SpecExit: Accelerating Large Reasoning Model via Speculative Exit
This addresses a key limitation for deploying large reasoning models in real-world applications by improving efficiency, though it is an incremental advance over existing early-exit and speculative decoding methods.
The paper tackles the problem of overthinking in large reasoning models, which causes unnecessarily long outputs and high latency, by proposing SpecExit, a framework that reduces average generation length by 66% and achieves a 2.5x speedup in end-to-end latency without compromising accuracy.
Despite their strong performance on reasoning tasks, large reasoning models (LRMs) often suffer from overthinking, producing unnecessarily long outputs and incurring high end-to-end latency, a significant limitation to their real-world deployment. To address overthinking, early-exit mechanisms have been proposed to terminate reasoning before typical completion, showing that this approach can effectively shorten generation length with minimal impact on accuracy. However, their reliance on probing mechanisms introduces a detection overhead that limits their end-to-end latency gains and compromises their generalizability across diverse problems. Inspired by the use of hidden states in speculative decoding, we propose SpecExit, a novel framework that predicts both future tokens and an early-exit signal directly from a lightweight draft model without probing overhead. Our method offers significant improvements, reducing average generation length by 66\% and achieving a 2.5x speedup in end-to-end latency compared to the speculative decoding baseline, without compromising accuracy. Our method leverages the inherent signals from hidden states to provide effective early-exit signals, suggesting broader use of hidden states for efficient reasoning. Our code is available at https://github.com/Tencent/AngelSlim.