ElasticMem enables LLM agents to learn adaptive latent memory retrieval and elastic budget allocation, improving QA accuracy by 24-26% and ALFWorld success by 27-66% over baselines with lower token cost.
hub Canonical reference
The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models
Canonical reference. 80% of citing Pith papers cite this work as background.
80
Pith papers citing it
Background
80% of classified citations
abstract
This paper aims to overcome a major obstacle in scaling RL for reasoning with LLMs, namely the collapse of policy entropy. Such phenomenon is consistently observed across vast RL runs without entropy intervention, where the policy entropy dropped sharply at the early training stage, this diminished exploratory ability is always accompanied with the saturation of policy performance. In practice, we establish a transformation equation R=-a*e^H+b between entropy H and downstream performance R. This empirical law strongly indicates that, the policy performance is traded from policy entropy, thus bottlenecked by its exhaustion, and the ceiling is fully predictable H=0, R=-a+b. Our finding necessitates entropy management for continuous exploration toward scaling compute for RL. To this end, we investigate entropy dynamics both theoretically and empirically. Our derivation highlights that, the change in policy entropy is driven by the covariance between action probability and the change in logits, which is proportional to its advantage when using Policy Gradient-like algorithms. Empirical study shows that, the values of covariance term and entropy differences matched exactly, supporting the theoretical conclusion. Moreover, the covariance term stays mostly positive throughout training, further explaining why policy entropy would decrease monotonically. Through understanding the mechanism behind entropy dynamics, we motivate to control entropy by restricting the update of high-covariance tokens. Specifically, we propose two simple yet effective techniques, namely Clip-Cov and KL-Cov, which clip and apply KL penalty to tokens with high covariances respectively. Experiments show that these methods encourage exploration, thus helping policy escape entropy collapse and achieve better downstream performance.
hub tools
citation-role summary
background 23
baseline 1
method 1
citation-polarity summary
claims ledger
- abstract This paper aims to overcome a major obstacle in scaling RL for reasoning with LLMs, namely the collapse of policy entropy. Such phenomenon is consistently observed across vast RL runs without entropy intervention, where the policy entropy dropped sharply at the early training stage, this diminished exploratory ability is always accompanied with the saturation of policy performance. In practice, we establish a transformation equation R=-a*e^H+b between entropy H and downstream performance R. This empirical law strongly indicates that, the policy performance is traded from policy entropy, thus b
- background [145] Yingqian Cui, Pengfei He, Jingying Zeng, Hui Liu, Xianfeng Tang, Zhenwei Dai, Yan Han, Chen Luo, Jing Huang, Zhen Li, et al. Stepwise perplexity-guided refinement for efficient chain-of-thought reasoning in large language models. arXiv preprint arXiv:2502.13260, 2025. [146] Yu Cui and Cong Zuo. Practical reasoning interruption attacks on reasoning large language models. arXiv preprint arXiv:2505.06643, 2025. [147] Yu Cui, Bryan Hooi, Yujun Cai, and Yiwei Wang. Process or result? manipulate
co-cited works
representative citing papers
Extrapolative weight averaging of RL checkpoints trained under nested unit-test coverage extends a correctness-efficiency frontier and boosts ensemble pass rates in code generation across model scales and inference modes.
Distribution-Aware Reward optimizes LLM regression by treating rollouts as empirical predictive distributions and rewarding marginal improvements in CRPS quality rather than point accuracy alone.
PPR-GDE is a new RL approach that integrates pairwise preference rewards with group-based diversity enhancement in a unified objective to improve both alignment quality and expressive diversity in open-ended generation tasks such as role-playing.
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
DISA decouples partition function estimation using offline importance sampling for distribution-matching LLM-RL, matching or exceeding online baselines like FlowRL on math and code benchmarks while retaining more strategy diversity.
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
GCPO uses team-level credit assignment via determinant volume over reward-weighted semantic embeddings to promote non-redundant correct reasoning paths, improving both accuracy and diversity in LLM training.
RLRT augments GRPO by reinforcing tokens on correct student rollouts that the teacher would not have predicted, outperforming standard self-distillation and exploration baselines on Qwen3 models.
The cancellation hypothesis shows how rollout-level rewards produce token-level credit assignment in critic-free RL through cancellation of opposing signals on shared tokens, with empirical support and batching interventions that enhance performance.
VaP-CSMV uses a cross-semantic encoder and multi-view decoder to unify DRL solving of HFVRP variants, outperforming prior neural solvers while matching heuristics at much lower inference time and generalizing zero-shot to unseen scales.
SDRL trains LLMs via self-generated multi-path debates and joint optimization of standalone plus debate-conditioned responses to boost both single-model reasoning and multi-agent debate performance.
A 400k+ GPU-hour study shows RL scaling in LLMs follows predictable sigmoidal trajectories, with most design choices affecting efficiency rather than the performance asymptote, enabling accurate large-scale predictions via the ScaleRL recipe.
Introduces a hierarchical latent selection model showing SFT supplies raw module materials in compound traces while RL decomposes them to identify atomic modules and enable recombination for new reasoning configurations.
CPT is introduced as a pairwise reasoning-trace comparison stage that improves the reasoning-metacognition trade-off over standard SFT+RL pipelines across model scales.
ExpGraph builds a graph of summarized agent experiences and uses graph diffusion plus an RL-trained retrieval copilot to improve frozen LLM executors on QA, math, code, and agentic tasks without parameter updates.
NFPO augments the PPO surrogate with N-step forward traces to bridge local approximations and exact policy gradients, delivering tighter policy-improvement bounds and improved results on reasoning benchmarks.
DMPO approximates forward KL minimization in on-policy RL by aligning the policy to a group-level reward-proportional target distribution, yielding 9-12% relative gains over GRPO on NP-Bench and smaller gains on math reasoning.
CES applies conditional bidirectional entropy control on top of DAPO to improve accuracy and shorten responses on mathematical benchmarks for 7B and 1.5B LLMs.
ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points.
Policy entropy remains constant in flow-matching models during RLHF due to fixed noise schedules while perceptual diversity collapses from mode-seeking policy gradients, so perceptual entropy constraints are introduced to preserve diversity and improve quality.
Entropy polarity is a signed token-level quantity derived from a first-order approximation of entropy change that predicts whether RL updates expand or contract policy entropy in LLM fine-tuning, revealing an asymmetry between high- and low-probability tokens.
Covariance-weighted GRPO with Gaussian-kernel reweighting tames extreme tokens to stabilize training and boost reasoning performance over standard GRPO.
OPEFO prevents entropy collapse in RLVR by rescaling token updates according to their entropy change contributions, yielding more stable optimization and better results on math benchmarks.
citing papers explorer
-
ElasticMem: Latent Memory as a Learnable Resource for LLM Agents
ElasticMem enables LLM agents to learn adaptive latent memory retrieval and elastic budget allocation, improving QA accuracy by 24-26% and ALFWorld success by 27-66% over baselines with lower token cost.
-
Extrapolative Weight Averaging Reveals Correctness-Efficiency Frontiers in Code RL
Extrapolative weight averaging of RL checkpoints trained under nested unit-test coverage extends a correctness-efficiency frontier and boosts ensemble pass rates in code generation across model scales and inference modes.
-
Distribution-Aware Reward: Reinforcement Learning over Predictive Distributions for LLM Regression
Distribution-Aware Reward optimizes LLM regression by treating rollouts as empirical predictive distributions and rewarding marginal improvements in CRPS quality rather than point accuracy alone.
-
Pairwise Preference Reward and Group-Based Diversity Enhancement for Superior Open-Ended Generation
PPR-GDE is a new RL approach that integrates pairwise preference rewards with group-based diversity enhancement in a unified objective to improve both alignment quality and expressive diversity in open-ended generation tasks such as role-playing.
-
Reasoning Portability: Guiding Continual Learning for MLLMs in the RLVR Era
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
-
DISA: Offline Importance Sampling for Distribution-Matching LLM-RL
DISA decouples partition function estimation using offline importance sampling for distribution-matching LLM-RL, matching or exceeding online baselines like FlowRL on math and code benchmarks while retaining more strategy diversity.
-
Learning, Fast and Slow: Towards LLMs That Adapt Continually
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
-
Breaking $\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning
GCPO uses team-level credit assignment via determinant volume over reward-weighted semantic embeddings to promote non-redundant correct reasoning paths, improving both accuracy and diversity in LLM training.
-
Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR
RLRT augments GRPO by reinforcing tokens on correct student rollouts that the teacher would not have predicted, outperforming standard self-distillation and exploration baselines on Qwen3 models.
-
The Cancellation Hypothesis in Critic-Free RL: From Outcome Rewards to Token Credits
The cancellation hypothesis shows how rollout-level rewards produce token-level credit assignment in critic-free RL through cancellation of opposing signals on shared tokens, with empirical support and batching interventions that enhance performance.
-
Vehicle-as-Prompt: A Unified Deep Reinforcement Learning Framework for Heterogeneous Fleet Vehicle Routing Problem
VaP-CSMV uses a cross-semantic encoder and multi-view decoder to unify DRL solving of HFVRP variants, outperforming prior neural solvers while matching heuristics at much lower inference time and generalizing zero-shot to unseen scales.
-
Learning from Self-Debate: Preparing Reasoning Models for Multi-Agent Debate
SDRL trains LLMs via self-generated multi-path debates and joint optimization of standalone plus debate-conditioned responses to boost both single-model reasoning and multi-agent debate performance.
-
The Art of Scaling Reinforcement Learning Compute for LLMs
A 400k+ GPU-hour study shows RL scaling in LLMs follows predictable sigmoidal trajectories, with most design choices affecting efficiency rather than the performance asymptote, enabling accurate large-scale predictions via the ScaleRL recipe.
-
From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning
Introduces a hierarchical latent selection model showing SFT supplies raw module materials in compound traces while RL decomposes them to identify atomic modules and enable recombination for new reasoning configurations.
-
Enhancing LLM Metacognition via Cognitive Pairwise Training
CPT is introduced as a pairwise reasoning-trace comparison stage that improves the reasoning-metacognition trade-off over standard SFT+RL pipelines across model scales.
-
ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents
ExpGraph builds a graph of summarized agent experiences and uses graph diffusion plus an RL-trained retrieval copilot to improve frozen LLM executors on QA, math, code, and agentic tasks without parameter updates.
-
Multi-Step Likelihood-Ratio Correction for Reinforcement Learning with Verifiable Rewards
NFPO augments the PPO surrogate with N-step forward traces to bridge local approximations and exact policy gradients, delivering tighter policy-improvement bounds and improved results on reasoning benchmarks.
-
Beyond Mode Collapse: Distribution Matching for Diverse Reasoning
DMPO approximates forward KL minimization in on-policy RL by aligning the policy to a group-level reward-proportional target distribution, yielding 9-12% relative gains over GRPO on NP-Bench and smaller gains on math reasoning.
-
Taming the Thinker: Conditional Entropy Shaping for Adaptive LLM Reasoning
CES applies conditional bidirectional entropy control on top of DAPO to improve accuracy and shorten responses on mathematical benchmarks for 7B and 1.5B LLMs.
-
Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy
ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points.
-
When Policy Entropy Constraint Fails: Preserving Diversity in Flow-based RLHF via Perceptual Entropy
Policy entropy remains constant in flow-matching models during RLHF due to fixed noise schedules while perceptual diversity collapses from mode-seeking policy gradients, so perceptual entropy constraints are introduced to preserve diversity and improve quality.
-
Entropy Polarity in Reinforcement Fine-Tuning: Direction, Asymmetry, and Control
Entropy polarity is a signed token-level quantity derived from a first-order approximation of entropy change that predicts whether RL updates expand or contract policy entropy in LLM fine-tuning, revealing an asymmetry between high- and low-probability tokens.
-
Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting
Covariance-weighted GRPO with Gaussian-kernel reweighting tames extreme tokens to stabilize training and boost reasoning performance over standard GRPO.
-
Understanding and Preventing Entropy Collapse in RLVR with On-Policy Entropy Flow Optimization
OPEFO prevents entropy collapse in RLVR by rescaling token updates according to their entropy change contributions, yielding more stable optimization and better results on math benchmarks.
-
HTPO: Towards Exploration-Exploitation Balanced Policy Optimization via Hierarchical Token-level Objective Control
HTPO introduces hierarchical token-level objective control in RLVR to balance exploration and exploitation by grouping tokens according to difficulty, correctness, and entropy, yielding up to 8.6% gains on AIME benchmarks over DAPO.
-
Gradient Extrapolation-Based Policy Optimization
GXPO approximates longer local lookahead in GRPO training via gradient extrapolation from two optimizer steps using three backward passes total, improving pass@1 accuracy by 1.65-5.00 points over GRPO and delivering up to 4x step speedup.
-
Teaching Thinking Models to Reason with Tools: A Full-Pipeline Recipe for Tool-Integrated Reasoning
A training recipe for tool-integrated reasoning models achieves state-of-the-art open-source results on math benchmarks such as 96.7% and 99.2% on AIME 2025 at 4B and 30B scales by balancing tool-use trajectories and optimizing for pass@k during SFT before stable RLVR.
-
ZAYA1-8B Technical Report
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
-
Addressing Performance Saturation for LLM RL via Precise Entropy Curve Control
Entrocraft uses rejection sampling to enforce precise entropy schedules in LLM RL by biasing advantages, enabling longer training, better generalization, and higher performance than baselines.
-
Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data
A parameter-free sampling strategy called CUTS combined with Mixed-CUTS training prevents mode collapse in RL for saturated LLM reasoning tasks and raises AIME25 Pass@1 accuracy by up to 15.1% over standard GRPO.
-
HEALing Entropy Collapse: Enhancing Exploration in Few-Shot RLVR via Hybrid-Domain Entropy Dynamics Alignment
HEAL mitigates entropy collapse in few-shot RLVR by selectively adding general-domain data and aligning trajectory-level entropy dynamics, matching full-shot performance with 32 target samples.
-
SPS: Steering Probability Squeezing for Better Exploration in Reinforcement Learning for Large Language Models
SPS interleaves RL and IRL to counteract probability squeezing in LLM reasoning trajectories, improving Pass@k on five benchmarks while identifying an empirical upper bound on multi-sample performance.
-
TIP: Token Importance in On-Policy Distillation
A two-axis taxonomy of student entropy and teacher-student divergence identifies informative tokens in on-policy distillation, allowing near-full performance with 10-50% of tokens.
-
HiRO-Nav: Hybrid ReasOning Enables Efficient Embodied Navigation
HiRO-Nav adaptively triggers reasoning only on high-entropy actions via a hybrid training pipeline and shows better success-token trade-offs than always-reason or never-reason baselines on the CHORES-S benchmark.
-
The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment
The Master Key Hypothesis states that capabilities are low-dimensional directions transferable across models through linear subspace alignment, with UNLOCK demonstrating gains such as 12.1% accuracy improvement on MATH when transferring CoT from 14B to 7B models.
-
Vocabulary Dropout for Curriculum Diversity in LLM Co-Evolution
Vocabulary dropout prevents diversity collapse in LLM co-evolution by masking proposer logits, yielding average +4.4 point solver gains on mathematical reasoning benchmarks at 8B scale.
-
CRISP: Compressed Reasoning via Iterative Self-Policy Distillation
CRISP achieves 57-59% token reduction on MATH-500 with 9-16 point accuracy gains on Qwen3 models via iterative self-distillation of concise reasoning behavior.
-
Why Code, Why Now: An Information-Theoretic Perspective on the Limits of Machine Learning
Task information structure determines ML scaling success, with code's dense verifiable signals enabling predictable progress while sparse-feedback tasks like typical RL do not.
-
Flexible Entropy Control in RLVR with a Gradient-Preserving Perspective
Dynamic clipping strategies based on importance sampling regions enable precise entropy management in RLVR, mitigating collapse and improving benchmark performance.
-
Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning
Entropy Ratio Clipping introduces a global entropy-ratio constraint that stabilizes RL policy updates in LLM post-training beyond local PPO clipping.
-
Auditing Data Membership in Reinforcement Learning With Verifiable Rewards
DIBA detects membership of prompts in RLVR training by measuring reward success changes and policy behavioral drift between pre- and post-RLVR model checkpoints.
-
Kimi Linear: An Expressive, Efficient Attention Architecture
Kimi Linear hybridizes linear attention with a new KDA module to beat full attention on tasks while slashing KV cache by 75% and speeding decoding up to 6x.
-
SCOPE-RL: Stable and Quantitative Control of Policy Entropy in RL Post-Training
SCOPE-RL adds a regularization term built from high-temperature positive samples to quantitatively control entropy dynamics and maintain exploration in RL post-training of reasoning LLMs.
-
LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?
LightReasoner distills supervision signals from SLM-LLM behavioral divergence to improve LLM reasoning on math benchmarks with up to 28.1% accuracy gains and 90-99% reductions in resources.
-
When Importance Sampling Misallocates Credit: Asymmetric Ratios for Outcome-Supervised RL
The paper identifies that importance sampling ratios in outcome-supervised RL misallocate credit by creating unbalanced token updates, and introduces ASPO to correct the asymmetry for positive-advantage tokens.
-
Emergent Slow Thinking in LLMs as Inverse Tree Freezing
RLVR drives a concept network in LLMs through nucleation and freezing into inverse trees that support slow thinking, and intervening with brief SFT at peak frustration outperforms standard RLVR while post-freeze SFT causes forgetting.
-
GIFT: Guided Importance-Aware Fine-Tuning for Diffusion Language Models
GIFT weights tokens by entropy during fine-tuning of diffusion language models and reports better performance than standard SFT on reasoning benchmarks across multiple settings.
-
CE-GPPO: Coordinating Entropy via Gradient-Preserving Clipping Policy Optimization in Reinforcement Learning
CE-GPPO preserves bounded gradients from clipped tokens in PPO to regulate entropy evolution and improve performance on mathematical reasoning benchmarks.
-
SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning
SimpleVLA-RL applies tailored reinforcement learning to VLA models, reaching SoTA on LIBERO, outperforming π₀ on RoboTwin, and surpassing SFT in real-world tasks while reducing data needs and identifying a 'pushcut' phenomenon.
-
InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency
InternVL3.5 advances open-source multimodal models with Cascade RL for +16% reasoning gains and ViR for 4x inference speedup, with the 241B model reaching SOTA among open-source MLLMs on multimodal, reasoning, and agentic tasks.