Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs

TL;DR

This paper introduces Adaptive Semantic-Aware Typicality Sampling (ASTS), an improved decoding strategy for large language models that enhances diversity, coherence, and fluency in generated text through dynamic adjustments and multi-objective scoring.

Contribution

The paper proposes ASTS, a novel decoding algorithm that advances Locally Typical Sampling by incorporating adaptive thresholds and multi-objective optimization for better text generation.

Findings

ASTS outperforms existing methods in reducing repetition.

ASTS improves semantic alignment and fluency.

Experimental results show enhanced diversity and coherence.

Abstract

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.