Improving the Diversity of LLM Outputs without a Trade-off
Not on the map. Why?
Abstract (arXiv)
We propose DAST (Diversifying Arithmetic Sampling with TokenTour), a method that increases the diversity of LLM outputs without any change to the marginal distribution and with negligible generation-time overhead (a few microseconds). We observe that token IDs are often arranged in a meaningless order and reassign them so that tokens with similar meanings appear consecutively. This can be done in advance in a few hundred seconds per model, and the resulting order can be reused for all subsequent generations. By combining this order with arithmetic sampling (or quasi-Monte Carlo methods), we make similar tokens less likely to be generated across runs while preserving the distribution. Our method not only produces qualitatively good ideas but also significantly improves performance on the downstream task of ProtoQA.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 33 of its 37 references that are in the map. First posted on arXiv 27 Sep 2026.
Builds on
33 of 37 references.
References · 33
- 2024Qwen2.5 Technical Reportcited by 4,942
- 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,810
- 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 991
- 2025SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Modelcited by 335
- 2019Stochastic Beams and Where to Find Them: The Gumbel-Top-k Trick for Sampling Sequences Without Replacementcited by 211
- 2021High Quality Rather than High Model Probability: Minimum Bayes Risk Decoding with Neural Metricscited by 104
- 2018Language GANs Falling Shortcited by 97
- 2025Jointly Reinforcing Diversity and Quality in Language Model Generationscited by 80
- 2019Comparison of Diverse Decoding Methods from Conditional Language Modelscited by 78
- 2025No Prompt Left Behind: Exploiting Zero-Variance Prompts in LLM Reinforcement Learning via Entropy-Guided Advantage Shapingcited by 50
- 2020ProtoQA: A Question Answering Dataset for Prototypical Common-Sense Reasoningcited by 46
- 2025Diversity-Aware Policy Optimization for Large Language Model Reasoningcited by 41
21 more
- 2025Modifying Large Language Model Post-Training for Diverse Creative Writingcited by 40
- 2025Multilingual Prompting for Improving LLM Generation Diversitycited by 25
- 2020Incremental Sampling Without Replacement for Sequence Modelscited by 23
- 2022Arithmetic Sampling: Parallel Diverse Decoding for Large Language Modelscited by 19
- 2021Determinantal Beam Searchcited by 18
- 2025On the Effect of Sampling Diversity in Scaling LLM Inferencecited by 18
- 2025Representation-Based Exploration for Language Models: From Test-Time to Post-Trainingcited by 16
- 2025G2: Guided Generation for Enhanced Output Diversity in LLMscited by 14
- 2025How Diversely Can Language Models Solve Problems? Exploring the Algorithmic Diversity of Model-Generated Codecited by 11
- 2026SetPO: Set-Level Policy Optimization for Diversity-Preserving LLM Reasoningcited by 11
- 2026Prune as You Generate: Online Rollout Pruning for Faster and Better RLVRcited by 10
- 2025Semantic-guided Diverse Decoding for Large Language Modelcited by 9
- 2025Asking a Language Model for Diverse Responsescited by 9
- 2025Optimizing Diversity and Quality through Base-Aligned Model Collaborationcited by 8
- 20243D-Prover: Diversity Driven Theorem Proving With Determinantal Point Processescited by 6
- 2026Beyond Mode Collapse: Distribution Matching for Diverse Reasoningcited by 6
- 2026Free Lunch for Pass@k? Low Cost Diverse Sampling for Diffusion Language Modelscited by 5
- 2024Quasi-random Multi-Sample Inference for Large Language Modelscited by 4
- 2026QuasiMoTTo: Quasi-Monte Carlo Test-Time Scalingcited by 4
- 2026Exploring Diverse Generation Paths via Inference-time Stiefel Activation Steeringcited by 2
- 2022Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problemcited by 1
Built on by
Not cited yet.