Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sen, Jaydip, Sengupta, Saptarshi, Dasgupta, Subhasis
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913178368081920
author Sen, Jaydip
Sengupta, Saptarshi
Dasgupta, Subhasis
author_facet Sen, Jaydip
Sengupta, Saptarshi
Dasgupta, Subhasis
contents This chapter explores advancements in decoding strategies for large language models (LLMs), focusing on enhancing the Locally Typical Sampling (LTS) algorithm. Traditional decoding methods, such as top-k and nucleus sampling, often struggle to balance fluency, diversity, and coherence in text generation. To address these challenges, Adaptive Semantic-Aware Typicality Sampling (ASTS) is proposed as an improved version of LTS, incorporating dynamic entropy thresholding, multi-objective scoring, and reward-penalty adjustments. ASTS ensures contextually coherent and diverse text generation while maintaining computational efficiency. Its performance is evaluated across multiple benchmarks, including story generation and abstractive summarization, using metrics such as perplexity, MAUVE, and diversity scores. Experimental results demonstrate that ASTS outperforms existing sampling techniques by reducing repetition, enhancing semantic alignment, and improving fluency.
format Preprint
id arxiv_https___arxiv_org_abs_2506_05387
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs
Sen, Jaydip
Sengupta, Saptarshi
Dasgupta, Subhasis
Computation and Language
Artificial Intelligence
This chapter explores advancements in decoding strategies for large language models (LLMs), focusing on enhancing the Locally Typical Sampling (LTS) algorithm. Traditional decoding methods, such as top-k and nucleus sampling, often struggle to balance fluency, diversity, and coherence in text generation. To address these challenges, Adaptive Semantic-Aware Typicality Sampling (ASTS) is proposed as an improved version of LTS, incorporating dynamic entropy thresholding, multi-objective scoring, and reward-penalty adjustments. ASTS ensures contextually coherent and diverse text generation while maintaining computational efficiency. Its performance is evaluated across multiple benchmarks, including story generation and abstractive summarization, using metrics such as perplexity, MAUVE, and diversity scores. Experimental results demonstrate that ASTS outperforms existing sampling techniques by reducing repetition, enhancing semantic alignment, and improving fluency.
title Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.05387