Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models
This addresses the deployment cost issue for users of large reasoning models by improving efficiency without sacrificing performance, though it is incremental as it builds on existing reinforcement learning and reasoning paradigms.
The paper tackles the problem of excessive verbosity in large reasoning models trained with reinforcement learning, which increases deployment costs, by introducing Dynamic Outlier Truncation (DOT) to suppress redundant tokens during training. The result is a 78% reduction in inference token usage on AIME-24 while increasing accuracy compared to the initial policy and outperforming state-of-the-art efficient reasoning methods.
Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivial inputs during training. To address this, we introduce Dynamic Outlier Truncation (DOT), a training-time intervention that selectively suppresses redundant tokens. This method targets only the extreme tail of response lengths within fully correct rollout groups while preserving long-horizon reasoning capabilities for complex problems. To complement this intervention and ensure stable convergence, we further incorporate auxiliary KL regularization and predictive dynamic sampling. Experimental results across multiple model scales demonstrate that our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.