AttnSR: Attention Step Reward for Long-Horizon Agentic RL
cs.LG, cs.AI
Submitted: 2026-11-05
Updated: 2026-11-05
Code: https://github.com/fairyshine/AttnSR
Terminology
Sources
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- How Does Reasoning Flow? Tracing Attention-Induced Information Flow for Targeted RL in LLMs
- Agentic Reinforcement Learning with Implicit Step Rewards
- Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents
- Qwen2.5 Technical Report
- High-Dimensional Continuous Control Using Generalized Advantage Estimation
- Proximal Policy Optimization Algorithms
- Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
- Hindsight Credit Assignment for Long-Horizon LLM Agents
- Process-Supervised Reinforcement Learning for Interactive Multimodal Tool-Use Agents
- SPA-RL: Reinforcing LLM Agents via Stepwise Progress Attribution
- RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning
- Reinforcing Language Agents via Policy Optimization with Action Decomposition
- AgentPRM: Process Reward Models for LLM Agents via Step-Wise Promise and Progress
- Pinpointing crucial steps: Attribution-based Credit Assignment for Verifiable Reinforcement Learning
- From Reasoning to Agentic: Credit Assignment in Reinforcement Learning for Large Language Models
- SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis