arXiv Computation and Language

Beyond Repeated Sampling: Learning Search Policies for LLM Reasoning

The paper proposes a new approach to large language model (LLM) reasoning that moves beyond naive repeated sampling. Instead of generating many independent solutions, it first samples problem‑specific concepts, hints, or strategies and conditions answer generation on them, producing a single trajectory of diverse concepts. A small concept generator is then trained via reinforcement learning to maximize downstream success, leading to significant improvements in pass@k on hard mathematical reasoning tasks compared to both naive sampling and concepts from larger untuned models, and the trained generator transfers to unseen answer generators, including those from different model families.

arXiv Computation and Language
Aug 28

Boosting LLM Exploration via Weak-Model Guidance in RLVR

The paper introduces a method to enhance large language model (LLM) exploration in Reinforcement Learning with Verifiable Rewards (RLVR) by guiding the target model with partial reasoning trajectories from smaller, weaker language models. This weak-model guidance disrupts over‑confidence, preserves generative diversity, and mitigates entropy collapse without extra fine‑tuning or complex reward designs. Experiments on mathematical benchmarks show consistent improvements over vanilla RLVR, especially as the number of allowed attempts ($k$) increases, indicating broader reasoning coverage.

By Xingyu Shen, Huishuai Zhang, Peng Li, Yinchun Wang, Dongyan Zhao
arXiv AI
Sep 25

To Think or Not to Think: Allocating Reasoning Where It Helps

The paper introduces CARE, a contrastive accuracy reward estimation method that adaptively adjusts reasoning length for large language models. By comparing beneficial length adjustments from online sampled responses, CARE applies adaptive length rewards within Group Relative Policy Optimization without extra hyperparameters or inference cost. Experiments on multiple reasoning benchmarks show that CARE improves Pass@1 by up to 4% while reducing reasoning length by 37%, achieving higher token efficiency.

By Zhengdong He, Yunfan Zhou, Jianguo Yao, Haibing Guan, Xijun Li
arXiv Machine Learning
Oct 8

Decoupling Exploration from Optimization in RLVR

The paper introduces Exploration-Distillation (ExpDis), a method that separates exploration from optimization in reinforcement learning with verifiable rewards (RLVR) for language models. ExpDis trains explorer policies with a novelty bonus, filters their outputs for correctness, and distills the results into a student policy that is optimized without the novelty bonus. This iterative process enables aggressive exploration while preserving model quality, outperforming DAPO on multiple mathematical reasoning benchmarks and improving pass@$k$ scaling.

By Saif Punjwani, Micah Goldblum
Hugging Face Trending Papers
Jul 30

LEEPS: Latent-Guided Explore-Exploit Prompt Sampling for Efficient RLVR in Large Language Models

Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities of large language models, but prompt groups with identical rollout rewards consume generation budget without effective learning signals. Pre-rollout prompt selection can reduce this waste by screening prompts before rollout generation.

arXiv AI
Jul 16

Representation-Based Exploration for Language Models: From Test-Time to Post-Training

arXiv:2510. 11686v2 Announce Type: replace-cross Abstract: Reinforcement learning (RL) promises to expand the capabilities of language models, but it is unclear if current RL techniques promote the discovery of novel behaviors, or simply sharpen those already present in the base model.

By Jens Tuyls, Dylan J. Foster, Akshay Krishnamurthy, Jordan T. Ash
arXiv Machine Learning
Oct 22

ReAPO: Retry-Aware Policy Optimization for Efficient Exploration in LLM Reasoning

ReAPO: Retry-Aware Policy Optimization for Efficient Exploration in LLM Reasoning proposes a new approach to improve language‑model reasoning by addressing the inefficiencies of independent stochastic rollouts. By framing credit assignment as a Stochastic Shortest Path problem, ReAPO modifies GRPO to allocate more generation to failed trajectories, reinforcing successful retries without rewarding failures. Experiments on Qwen3 and OLMo‑2 models across eight mathematical and scientific benchmarks show that ReAPO outperforms GRPO and other baselines, achieving higher first‑attempt accuracy and reducing generated tokens needed to reach peak performance.

By Nghi Dao, Calder Morton-Ferguson, Noah Golowich, Adam Block