A verifiable empirical win rate reward combined with gradient masking enables RL training of a 7B model to reach betting-market calibration on NFL win probabilities using only outcome data.
hub
Gtpo and grpo-s: Token and sequence-level reward shaping with policy entropy.arXiv preprint arXiv:2508.04349
15 Pith papers cite this work. Polarity classification is still indexing.
15
Pith papers citing it
hub tools
citation-role summary
background 2
citation-polarity summary
roles
background 2representative citing papers
TACO soft-suppresses positive GRPO credit on high tail-risk tokens (surprisal above local entropy) and consistently beats GRPO-style baselines on three LLMs and eight reasoning benchmarks while stabilizing long training.
BALTO projects claim-level verification into balanced token-level rewards for RL-based hallucination mitigation in LLMs.
ISPO densifies GRPO rewards with sequence-level informativeness and token-level directional signals from policy probabilities to reduce zero-advantage collapse and hallucinated certainty on math benchmarks.
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.
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.
RTT bridges response-level rubrics to token-level rewards via a relevance discriminator and intra-sample group normalization, yielding higher instruction and rubric accuracy than baselines.
Group Causal Counterfactual Policy Optimization trains LLMs on generalizable reasoning by defining episodic rewards for counterfactual robustness and transferability then optimizing the policy with token-level advantages.
EGM enables 8B VLMs to reach 91.4 IoU on RefCOCO at 737 ms latency, outperforming a 235B model at 4320 ms, by substituting volume of mid-quality tokens for model scale.
MRPO is a step-aware RL method that penalizes early reasoning errors exponentially more when the final answer is incorrect, reducing early-stage failures from 64% to 13% and outperforming baselines including larger models on medical VQA tasks.
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
IMAX trains soft prefixes with an InfoMax reward to drive diverse exploration in RLVR, yielding up to 11.60% gains in Pass@4 over standard RLVR across model scales.
Derives a token-level entropy change approximation revealing four factors, identifies limitations in prior entropy interventions, and proposes STEER which adaptively reweights tokens to mitigate collapse and improve performance on math and coding benchmarks.
VaseMuseum is a training-free multimodal agent that combines DeepResearch-style retrieval, source/response reliability control, and best-of-K reranking to improve citation validity and reduce hallucination for museum VQA on ancient Greek pottery.
citing papers explorer
-
Verifiable Rewards for Calibrated Probabilistic Forecasting
A verifiable empirical win rate reward combined with gradient masking enables RL training of a 7B model to reach betting-market calibration on NFL win probabilities using only outcome data.
-
When Implausible Tokens Get Reinforced: Tail-Aware Credit Calibration for LLM Reinforcement Learning
TACO soft-suppresses positive GRPO credit on high tail-risk tokens (surprisal above local entropy) and consistently beats GRPO-style baselines on three LLMs and eight reasoning benchmarks while stabilizing long training.
-
BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation
BALTO projects claim-level verification into balanced token-level rewards for RL-based hallucination mitigation in LLMs.
-
Momentum for Reasoning: Dense Intrinsic Signals in Policy Optimization
ISPO densifies GRPO rewards with sequence-level informativeness and token-level directional signals from policy probabilities to reduce zero-advantage collapse and hallucinated certainty on math benchmarks.
-
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.
-
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.
-
Rubrics to Tokens: Bridging Response-level Rubrics and Token-level Rewards in Instruction Following Tasks
RTT bridges response-level rubrics to token-level rewards via a relevance discriminator and intra-sample group normalization, yielding higher instruction and rubric accuracy than baselines.
-
Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoning
Group Causal Counterfactual Policy Optimization trains LLMs on generalizable reasoning by defining episodic rewards for counterfactual robustness and transferability then optimizing the policy with token-level advantages.
-
EGM: Efficient Visual Grounding Language Models
EGM enables 8B VLMs to reach 91.4 IoU on RefCOCO at 737 ms latency, outperforming a 235B model at 4320 ms, by substituting volume of mid-quality tokens for model scale.
-
Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal Reasoning
MRPO is a step-aware RL method that penalizes early reasoning errors exponentially more when the final answer is incorrect, reducing early-stage failures from 64% to 13% and outperforming baselines including larger models on medical VQA tasks.
-
Trust Region On-Policy Distillation
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
-
How You Begin is How You Reason: Driving Exploration in RLVR via Prefix-Tuned Priors
IMAX trains soft prefixes with an InfoMax reward to drive diverse exploration in RLVR, yielding up to 11.60% gains in Pass@4 over standard RLVR across model scales.
-
Rethinking Entropy Interventions in RLVR: An Entropy Change Perspective
Derives a token-level entropy change approximation revealing four factors, identifies limitations in prior entropy interventions, and proposes STEER which adaptively reweights tokens to mitigate collapse and improve performance on math and coding benchmarks.
-
VaseMuseum: Digital Intelligent Museum for Ancient Greek Pottery
VaseMuseum is a training-free multimodal agent that combines DeepResearch-style retrieval, source/response reliability control, and best-of-K reranking to improve citation validity and reduce hallucination for museum VQA on ancient Greek pottery.
- Know When to Stop: Segment-Level Credit Assignment for Reducing Overthinking