N-GRPO: Embedding-Level Neighbor Mixing for Enhanced Policy Optimization

Hugging Face Daily Papers Papers

Summary

N-GRPO introduces semantic neighbor mixing in the GRPO framework to enhance mathematical reasoning diversity while preserving semantic consistency, achieving improvements on math benchmarks and out-of-distribution tasks.

The success of Large Language Models in mathematical reasoning relies heavily on the generation of diverse and valid solution paths during the rollout phase. However, current rollout techniques face a fundamental trade-off: token-level sampling often yields redundant trajectories that differ only in rephrasing, while embedding-level methods utilizing random noise frequently disrupt semantic consistency. To resolve this, we introduce N-GRPO, a novel exploration strategy integrated into the Group Relative Policy Optimization (GRPO) framework. Rather than relying on token-level sampling or native embedding-level noise, our approach leverages Semantic Neighbor Mixing. This mechanism dynamically constructs input representations by mixing the embeddings of an anchor token and its nearest semantic neighbors, thereby injecting diversity while strictly adhering to the local semantic manifold. Experimental evaluations on the DeepSeek-R1-Distill-Qwen models across different sizes show that N-GRPO not only achieves consistent improvements over strong baselines on math reasoning benchmarks but also exhibits robust generalization capabilities on out-of-distribution tasks.
Original Article
View Cached Full Text

Cached at: 06/12/26, 02:52 AM

Paper page - N-GRPO: Embedding-Level Neighbor Mixing for Enhanced Policy Optimization

Source: https://huggingface.co/papers/2606.10768

Abstract

N-GRPO, a novel exploration strategy within GRPO framework, enhances mathematical reasoning in large language models through semantic neighbor mixing that maintains semantic consistency while injecting diversity.

The success ofLarge Language Modelsinmathematical reasoningrelies heavily on the generation of diverse and valid solution paths during therollout phase. However, current rollout techniques face a fundamental trade-off:token-level samplingoften yields redundant trajectories that differ only in rephrasing, whileembedding-level methodsutilizing random noise frequently disrupt semantic consistency. To resolve this, we introduce N-GRPO, a novel exploration strategy integrated into theGroup Relative Policy Optimization(GRPO) framework. Rather than relying ontoken-level samplingor native embedding-level noise, our approach leveragesSemantic Neighbor Mixing. This mechanism dynamically constructs input representations by mixing the embeddings of an anchor token and its nearest semantic neighbors, thereby injecting diversity while strictly adhering to the localsemantic manifold. Experimental evaluations on the DeepSeek-R1-Distill-Qwen models across different sizes show that N-GRPO not only achieves consistent improvements over strong baselines onmath reasoning benchmarksbut also exhibits robust generalization capabilities onout-of-distribution tasks.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2606\.10768

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.10768 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.10768 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.10768 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Group Entropy-Controlled Policy Optimization

Hugging Face Daily Papers

This paper proposes Group Entropy-Controlled Policy Optimization (GEPO), a lightweight extension to GRPO that uses group entropy to perform entropy-conditioned asymmetric advantage shaping, addressing heterogeneous entropy regimes across tasks during RL-based alignment of LLMs. Experiments show consistent improvements over GRPO and recent entropy-controlled methods across multiple benchmarks.