Nonsense Helps: Prompt Space Perturbation Broadens Reasoning Exploration

Hugging Face Daily Papers Papers

Summary

This paper introduces LoPE, a training framework that uses prompt-space perturbations to address the zero-advantage problem in reinforcement learning with verifiable rewards, thereby enhancing reasoning exploration in large language models.

Reinforcement learning with verifiable rewards, particularly Group Relative Policy Optimization (GRPO), has significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, in complex tasks, GRPO frequently suffers from the ``zero-advantage problem'': when all sampled rollouts for a query fail, the relative advantage collapses to zero. Consequently, the model loses effective training signals for these questions, wasting the training data and computational budget. While simply increasing the sampling budget for these questions is a common remedy, the static sampling policy inherently constrains reasoning exploration, limiting the success rate. In this paper, we propose Lorem Perturbation for Exploration (LoPE), a simple yet effective training framework to break this exploration bottleneck. We posit that task-irrelevant prompt-space perturbations can shift the model's output distribution enough to unlock orthogonal reasoning pathways for hard questions. Specifically, LoPE prepends sequences stochastically assembled from Lorem Ipsum vocabulary (a pseudo-Latin placeholder text) to the prompts before resampling. Experiments across 1.7B, 4B, and 7B models demonstrate that LoPE significantly outperforms resampling with the original prompts. Further analysis reveals that other Latin-based random sequences with low perplexity are also effective perturbations. Our results establish LoPE as a strong baseline for broadening exploration in LLM reinforcement learning.
Original Article
View Cached Full Text

Cached at: 05/08/26, 07:15 AM

Paper page - Nonsense Helps: Prompt Space Perturbation Broadens Reasoning Exploration

Source: https://huggingface.co/papers/2605.05566

Abstract

LoPE addresses the zero-advantage problem in reinforcement learning with verifiable rewards by usingLorem Ipsum perturbations to enhance exploration in large language model training.

Reinforcement learningwith verifiable rewards, particularlyGroup Relative Policy Optimization(GRPO), has significantly advanced the reasoning capabilities ofLarge Language Models(LLMs). However, in complex tasks, GRPO frequently suffers from the ``zero-advantage problem’’: when all sampled rollouts for a query fail, the relative advantage collapses to zero. Consequently, the model loses effective training signals for these questions, wasting the training data and computational budget. While simply increasing the sampling budget for these questions is a common remedy, the static sampling policy inherently constrains reasoning exploration, limiting the success rate. In this paper, we propose Lorem Perturbation for Exploration (LoPE), a simple yet effective training framework to break this exploration bottleneck. We posit that task-irrelevantprompt-space perturbationscan shift the model’s output distribution enough to unlock orthogonal reasoning pathways for hard questions. Specifically, LoPE prepends sequences stochastically assembled fromLorem Ipsumvocabulary (a pseudo-Latin placeholder text) to the prompts beforeresampling. Experiments across 1.7B, 4B, and 7B models demonstrate that LoPE significantly outperformsresamplingwith the original prompts. Further analysis reveals that other Latin-based random sequences with lowperplexityare also effective perturbations. Our results establish LoPE as a strong baseline for broadening exploration in LLMreinforcement learning.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2605\.05566

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/2605.05566 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.05566 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.05566 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

Prompt-Driven Exploration

arXiv cs.LG

The paper introduces Prompt-Driven Exploration (PDE), a method that uses a vision-language model to iteratively refine natural language prompts for reinforcement learning policies, enabling global exploration and successful policy learning even from zero-reward starts.

SPS: Steering Probability Squeezing for Better Exploration in Reinforcement Learning for Large Language Models

arXiv cs.CL

Researchers propose SPS (Steering Probability Squeezing), a training paradigm combining reinforcement learning with inverse reinforcement learning to address probability squeezing in LLM reasoning training, where probability mass concentrates too narrowly on high-reward trajectories, limiting exploration and multi-sample performance (Pass@k). Experiments on five reasoning benchmarks demonstrate improved exploration and Pass@k metrics.