EUVIMEDEUVIMED entdecken
Crossref

SABER: Switchable and Balanced Training for Efficient LLM Reasoning

Kai Zhao, Yanjun Zhao, Jiaming Song, Shien He, Lusheng Zhang, Qiang Zhang, Tianjiao Li

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (41) · 34950-34958

DOI: 10.1609/aaai.v40i41.40799 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Kai Zhao und 6 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large language models (LLMs) empowered by chain-of-thought reasoning have achieved impressive accuracy on complex tasks but suffer from excessive inference costs and latency when applied uniformly to all problems. We propose SABER (Switchable and Balanced Training for Efficient LLM Reasoning), a reinforcement learning framework that endows LLMs with user‑controllable, token‑budgeted reasoning. SABER first profiles each training example’s base‑model thinking token usage and assigns it to one of the predefined budget tiers. During fine‑tuning, the model is guided by system prompts and length‑aware rewards to respect its assigned budget. In parallel, we incorporate no‑think examples to ensure the model remains reliable even when explicit reasoning is turned off. SABER further supports four discrete inference modes—NoThink, FastThink, CoreThink, and DeepThink, enabling flexible trade‑offs between latency and reasoning depth. Extensive evaluations on math reasoning (MATH, GSM8K), code generation (MBPP), and logical reasoning (LiveBench-Reasoning) demonstrate that SABER achieves high accuracy under tight budgets, graceful degradation, and effective cross-scale and cross‑domain generalization. In particular, SABER‑FastThink cuts reasoning length by 65.4% and yields a 3.6% accuracy gain compared with the base model on the MATH benchmark.

Zitieren

APA

Zhao, K., Zhao, Y., Song, J., He, S., Zhang, L., Zhang, Q., & Li, T. (2026). SABER: Switchable and Balanced Training for Efficient LLM Reasoning. Proceedings of the AAAI Conference on Artificial Intelligence, 40(41), 34950-34958. https://doi.org/10.1609/aaai.v40i41.40799

Vancouver

Zhao K, Zhao Y, Song J, He S, Zhang L, Zhang Q, et al. SABER: Switchable and Balanced Training for Efficient LLM Reasoning. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(41):34950-34958. doi:10.1609/aaai.v40i41.40799

BibTeX
@article{Zhao2026,
  author = {Zhao, Kai and Zhao, Yanjun and Song, Jiaming and He, Shien and Zhang, Lusheng and Zhang, Qiang and Li, Tianjiao},
  title = {SABER: Switchable and Balanced Training for Efficient LLM Reasoning},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {41},
  pages = {34950--34958},
  doi = {10.1609/aaai.v40i41.40799}
}
Kostenlos starten →