ReLibra uses pre-known token-to-expert routing from RL rollouts to perform inter-batch expert reordering and intra-batch replication, delivering up to 1.6x higher throughput than Megatron-LM and 1.2x over oracle-equipped EPLB while staying within 6-10% of an ideal balanced baseline.
hub
Stabilizing reinforcement learning with llms: Formulation and practices.arXiv preprint arXiv:2512.01374, 2025a
19 Pith papers cite this work. Polarity classification is still indexing.
19
Pith papers citing it
hub tools
citation-role summary
background 3
method 1
citation-polarity summary
polarities
background 4representative citing papers
TaskMem uses RL in two phases to learn a task-focused memorization policy for multimodal agents, yielding 5.3-7.0% VQA accuracy gains on reformulated streaming benchmarks from VideoMME, EgoLife, and EgoTempo.
GRPO suffers advantage collapse on uniform-reward groups; ACR quantifies it and AVSPO adds virtual samples to restore gradients, yielding 4-6% accuracy gains on math benchmarks across 0.5B-14B models.
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.
Training-inference mismatch in separated rollout and optimization stages of LLM RL can independently cause training collapse.
ICRL uses joint RL training of solver and critic with distribution-calibration re-weighting and role-wise advantage estimation to internalize critique into unassisted LLM performance, yielding 6.4-point gains on agentic tasks and 7.0 on math reasoning with Qwen3 models.
Missing old logits in async agentic RL entangle discrepancy and staleness terms in PPO off-policy correction; exact acquisition methods and revised PPO-EWMA restore decoupled updates with reported gains in speed and performance.
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.
T²PO improves stability and performance in multi-turn agentic RL by using uncertainty dynamics at token and turn levels to guide exploration and avoid wasted rollouts.
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.
OpsLLM's pipeline (HITL data curation, SFT, GRPO RL with a domain process reward model) improves LLM accuracy on software-operations QA and RCA, especially on in-distribution root-cause-analysis tasks.
An ELBO-based likelihood estimator from the final generated sample dominates other RL design factors for diffusion models, raising GenEval from 0.24 to 0.95 in 90 GPU hours with better efficiency than prior methods.
A single LLM agent that injects and repairs its own bugs in real repositories improves SWE-bench Verified and Pro by +10.4 and +7.8 points, outperforming a human-data RL baseline.
The paper reformulates industrial continual learning for LLMs as a closed-loop ecosystem problem, identifies three core challenges, and organizes solutions around five lifecycle design principles.
DRPO introduces a smooth quadratic regularizer on policy divergence that preserves DPPO's trust-region geometry while providing continuous corrective gradients instead of hard masking.
PEFT adapters are positioned as persistent personal state on foundation models, organized via Scale Up, Scale Down, and Scale Out axes, with MinT as an infrastructure example for managing them.
POPO uses recency-based prioritized group replay and decoupled off-policy optimization to avoid zero-variance ineffective samples in RLVR, accelerating LLM reasoning finetuning with fewer rollouts.
Introduces Discrepancy-Constrained MDP (DCMDP) with Lagrangian relaxation to optimize LLM RL under train-inference discrepancy constraints, claiming performance gains on 8B and 30B models.
GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.
citing papers explorer
-
ReLibra: Routing-Replay-Guided Load Balancing for MoE Training in Reinforcement Learning
ReLibra uses pre-known token-to-expert routing from RL rollouts to perform inter-batch expert reordering and intra-batch replication, delivering up to 1.6x higher throughput than Megatron-LM and 1.2x over oracle-equipped EPLB while staying within 6-10% of an ideal balanced baseline.
-
Task-Focused Memorization for Multimodal Agents
TaskMem uses RL in two phases to learn a task-focused memorization policy for multimodal agents, yielding 5.3-7.0% VQA accuracy gains on reformulated streaming benchmarks from VideoMME, EgoLife, and EgoTempo.
-
Advantage Collapse in Group Relative Policy Optimization: Diagnosis and Mitigation
GRPO suffers advantage collapse on uniform-reward groups; ACR quantifies it and AVSPO adds virtual samples to restore gradients, yielding 4-6% accuracy gains on math benchmarks across 0.5B-14B models.
-
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.
-
Diagnosing Training Inference Mismatch in LLM Reinforcement Learning
Training-inference mismatch in separated rollout and optimization stages of LLM RL can independently cause training collapse.
-
ICRL: Learning to Internalize Self-Critique with Reinforcement Learning
ICRL uses joint RL training of solver and critic with distribution-calibration re-weighting and role-wise advantage estimation to internalize critique into unassisted LLM performance, yielding 6.4-point gains on agentic tasks and 7.0 on math reasoning with Qwen3 models.
-
Missing Old Logits in Asynchronous Agentic RL: Semantic Mismatch and Repair Methods for Off-Policy Correction
Missing old logits in async agentic RL entangle discrepancy and staleness terms in PPO off-policy correction; exact acquisition methods and revised PPO-EWMA restore decoupled updates with reported gains in speed and performance.
-
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.
-
T$^2$PO: Uncertainty-Guided Exploration Control for Stable Multi-Turn Agentic Reinforcement Learning
T²PO improves stability and performance in multi-turn agentic RL by using uncertainty dynamics at token and turn levels to guide exploration and avoid wasted rollouts.
-
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.
-
OpsLLM: Construction of Large Language Model for Software Operations with Multi-stage Learning
OpsLLM's pipeline (HITL data curation, SFT, GRPO RL with a domain process reward model) improves LLM accuracy on software-operations QA and RCA, especially on in-distribution root-cause-analysis tasks.
-
Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design
An ELBO-based likelihood estimator from the final generated sample dominates other RL design factors for diffusion models, raising GenEval from 0.24 to 0.95 in 90 GPU hours with better efficiency than prior methods.
-
Toward Training Superintelligent Software Agents through Self-Play SWE-RL
A single LLM agent that injects and repairs its own bugs in real repositories improves SWE-bench Verified and Pro by +10.4 and +7.8 points, outperforming a human-data RL baseline.
-
LLM Evolution as an Industry-Scale Ecosystem: A Lifecycle Perspective on Continual Learning
The paper reformulates industrial continual learning for LLMs as a closed-loop ecosystem problem, identifies three core challenges, and organizes solutions around five lifecycle design principles.
-
Rethinking the Divergence Regularization in LLM RL
DRPO introduces a smooth quadratic regularizer on policy divergence that preserves DPPO's trust-region geometry while providing continuous corrective gradients instead of hard masking.
-
On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters
PEFT adapters are positioned as persistent personal state on foundation models, organized via Scale Up, Scale Down, and Scale Out axes, with MinT as an infrastructure example for managing them.
-
RLVR without Ineffective Samples: Group Prioritized Off-Policy Optimization for LLM Reasoning
POPO uses recency-based prioritized group replay and decoupled off-policy optimization to avoid zero-variance ineffective samples in RLVR, accelerating LLM reasoning finetuning with fewer rollouts.
-
Reformulate LLM Reinforcement Learning for Efficient Training under Black-box Discrepancy
Introduces Discrepancy-Constrained MDP (DCMDP) with Lagrangian relaxation to optimize LLM RL under train-inference discrepancy constraints, claiming performance gains on 8B and 30B models.
-
Sharpness-Guided Group Relative Policy Optimization via Probability Shaping
GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.