Year: 2026
Type: conference-paper
Abstract: Test-time scaling (TTS) enhances LLM reasoning capabilities by sampling and aggregating diverse solution trajectories.However, existing approaches often rely on external verifiers and one-shot independent sampling, which results in inefficient budget allocation and underutilizes interim high-quality trajectories.We propose ConMA, a trainingfree, verifier-free TTS framework that reallocates a fixed inference budget into iterative sample-filter-diversify-select cycles: it filters answer groups based on intrinsic tokenprobability confidence, enriches candidates through diversity-aware expansion, and employs repeated single-choice selection for multistage refinement.Across multiple benchmarks, ConMA consistently improves accuracy under fixed budgets.With a maximum budget of N = 64, ConMA boosts Qwen3-4B to 80% accuracy on AIME25, significantly outperforming strong baselines while converging early with only 18 samples on average, substantially reducing inference cost.
Institution: Beijing Institute of Technology
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166877517