Addressing Performance Saturation for LLM RL via Precise Entropy Curve Control

Hugging Face Daily Papers Papers

Summary

This paper introduces Entrocraft, a rejection-sampling method for RL that controls entropy schedules to prevent performance saturation in LLMs. It demonstrates improved generalization and training longevity, allowing smaller models to outperform larger baselines.

Reinforcement learning (RL) has enabled complex reasoning abilities in large language models (LLMs). However, most RL algorithms suffer from performance saturation, preventing continued gains as RL training scales. This problem can be characterized by the collapse of entropy, a key diagnostic for exploration in RL. Existing attempts focus on preventing entropy collapse through regularization or clipping. However, their resulting entropy curves often exhibit instability in the long term, which hinders performance gains. In this paper, we introduce Entrocraft, a simple rejection-sampling approach that realizes user-customized entropy schedule by biasing the advantage distributions. Entrocraft requires no objective regularization and is advantage-estimator-agnostic. Theoretically, we relate per-step entropy change to the advantage distribution under minimal assumptions. This explains the behavior of existing RL and entropy-preserving methods. Entrocraft also enables a systematic study of entropy schedules, which reveals that linear annealing, which starts high and decays to a slightly lower target, performs best. Empirically, Entrocraft addresses performance saturation, significantly improving generalization, output diversity, and long-term training. It enables a 4B model to outperform an 8B baseline, sustains improvement for up to 4x longer before plateauing, and raises pass@K by 50% over the baseline.
Original Article
View Cached Full Text

Cached at: 05/13/26, 12:20 AM

Paper page - Addressing Performance Saturation for LLM RL via Precise Entropy Curve Control

Source: https://huggingface.co/papers/2604.26326

Abstract

Entrocraft, a rejection-sampling approach for reinforcement learning, addresses performance saturation in large language models by customizing entropy schedules and improving generalization and training longevity.

Reinforcement learning(RL) has enabled complex reasoning abilities in large language models (LLMs). However, most RL algorithms suffer fromperformance saturation, preventing continued gains as RL training scales. This problem can be characterized by the collapse of entropy, a key diagnostic for exploration in RL. Existing attempts focus on preventingentropy collapsethrough regularization or clipping. However, their resulting entropy curves often exhibit instability in the long term, which hinders performance gains. In this paper, we introduce Entrocraft, a simplerejection-samplingapproach that realizes user-customizedentropy scheduleby biasing theadvantage distributions. Entrocraft requires no objective regularization and isadvantage-estimator-agnostic. Theoretically, we relate per-step entropy change to the advantage distribution under minimal assumptions. This explains the behavior of existing RL and entropy-preserving methods. Entrocraft also enables a systematic study ofentropy schedules, which reveals that linear annealing, which starts high and decays to a slightly lower target, performs best. Empirically, Entrocraft addressesperformance saturation, significantly improvinggeneralization,output diversity, and long-term training. It enables a 4B model to outperform an 8B baseline, sustains improvement for up to 4x longer before plateauing, and raises pass@K by 50% over the baseline.

View arXiv pageView PDFProject pageGitHub2Add to collection

Get this paper in your agent:

hf papers read 2604\.26326

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/2604.26326 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2604.26326 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2604.26326 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

Revisiting Entropy Regularization: Adaptive Coefficient Unlocks Its Potential for LLM Reinforcement Learning

arXiv cs.CL

This paper proposes Adaptive Entropy Regularization (AER), a framework that dynamically balances exploration and exploitation in LLM reinforcement learning by addressing policy entropy collapse through difficulty-aware coefficient allocation and initial-anchored target entropy. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements in both accuracy and exploration capability.