Papers arxiv:2605.08737 The Extrapolation Cliff in On-Policy Distillation of Near-Deterministic Structured Outputs Published on May 9 Submitted by XinLi on May 14 Nanyang Technological University Authors: Xin Li , , , , Abstract On-policy distillation with reward extrapolation exhibits a safety thr… …
Papers arxiv:2605.12995 F-GRPO: Factorized Group-Relative Policy Optimization for Unified Candidate Generation and Ranking Published on May 13 Submitted by Rohan Surana on May 14 McAuley-Lab Authors: Rohan Surana , , , , , , , , , , , Abstract A unified framework combines candidate generation and r… …
… Also, the fix only applies when on policy=True. How do you handle cases where the policy has drifted? Is there a threshold or mechanism to detect when to switch to off-policy mode? …
… Each code acts like a small steering signal for a sub-policy inside the model. …
… The system logged 7.5M agent invocations, ~300K onchain actions, ~$20M in volume, 70B inference tokens, and 99.9% settlement success for policy-valid transactions. …
… The following papers were recommended by the Semantic Scholar API AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement Learning 2026 DPEPO: Diverse Parallel Exploration Policy Optimization for LLM-based Agents 2026 GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifia… …
… The following papers were recommended by the Semantic Scholar API PAINT: Partial-Solution Adaptive Interpolated Training for Self-Distilled Reasoners 2026 Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR 2026 On-Policy Distillation with Best-of-N Teac… …
Papers arxiv:2605.06200 A^2TGPO: Agentic Turn-Group Policy Optimization with Adaptive Turn-level Clipping Published on May 7 Submitted by Dingwei Chen on May 8 Tencent Authors: Dingwei Chen , , , , , , , Abstract Reinforcement learning for agentic LLMs suffers from sparse rewards and challenges in … …
… MinT thus manages million-scale LoRA policy catalogs while training and serving selected adapter revisions over shared 1T-class base models. …
… To address this, we introduce RLDX-1, a general-purpose robotic policy for dexterous manipulation built on the Multi-Stream Action Transformer MSAT , an architecture that unifies these capabilities by integrating heterogeneous modalities through modality-specific streams with cross-modal joint self… …