ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inference
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912895063818240 |
|---|---|
| author | Wang, Junda Yang, Zhichao Zhang, Dongxu Batra, Sanjit Singh Tillman, Robert E. |
| author_facet | Wang, Junda Yang, Zhichao Zhang, Dongxu Batra, Sanjit Singh Tillman, Robert E. |
| contents | Large reasoning models (LRMs) achieve state-of-the-art performance by generating long chains-of-thought, but often waste computation on redundant reasoning after the correct answer has already been reached. We introduce Early-Stopping for Token-Aware Reasoning (ESTAR), which detects and reduces such reasoning redundancy to improve efficiency without sacrificing accuracy. Our method combines (i) a trajectory-based classifier that identifies when reasoning can be safely stopped, (ii) supervised fine-tuning to teach LRMs to propose self-generated <stop> signals, and (iii) <stop>-aware reinforcement learning that truncates rollouts at self-generated stop points with compute-aware rewards. Experiments on four reasoning datasets show that ESTAR reduces reasoning length by about 3.7x (from 4,799 to 1,290) while preserving accuracy (74.9% vs. 74.2%), with strong cross-domain generalization. These results highlight early stopping as a simple yet powerful mechanism for improving reasoning efficiency in LRMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_10004 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inference Wang, Junda Yang, Zhichao Zhang, Dongxu Batra, Sanjit Singh Tillman, Robert E. Artificial Intelligence Large reasoning models (LRMs) achieve state-of-the-art performance by generating long chains-of-thought, but often waste computation on redundant reasoning after the correct answer has already been reached. We introduce Early-Stopping for Token-Aware Reasoning (ESTAR), which detects and reduces such reasoning redundancy to improve efficiency without sacrificing accuracy. Our method combines (i) a trajectory-based classifier that identifies when reasoning can be safely stopped, (ii) supervised fine-tuning to teach LRMs to propose self-generated <stop> signals, and (iii) <stop>-aware reinforcement learning that truncates rollouts at self-generated stop points with compute-aware rewards. Experiments on four reasoning datasets show that ESTAR reduces reasoning length by about 3.7x (from 4,799 to 1,290) while preserving accuracy (74.9% vs. 74.2%), with strong cross-domain generalization. These results highlight early stopping as a simple yet powerful mechanism for improving reasoning efficiency in LRMs. |
| title | ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inference |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2602.10004 |