MHLA: Restoring Expressivity of Linear Attention via Token-Level Multi-Head
Paper
•
2601.07832
•
Published
•
52
Optimization
-
-
Towards Automated Kernel Generation in the Era of LLMs
Paper • 2601.15727 • Published • 18 -
HERMES: KV Cache as Hierarchical Memory for Efficient Streaming Video Understanding
Paper • 2601.14724 • Published • 74 -
Post-LayerNorm Is Back: Stable, ExpressivE, and Deep
Paper • 2601.19895 • Published • 23 -
Scalable Power Sampling: Unlocking Efficient, Training-Free Reasoning for LLMs via Distribution Sharpening
Paper • 2601.21590 • Published • 12