Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute

2025/04/01 by Jianhao Chen, Chen, Jianhao, Bocheng Zhou +18 · 12 citations
Computer Science · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #Consistency (knowledge bases) #FOS: Computer and information sciences #Inference #Leverage (statistics) #Machine Learning in Healthcare #Sampling (signal processing) #Scaling #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2504.00762

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/04/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

This paper presents a simple, effective, and cost-efficient strategy to improve LLM performance by scaling test-time compute. Our strategy builds upon the repeated-sampling-then-voting framework, with a novel twist: incorporating multiple models, even weaker ones, to leverage their complementary strengths that potentially arise from diverse training data and paradigms. By using consistency as a signal, our strategy dynamically switches between models. Theoretical analysis highlights the efficiency and performance advantages of our strategy. Extensive experiments on six datasets demonstrate that our strategy not only outperforms self-consistency and state-of-the-art multi-agent debate approaches, but also significantly reduces inference costs. Additionally, ModelSwitch requires only a few comparable LLMs to achieve optimal performance and can be extended with verification methods, demonstrating the potential of leveraging multiple LLMs in the generation-verification paradigm.

Cited by

• Budget-Aware Tool-Use Enables Effective Agent Scaling
• C3PO: Optimized Large Language Model Cascades with Probabilistic Cost Constraints for Reasoning
• FEval-TTC: Fair Evaluation Protocol for Test-Time Compute
• Flip-Flop Consistency: Unsupervised Training for Robustness to Prompt Perturbations in LLMs
• Learning Compact Representations of LLM Abilities via Item Response Theory
• Making, not Taking, the Best of N
• Semantic Agreement Enables Efficient Open-Ended LLM Cascades
• Mixture of Thoughts: Learning to Aggregate What Experts Think, Not Just What They Say
• TalkPlayData 2: An Agentic Synthetic Data Pipeline for Multimodal Conversational Music Recommendation
• When Life Gives You Samples: The Benefits of Scaling up Inference Compute for Multilingual LLMs
• The Avengers: A Simple Recipe for Uniting Smaller Language Models to Challenge Proprietary Giants
• Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time Scaling

Related

• Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
• Scaling Test-time Compute for LLM Agents
• Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling
• Learning to Reason Across Parallel Samples for LLM Reasoning
• A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning