On Next-Token Prediction in LLMs: How End Goals Determine the Consistency of Decoding Algorithms
Not on the map. Why?
Abstract (arXiv)
Probabilistic next-token prediction trained using cross-entropy loss is the basis of most large language models. Given a sequence of previous values, next-token prediction assigns a probability to each possible next value in the vocabulary. There are many ways to use next-token prediction to output token sequences. This paper examines a few of these algorithms (greedy, lookahead, random sampling, and temperature-scaled random sampling) and studies their consistency with respect to various goals encoded as loss functions. Although consistency of surrogate losses with respect to a target loss function is a well researched topic, we are the first to study it in the context of LLMs (to the best of our knowledge). We find that, so long as next-token prediction converges to its true probability distribution, random sampling is consistent with outputting sequences that mimic sampling from the true probability distribution. For the other goals, such as minimizing the 0-1 loss on the entire sequence, we show no polynomial-time algorithm is optimal for all probability distributions and all decoding algorithms studied are only optimal for a subset of probability distributions. When analyzing these results, we see that there is a dichotomy created between the goals of information retrieval and creative generation for the decoding algorithms. This shows that choosing the correct decoding algorithm based on the desired goal is extremely important and many of the ones used are lacking theoretical grounding in numerous scenarios.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 21 of its 34 references that are in the map. First posted on arXiv 16 May 2025.
Builds on
21 of 34 references.
References · 21
- 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
- 2020Language Models are Few-Shot Learnerscited by 32,419
- 2023GPT-4 Technical Reportcited by 18,396
- 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
- 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
- 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
- 2019Language Models as Knowledge Bases?cited by 1,936
- 2024Break the Sequential Dependency of LLM Inference Using Lookahead Decodingcited by 298
- 2024The pitfalls of next-token predictioncited by 160
- 2024A Thorough Examination of Decoding Methods in the Era of LLMscited by 114
- 2022On Decoding Strategies for Neural Text Generatorscited by 90
- 2020A Mathematical Exploration of Why Language Models Help Solve Downstream Taskscited by 72
9 more
- 2023Hot or Cold? Adaptive Temperature Sampling for Code Generation with Large Language Modelscited by 38
- 2024Mechanics of Next Token Prediction with Self-Attentioncited by 34
- 2014Convex Calibration Dimension for Multiclass Loss Matricescited by 21
- 2024Improving Open-Ended Text Generation via Adaptive Decodingcited by 21
- 2020Multi-label classification: do Hamming loss and subset accuracy really conflict with each other?cited by 14
- 2024Implicit Optimization Bias of Next-Token Prediction in Linear Modelscited by 14
- 2024Adaptive Decoding via Latent Preference Optimizationcited by 10
- 2023Foundation Models for Natural Language Processing -- Pre-trained Language Models Integrating Mediacited by 8
- 2025Reasoning Bias of Next Token Prediction Trainingcited by 3
Built on by
Citing papers per year: 2026: 1.
1 paper.