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 MoE Reinforcement Learning by Aligning Training and Inference Routers
16 Pith papers cite this work. Polarity classification is still indexing.
16
Pith papers citing it
hub tools
citation-role summary
background 3
method 1
citation-polarity summary
years
2026 16polarities
background 4representative citing papers
Moebius enables runtime switching between EP and TP for MoE LLMs by resharing weights and KV cache, matching the best static choice and improving RL rollouts by 1.16-1.25x with 215-434 ms switches and 2.4% memory overhead.
ForeMoE uses routing foresight from the rollout stage to enable micro-step load balancing in MoE RL post-training via a hierarchical planner and transfer engine, claiming up to 1.45x speedup on 64 GPUs.
Replacing the importance sampling ratio with a stop-gradient self-anchored ratio for positive advantages yields unclipped, REINFORCE-equivalent gradients that improve exploration without training instability.
GDCR assigns step-level rewards via distance to the answer node in a training-time ER graph and SAPO combines these with trajectory advantages for credit assignment in agentic search.
A new RL objective adapts trust-region and off-policy handling automatically via normalized effective sample size of batch policy ratios, matching tuned baselines without new hyperparameters.
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.
AeSlides is a GRPO-based RL framework that uses verifiable aesthetic metrics to optimize LLM slide generation, achieving large gains in layout quality metrics and human scores with only 5K prompts.
Balanced Aggregation fixes sign-length coupling and length downweighting in GRPO by computing separate token means for positive and negative subsets and combining them with sequence-count weights, yielding more stable training and higher benchmark scores.
Relax is a new RL training engine with omni-native design and async execution that delivers up to 2x speedups over baselines like veRL while converging to equivalent reward levels on Qwen3 models.
FP8-RL delivers up to 44% faster rollouts in LLM RL by using blockwise FP8 quantization, KV-cache recalibration, and importance-sampling corrections while keeping learning behavior close to BF16 baselines.
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.
PADD distills from dense teachers to MoE students via neuron clustering, expert warmup, online adaptive distillation, path-refined policy optimization, and reward-augmented load balancing, yielding gains on math reasoning benchmarks.
MinT is a system for managing million-scale LoRA adapter catalogs on shared 1T-parameter base models, with reported efficiency gains in adapter movement, multi-policy training, and catalog addressability.
ASymPO normalizes token losses by average current-policy negative log-probability to restore zero-sum balance in asynchronous LLM RL without behavior information.
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.
-
Moebius: Serving Mixture-of-Expert Models with Seamless Runtime Parallelism Switch
Moebius enables runtime switching between EP and TP for MoE LLMs by resharing weights and KV cache, matching the best static choice and improving RL rollouts by 1.16-1.25x with 215-434 ms switches and 2.4% memory overhead.
-
Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training
ForeMoE uses routing foresight from the rollout stage to enable micro-step load balancing in MoE RL post-training via a hierarchical planner and transfer engine, claiming up to 1.45x speedup on 64 GPUs.
-
UP: Unbounded Positive Asymmetric Optimization for Breaking the Exploration-Stability Dilemma
Replacing the importance sampling ratio with a stop-gradient self-anchored ratio for positive advantages yields unclipped, REINFORCE-equivalent gradients that improve exploration without training instability.
-
Beyond Trajectory Rewards: Step-level Credit Assignment for Agentic Search via Graph Modeling
GDCR assigns step-level rewards via distance to the answer node in a training-time ER graph and SAPO combines these with trajectory advantages for credit assignment in agentic search.
-
Trust the Batch, On- or Off-Policy: Adaptive Policy Optimization for RL Post-Training
A new RL objective adapts trust-region and off-policy handling automatically via normalized effective sample size of batch policy ratios, matching tuned baselines without new hyperparameters.
-
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.
-
AeSlides: Incentivizing Aesthetic Layout in LLM-Based Slide Generation via Verifiable Rewards
AeSlides is a GRPO-based RL framework that uses verifiable aesthetic metrics to optimize LLM slide generation, achieving large gains in layout quality metrics and human scores with only 5K prompts.
-
Balanced Aggregation: Understanding and Fixing Aggregation Bias in GRPO
Balanced Aggregation fixes sign-length coupling and length downweighting in GRPO by computing separate token means for positive and negative subsets and combining them with sequence-count weights, yielding more stable training and higher benchmark scores.
-
Relax: An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale
Relax is a new RL training engine with omni-native design and async execution that delivers up to 2x speedups over baselines like veRL while converging to equivalent reward levels on Qwen3 models.
-
FP8-RL: A Practical and Stable Low-Precision Stack for LLM Reinforcement Learning
FP8-RL delivers up to 44% faster rollouts in LLM RL by using blockwise FP8 quantization, KV-cache recalibration, and importance-sampling corrections while keeping learning behavior close to BF16 baselines.
-
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.
-
PADD: Path-Aligned Decompression Distillation for Non-Router Teacher to Guide MoE Student Learning
PADD distills from dense teachers to MoE students via neuron clustering, expert warmup, online adaptive distillation, path-refined policy optimization, and reward-augmented load balancing, yielding gains on math reasoning benchmarks.
-
MinT: Managed Infrastructure for Training and Serving Millions of LLMs
MinT is a system for managing million-scale LoRA adapter catalogs on shared 1T-parameter base models, with reported efficiency gains in adapter movement, multi-policy training, and catalog addressability.
-
ASymPO: Asymmetric-Scale Policy Optimization for Asynchronous LLM Post-Training Without Behavior Information
ASymPO normalizes token losses by average current-policy negative log-probability to restore zero-sum balance in asynchronous LLM RL without behavior information.