MarkovScale: Towards Optimal Sequential Scaling at Inference Time
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908802615345152 |
|---|---|
| author | Wang, Youkang Wang, Jian Chen, Rubing Zeng, Tianyi Wei, Xiao-Yong Li, Qing |
| author_facet | Wang, Youkang Wang, Jian Chen, Rubing Zeng, Tianyi Wei, Xiao-Yong Li, Qing |
| contents | Sequential scaling is a prominent inference-time scaling paradigm, yet its performance improvements are typically modest and not well understood, largely due to the prevalence of heuristic, non-principled approaches that obscure clear optimality bounds. To address this, we propose a principled framework that models sequential scaling as a two-state Markov process. This approach reveals the underlying properties of sequential scaling and yields closed-form solutions for essential aspects, such as the specific conditions under which accuracy is improved and the theoretical upper, neutral, and lower performance bounds. Leveraging this formulation, we develop MarkovScale, a practical system that applies these optimality criteria to achieve a theoretically grounded balance between accuracy and efficiency. Comprehensive experiments across 3 backbone LLMs, 5 benchmarks, and over 20 configurations show that MarkovScale consistently outperforms state-of-the-art parallel and sequential scaling methods, representing a significant step toward optimal and resource-efficient inference in LLMs. The source code will be open upon acceptance at https://open-upon-acceptance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_01120 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | MarkovScale: Towards Optimal Sequential Scaling at Inference Time Wang, Youkang Wang, Jian Chen, Rubing Zeng, Tianyi Wei, Xiao-Yong Li, Qing Machine Learning Artificial Intelligence Computation and Language Sequential scaling is a prominent inference-time scaling paradigm, yet its performance improvements are typically modest and not well understood, largely due to the prevalence of heuristic, non-principled approaches that obscure clear optimality bounds. To address this, we propose a principled framework that models sequential scaling as a two-state Markov process. This approach reveals the underlying properties of sequential scaling and yields closed-form solutions for essential aspects, such as the specific conditions under which accuracy is improved and the theoretical upper, neutral, and lower performance bounds. Leveraging this formulation, we develop MarkovScale, a practical system that applies these optimality criteria to achieve a theoretically grounded balance between accuracy and efficiency. Comprehensive experiments across 3 backbone LLMs, 5 benchmarks, and over 20 configurations show that MarkovScale consistently outperforms state-of-the-art parallel and sequential scaling methods, representing a significant step toward optimal and resource-efficient inference in LLMs. The source code will be open upon acceptance at https://open-upon-acceptance. |
| title | MarkovScale: Towards Optimal Sequential Scaling at Inference Time |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2602.01120 |