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

TL;DR

ReASC addresses the cost of Self-Consistency in LLM reasoning by replacing purely count-based adaptive sampling with reliability-aware evidence accumulation. It introduces a two-stage inference process: Stage 1 uses a single-sample decision based on Bottom $10\%$ Group Confidence to quickly resolve easy cases, and Stage 2 aggregates evidence with a confidence-weighted Beta update, weighting each sample by a standardized confidence score. The method integrates offline and online calibration to set gating thresholds and uses a Beta-based stopping rule tied to $P(p_1>p_2\mid V)\ge C_{\mathrm{threshold}}$, enabling earlier stopping when high-confidence evidence emerges. Across five models and four datasets, ReASC consistently improves the accuracy-cost trade-off (Acc/TF), achieving up to about $70\%$ cost reduction relative to standard Self-Consistency while preserving accuracy, and demonstrating robust performance from $3\mathrm{B}$ to $27\mathrm{B}$ parameters.

Abstract

Self-Consistency improves reasoning reliability through multi-sample aggregation, but incurs substantial inference cost. Adaptive self-consistency 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.

Figures (8)

• Figure 1: Count-based stopping may lead to inefficient evidence accumulation. Ignoring response reliability, count-based criteria may require unnecessary additional samples, while ReASC reaches the same decision with fewer samples.
• Figure 2: Comparison of two confidence signals. Using Gemma 3 4B-Instruct on MATH500, Bottom 10% Group Confidence shows a larger separation between correct and incorrect responses than Response-level Self-Certainty.
• Figure 3: Overview of ReASC. The model first attempts a Single-Sample Decision (Stage 1) by evaluating whether the response reliability is sufficient. If not, it proceeds to Reliability-Aware Accumulation (Stage 2), where responses are adaptively sampled and aggregated via confidence-weighted Beta updates.
• Figure 4: Stage 1 acceptance ratio versus model size. Acceptance increases with model scale across datasets.
• Figure 5: Confidence-weighted update improves sampling efficiency. Each sampled response updates a Beta posterior, shown as the shaded region in each plot. With sampling stopping at $p \geq 0.95$, ASC requires seven uniform updates, while ReASC reaches in four confidence-weighted updates, reducing sample cost by 43%.