ViT-Up: Faithful Feature Upsampling for Vision Transformers Paper • 2606.14024 • Published 18 days ago • 9
HorizonStream: Long-Horizon Attention for Streaming 3D Reconstruction Paper • 2605.23889 • Published May 22 • 4
SOD: Step-wise On-policy Distillation for Small Language Model Agents Paper • 2605.07725 • Published May 8 • 25
Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information Paper • 2605.11609 • Published May 12 • 196
Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models Paper • 2605.08472 • Published May 8 • 5
CiteVQA: Benchmarking Evidence Attribution for Trustworthy Document Intelligence Paper • 2605.12882 • Published May 13 • 274
Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models Paper • 2605.07721 • Published May 8 • 29