Tag
作者介绍 ALHR（Adaptive Learnable Hierarchical Routing），一种基于静态二叉树与可学习函数的稀疏注意力系统，在 MQAR 测试中每查询仅读取约 30 个键（相比稠密注意力的 512 个），实现约 35.3 倍 KV 压缩，同时保持接近的准确率，并将推理复杂度降至 NlogN、显存线性扩展。
This paper proposes SLM-Conditioned Hierarchical Relation Routing, an architecture that integrates small language models into graph neural networks to enable adaptive message selection in labeled property graphs, enhancing prediction accuracy by leveraging contextual semantic information.
HiRoute proposes a hierarchical routed prompt-tuning framework that separates category-agnostic safety control from category-specific response guidance for LLM safety alignment, reducing over-refusal while maintaining high safety rates.
Hierarchical Global Attention (HGA) is a drop-in replacement for dense causal attention in pretrained long-context transformers that uses hierarchical two-level routing to enable exact attention over a small routed working set, allowing models like Qwen3-30B to run at 64K context on a single RTX 5090 with minimal quality loss.
SkillTree is a tool that organizes an AI agent's multiple skills into a tree structure and loads them on demand, avoiding stuffing all instructions into the context at once and improving agent efficiency.