Balancing Diversity and Risk in LLM Sampling: How to Select Your Method and Parameter for Open-Ended Text Generation

2024-08-24

arXiv.org

AI Summary

To address the challenge of simultaneously ensuring safety and diversity in open-ended generation by large language models (LLMs), this paper proposes a context-aware dynamic truncation sampling optimization method. The core contribution is twofold: first, it introduces a prefix-tree structure built over complete-sentence contexts to quantitatively model truncation sampling capacity; second, it establishes a joint diversity–risk evaluation framework that enables interpretable, adaptive co-selection of temperature and top-k/top-p parameters. Experimental results on multiple open-generation benchmarks demonstrate significant improvements in generation stability: high-risk repetition and hallucination rates decrease by 12.7%, while the Dist-2 diversity metric improves by 3.2% over baseline methods. The approach thus achieves a principled trade-off between safety and expressiveness without compromising either dimension.

Technology Category

• Natural Language Processing: Safety and Robustness
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Sampling-based decoding strategies have been widely adopted for Large Language Models (LLMs) in numerous applications, targeting a balance between diversity and quality via temperature tuning and tail truncation. Considering the strong dependency of the candidate next tokens on different prefixes, recent studies propose to adaptively truncate the tail of LLMs' predicted distribution. Although improved results have been reported with these methods on open-ended text generation tasks, the results are highly dependent on the curated parameters and the limited exemplar text. In this paper, we propose a systematic way to estimate the capacity of a truncation sampling method by considering the trade-off between diversity and risk at each decoding step, based on our collected prefix tree which preserves the context of a full sentence. Our work offers a comprehensive comparison of existing truncation sampling methods and serves as a practical user guideline for their parameter selection.

Problem

• Large Language Models
• Safety
• Diversity

Innovation

• Dynamic Tail Trimming
• Sampling Decoding Optimization
• Text Generation Safety