Correctness vs Efficiency: Behavior Control with Weight-Space Extrapolation in Code RL
cs.LG, cs.AI, cs.CL
Submitted: 2026-05-27
Updated: 2026-11-09
Terminology
Sources
- OpenCodeReasoning-II: A Simple Test Time Scaling Approach via Self-Critique
- On Predictability of Reinforcement Learning Dynamics for Large Language Models
- BigO(Bench): Can LLMs Generate Code with Controlled Time and Space Complexity?
- The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models
- Process Supervision-Guided Policy Optimization for Code Generation
- Afterburner: Reinforcement Learning Facilitates Self-Improving Code Efficiency Optimization
- CWM: An Open-Weights LLM for Research on Code Generation with World Models
- Neural Thickets: Diverse Task Experts Are Dense Around Pretrained Weights
- ReVeal: Self-Evolving Code Agents via Reliable Self-Verification
- Scaling Test-Time Compute for Agentic Coding
- Kimi K2: Open Agentic Intelligence
- TACO: Topics in Algorithmic COde generation dataset
- Understanding R1-Zero-Like Training: A Critical Perspective
- SWE-fficiency: Can Language Models Optimize Real-World Repositories on Real Workloads?
- Self-Execution Simulation Improves Coding Models
- MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention
- AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
- Qwen2.5 Technical Report
- WARP: On the Benefits of Weight Averaged Rewarded Policies
- Linear Dynamics in the RLVR Training of Large Language Models
Related papers
- 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
- (Mis)generalization of Helpful-Only Fine-tuning