Paper Detail
Han Wang, Xiaodong Yu, Jialian Wu, Jiang Liu, Ximeng Sun, Mohit Bansal, Zicheng Liu
Large language models (LLMs) achieve strong reasoning performance by allocating substantial computation at inference time, often generating long and verbose reasoning traces. While recent work on efficient reasoning reduces this overhead through length-based rewards or pruning, many approaches are post-trained under a much shorter context window than base-model training, a factor whose effect has not been systematically isolated. We first show that short-context post-training alone, using standard GRPO without any length-aware objective, already induces substantial reasoning compression-but at the cost of increasingly unstable training dynamics and accuracy degradation. To address this, we propose Step-level Advantage Selection (SAS), which operates at the reasoning-step level and assigns a zero advantage to low-confidence steps in correct rollouts and to high-confidence steps in verifier-failed rollouts, where failures often arise from truncation or verifier issues rather than incorrect reasoning. Across diverse mathematical and general reasoning benchmarks, SAS improves average Pass@1 accuracy by 0.86 points over the strongest length-aware baseline while reducing average reasoning length by 16.3%, yielding a better accuracy-efficiency trade-off.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{wang2026stabilizing,
title = {Stabilizing Efficient Reasoning with Step-Level Advantage Selection},
author = {Han Wang and Xiaodong Yu and Jialian Wu and Jiang Liu and Ximeng Sun and Mohit Bansal and Zicheng Liu},
year = {2026},
abstract = {Large language models (LLMs) achieve strong reasoning performance by allocating substantial computation at inference time, often generating long and verbose reasoning traces. While recent work on efficient reasoning reduces this overhead through length-based rewards or pruning, many approaches are post-trained under a much shorter context window than base-model training, a factor whose effect has not been systematically isolated. We first show that short-context post-training alone, using standa},
url = {https://huggingface.co/papers/2604.24003},
keywords = {large language models, reasoning performance, inference time, length-based rewards, pruning, post-training, GRPO, reasoning steps, advantage selection, Pass@1 accuracy, reasoning length, code available, huggingface daily},
eprint = {2604.24003},
archiveprefix = {arXiv},
}
{}