# (G)I-DLE: Generative Inference via Distribution-preserving Logit Exclusion with KL Divergence Minimization for Constrained Decoding

arXiv: https://arxiv.org/abs/2503.18050 · HTML: https://mlsanity.com/paper/2503.18050/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 4 of its 5 references that are in the map.
- First posted on arXiv 23 Mar 2025.

## Builds on (most cited first, 4 of 4)

- [Neural Text Generation with Unlikelihood Training](https://mlsanity.com/paper/1908.04319/) (2019)
- [Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs](https://mlsanity.com/paper/2306.03081/) (2023)
- [Lexically Constrained Neural Machine Translation with Levenshtein Transformer](https://mlsanity.com/paper/2004.12681/) (2020)
- [Rejection sampling from shape-constrained distributions in sublinear time](https://mlsanity.com/paper/2105.14166/) (2021)

## Built on by (most cited first, 1 of 1)

- [Repair, Not Improvement: Decomposing Constrained Decoding in Tool-Call Abstention](https://mlsanity.com/paper/2608.13959/) (2026)

## Abstract (arXiv)

We propose (G)I-DLE, a new approach to constrained decoding that leverages KL divergence minimization to preserve the intrinsic conditional probability distribution of autoregressive language models while excluding undesirable tokens. Unlike conventional methods that naively set banned tokens' logits to -∞, which can distort the conversion from raw logits to posterior probabilities and increase output variance, (G)I-DLE re-normalizes the allowed token probabilities to minimize such distortion. We validate our method on the K2-Eval dataset, specifically designed to assess Korean language fluency, logical reasoning, and cultural appropriateness. Experimental results on Qwen2.5 models (ranging from 1.5B to 14B) demonstrate that G-IDLE not only boosts mean evaluation scores but also substantially reduces the variance of output quality.