Dual-Track CoT: Budget-Aware Stepwise Guidance for Small LMs

TL;DR

This paper introduces Dual-Track CoT, a method that enhances small language models' reasoning capabilities by using budget-aware, stepwise guidance, enabling reliable multi-step reasoning under limited tokens and compute.

Contribution

It proposes a novel approach that improves small language models' reasoning efficiency and accuracy through fine-grained, budget-aware stepwise guidance, reducing reliance on large models or extensive sampling.

Findings

Dual-Track CoT improves reasoning accuracy of small LMs under token constraints.

The method achieves better performance with fewer tokens compared to existing approaches.

It demonstrates practical benefits for low-resource or on-device reasoning tasks.

Abstract

Large Language Models (LLMs) solve many reasoning tasks via chain-of-thought (CoT) prompting, but smaller models (about 7 to 8B parameters) still struggle with multi-step reasoning under tight compute and token budgets. Existing test time reasoning methods such as self consistency (sampling multiple rationales and voting), Tree-of-Thoughts (search over intermediate thoughts), and critique revise loops improve performance, but often at high token cost and without fine-grained step-level control. This project1 aims to address that gap: can Small Language Models (SLMs) reason reliably using the same or fewer tokens? This question is both scientific and practical. Scientifically, it probes whether process supervision and simple test-time controls (such as token budgets and rejection of redundant steps) can substitute for model scale or large sampling counts. Practically, many deployments (on-device, low-latency, or cost-constrained settings) cannot afford huge models or dozens of sampled rationales per query. A method that improves SLM reasoning at fixed cost would therefore be directly useful.