Enregistré dans:
Détails bibliographiques
Auteurs principaux: Davoodi, Arash Gholami, Rezazadeh, Navid, Davoudi, Seyed Pouyan Mousavi, Pezeshkpour, Pouya
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2602.10346
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914563966894080
author Davoodi, Arash Gholami
Rezazadeh, Navid
Davoudi, Seyed Pouyan Mousavi
Pezeshkpour, Pouya
author_facet Davoodi, Arash Gholami
Rezazadeh, Navid
Davoudi, Seyed Pouyan Mousavi
Pezeshkpour, Pouya
contents Large language models (LLMs) must balance diversity and creativity against logical coherence in open-ended generation. Existing truncation-based samplers are effective but largely heuristic, relying mainly on probability mass and entropy while ignoring semantic geometry of the token space. We present Top-W, a geometry-aware truncation rule that uses Wasserstein distance-defined over token-embedding geometry-to keep the cropped distribution close to the original, while explicitly balancing retained probability mass against the entropy of the kept set. Our theory yields a simple closed-form structure for the fixed-potential subset update: depending on the mass-entropy trade-off, the optimal crop either collapses to a single token or takes the form of a one-dimensional prefix that can be found efficiently with a linear scan. We implement Top-W using efficient geometry-based potentials (nearest-set or k-NN) and pair it with an alternating decoding routine that keeps the standard truncation-and-sampling interface unchanged. Extensive experiments on four benchmarks (GSM8K, GPQA, AlpacaEval, and MT-Bench) across three instruction-tuned models show that Top-W consistently outperforms prior state-of-the-art decoding approaches achieving up to 33.7% improvement. Moreover, we find that Top-W not only improves accuracy-focused performance, but also boosts creativity under judge-based open-ended evaluation.
format Preprint
id arxiv_https___arxiv_org_abs_2602_10346
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Geometry-Aware Decoding with Wasserstein-Regularized Truncation and Mass Penalties for Large Language Models
Davoodi, Arash Gholami
Rezazadeh, Navid
Davoudi, Seyed Pouyan Mousavi
Pezeshkpour, Pouya
Computation and Language
Machine Learning
Large language models (LLMs) must balance diversity and creativity against logical coherence in open-ended generation. Existing truncation-based samplers are effective but largely heuristic, relying mainly on probability mass and entropy while ignoring semantic geometry of the token space. We present Top-W, a geometry-aware truncation rule that uses Wasserstein distance-defined over token-embedding geometry-to keep the cropped distribution close to the original, while explicitly balancing retained probability mass against the entropy of the kept set. Our theory yields a simple closed-form structure for the fixed-potential subset update: depending on the mass-entropy trade-off, the optimal crop either collapses to a single token or takes the form of a one-dimensional prefix that can be found efficiently with a linear scan. We implement Top-W using efficient geometry-based potentials (nearest-set or k-NN) and pair it with an alternating decoding routine that keeps the standard truncation-and-sampling interface unchanged. Extensive experiments on four benchmarks (GSM8K, GPQA, AlpacaEval, and MT-Bench) across three instruction-tuned models show that Top-W consistently outperforms prior state-of-the-art decoding approaches achieving up to 33.7% improvement. Moreover, we find that Top-W not only improves accuracy-focused performance, but also boosts creativity under judge-based open-ended evaluation.
title Geometry-Aware Decoding with Wasserstein-Regularized Truncation and Mass Penalties for Large Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2602.10346