this joint distribution, we propose a Langevin-within-Gibbs sampling algorithm using gradient-based discrete MCMC. Our method significantly improves constraint satisfaction while maintaining comparable or better fluency, all with even lower computational costs. We demonstrate the advantages of our controlled decoding method on sentiment control, language detoxification, and keyword-guided generation.