Reward-Guided Speculative Decoding for Efficient LLM Reasoning

We introduce Reward-Guided Speculative Decoding (RSD), a novel framework aimed at improving the efficiency of inference in large language models (LLMs). RSD synergistically combines a lightweight draft model with a more powerful target model, incorporating a controlled bias to prioritize high-reward outputs, in contrast to existing speculative decoding methods that enforce strict unbiasedness.RSD employs a process reward model to evaluate intermediate decoding steps and dynamically decide whether to invoke the target model, optimizing the trade-off between computational cost and output quality. We theoretically demonstrate that a threshold-based mixture strategy achieves an optimal balance between resource utilization and performance. Extensive evaluations on challenging reasoning benchmarks, including Olympiad-level tasks, show that RSD delivers significant efficiency gains against decoding with the target model only (up to 4.4X fewer FLOPs), while achieving significant better accuracy than parallel decoding method on average (up to +3.5). These results highlight RSD as a robust and cost-effective approach for deploying LLMs in resource-intensive scenarios.

Large language models can solve hard reasoning problems but often need a lot of computing power and time. This paper introduces Reward-Guided Speculative Decoding, which pairs a small, fast “draft” model with the full-strength model so most of the work happens quickly in the draft. A lightweight reward checker then decides when to call in the expensive model to refine only the most promising answers. By doing so, it keeps nearly the same answer quality but uses up to 4.4× less computation, making advanced language AI much faster and cheaper to run in real-world settings.