TRiMS: Real-Time Tracking of Minimal Sufficient Length for Efficient Reasoning via RL
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866918395325186048 |
|---|---|
| author | Bian, Tingcheng Luo, Jinchang Cheng, Mingquan Zhang, Jinyu Xia, Xiaoling Li, Ni Tao, Yan Wang, Haiwei |
| author_facet | Bian, Tingcheng Luo, Jinchang Cheng, Mingquan Zhang, Jinyu Xia, Xiaoling Li, Ni Tao, Yan Wang, Haiwei |
| contents | Large language models achieve breakthroughs in complex reasoning via long chain-of-thought sequences. However, this often leads to severe reasoning inflation, causing substantial computational redundancy. To maximize Intelligence per Token, we introduce a theoretical metric, MSL-Minimal Sufficient Length. MSL rigorously characterizes the shortest reasoning length that preserves answer correctness. We provide a recursive definition based on independently sampled sequences and prove the existence of its limit, establishing the first measurable lower bound for reasoning-chain compression. Building on an analysis of mainstream CoT compression strategies, we identify key structural factors enabling a model to approach MSL. Based on these insights, we propose TRiMS which employs the GRPO algorithm in conjunction with MSL-based estimation during training, while mitigating instabilities during the training process through dynamic batch aggregation and advantage computation using batch-level standard deviation. TRiMS achieves over 80% CoT token reduction with a minor accuracy boost across all benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_17449 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | TRiMS: Real-Time Tracking of Minimal Sufficient Length for Efficient Reasoning via RL Bian, Tingcheng Luo, Jinchang Cheng, Mingquan Zhang, Jinyu Xia, Xiaoling Li, Ni Tao, Yan Wang, Haiwei Computation and Language I.2.7; I.2.6 Large language models achieve breakthroughs in complex reasoning via long chain-of-thought sequences. However, this often leads to severe reasoning inflation, causing substantial computational redundancy. To maximize Intelligence per Token, we introduce a theoretical metric, MSL-Minimal Sufficient Length. MSL rigorously characterizes the shortest reasoning length that preserves answer correctness. We provide a recursive definition based on independently sampled sequences and prove the existence of its limit, establishing the first measurable lower bound for reasoning-chain compression. Building on an analysis of mainstream CoT compression strategies, we identify key structural factors enabling a model to approach MSL. Based on these insights, we propose TRiMS which employs the GRPO algorithm in conjunction with MSL-based estimation during training, while mitigating instabilities during the training process through dynamic batch aggregation and advantage computation using batch-level standard deviation. TRiMS achieves over 80% CoT token reduction with a minor accuracy boost across all benchmarks. |
| title | TRiMS: Real-Time Tracking of Minimal Sufficient Length for Efficient Reasoning via RL |
| topic | Computation and Language I.2.7; I.2.6 |
| url | https://arxiv.org/abs/2603.17449 |