Lune

ACL Findings2026

Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning

Junseok Kim, Nakyeong Yang, Kyungmin Min, Kyomin Jung

摘要

Self-Consistency improves reasoning reliability through multi-sample aggregation, but incurs substantial inference cost. Adaptive selfconsistency methods mitigate this issue by adjusting the sampling budget; however, they rely on count-based stopping rules that treat all responses equally, often leading to unnecessary sampling. We propose Reliability-Aware Adaptive Self-Consistency (ReASC), which addresses this limitation by reframing adaptive sampling from response counting to evidence sufficiency, leveraging response-level confidence for principled information aggregation. ReASC operates in two stages: a single-sample decision stage that resolves instances confidently answerable from a single response, and a reliability-aware accumulation stage that aggregates responses by jointly leveraging their frequency and confidence. Across five models and four datasets, ReASC consistently achieves the best accuracy-cost trade-off compared to existing baselines, yielding improved inference efficiency across model scales from 3B to 27B parameters. As a concrete example, ReASC reduces inference cost by up to 70% relative to self-consistency while preserving accuracy on GSM8K using Gemma-3-4B-it.

问 Lune

几分钟读透这篇 Paper,不用再花一下午。

Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。

可以从这些问题问起

Lune 可以查阅

全文 · 相关的 Paper

获取答案

免费开始,无需绑卡

脉络

  • Scalable Best-of-N Selection for Large Language Models via Self-Certainty 引用了 Self-Consistency Improves Chain of Thought Reasoning in Language Models
  • Scalable Best-of-N Selection for Large Language Models via Self-Certainty 引用了 Let's Verify Step by Step
  • Let's Verify Step by Step 引用了 Self-Consistency Improves Chain of Thought Reasoning in Language Models
  • Chain-of-Thought Reasoning Without Prompting 引用了 Self-Consistency Improves Chain of Thought Reasoning in Language Models
  • Chain-of-Thought Reasoning Without Prompting 引用了 Let's Verify Step by Step
  • Deep Think with Confidence 引用了 Scalable Best-of-N Selection for Large Language Models via Self-Certainty
它引用的引用它的从左到右按年份排列,引用越多,位置越高。

它引用的 Paper9

  1. Let's Verify Step by StepHunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards 等ICLR 2024 · 3,045 次引用
  2. Self-Consistency Improves Chain of Thought Reasoning in Language ModelsXuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le 等ICLR 2023 · 681 次引用
  3. Chain-of-Thought Reasoning Without PromptingXuezhi Wang, Denny ZhouNeurIPS 2024 · 305 次引用
  4. Scalable Best-of-N Selection for Large Language Models via Self-CertaintyZhewei Kang, Xuandong Zhao, Dawn SongNeurIPS 2025 · 211 次引用
  5. Deep Think with ConfidenceYichao Fu, Xuewei Wang, Hao Zhang, Yuandong Tian 等ICLR 2026 · 171 次引用

相关 Paper

  1. Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMsPranjal Aggarwal, Aman Madaan, Yiming Yang, MausamEMNLP 2023 · 5 次引用
  2. Task-and-Model-Aware Fractal-Consistency for Efficient LLM ReasoningZiqiu Luo, Jianmin Liu, Yukai Miao, Li Chen 等ICML 2026
  3. Optimal Self-Consistency for Efficient Reasoning with Large Language ModelsAustin Feng, Marius Alonso, Ambroise Odonnat, Vasilii Feofanov 等ICML 2026 · 6 次引用
  4. Optimal Bayesian Stopping for Efficient Inference of Consistent LLM AnswersJingkai Huang, Will Ma, Zhengyuan ZhouICML 2026 · 3 次引用
  5. THE PATH OF LEAST RESISTANCE: GUIDING LLM REASONING TRAJECTORIES WITH PREFIX CONSENSUSIshan Jindal, Sai Prashanth Akuthota, Jayant Taneja, Sachin Dev SharmaICLR 2026 · 1 次引用
黄昏的海面,两侧是细线勾勒的悬崖