Paper Detail
Prasanth YSS, Zhichen Ren, Rasa Hosseinzadeh, Ilan Gofman, Yuqi Chen, Zhaoyan Liu, Guangwei Yu, Jesse C. Cresswell, Satya Krishna Gorti
Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning, but GRPO-style optimization remains prone to collapse. We analyse this instability through token-level gradient dynamics, deriving a taxonomy that predicts how updates affect next-token probabilities and entropy. The taxonomy shows that stability depends jointly on the advantage sign and token distribution under the current policy. Motivated by this finding, we propose Winner Advantage Policy Optimization (WAPO), a simple online clipped policy-gradient objective that updates only on positive-advantage completions. Across mathematical reasoning and multi-hop QA benchmarks, WAPO improves training stability and matches or outperforms baselines across multiple model families. Full code can be found at https://github.com/layer6ai-labs/wapo.
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{yss2026gradient,
title = {A Gradient Perspective on RLVR Stability and Winner Advantage Policy Optimization},
author = {Prasanth YSS and Zhichen Ren and Rasa Hosseinzadeh and Ilan Gofman and Yuqi Chen and Zhaoyan Liu and Guangwei Yu and Jesse C. Cresswell and Satya Krishna Gorti},
year = {2026},
abstract = {Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning, but GRPO-style optimization remains prone to collapse. We analyse this instability through token-level gradient dynamics, deriving a taxonomy that predicts how updates affect next-token probabilities and entropy. The taxonomy shows that stability depends jointly on the advantage sign and token distribution under the current policy. Motivated by this finding, we propose Winner Advantage Policy Optimization (W},
url = {https://huggingface.co/papers/2606.16154},
keywords = {reinforcement learning, verifiable rewards, language-model reasoning, GRPO-style optimization, token-level gradient dynamics, advantage sign, token distribution, policy-gradient objective, positive-advantage completions, code available, huggingface daily},
eprint = {2606.16154},
archiveprefix = {arXiv},
}
{}