---
id: 20260710-T0-08
title: "TriRoute：统一学习路由实现自适应注意力、专家和KV缓存分配"
title_en: "TriRoute: Unified Learned Routing for Adaptive Attention, Experts, KV-Cache"
url: https://ai.daily.yangsir.net/daily/20260710-T0-08
issue_date: 2026-07-10
publish_date: 2026-07-09T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2607.06601
---

# TriRoute：统一学习路由实现自适应注意力、专家和KV缓存分配

新论文提出TriRoute，一种统一的基于学习的路由方法，可同时实现自适应注意力、混合专家（MoE）和KV缓存分配。与单独的MoE（稀疏化FFN）或MoD（跳过transformer块）不同，TriRoute在同一框架内协调多个条件计算维度，有望进一步降低推理成本。

## English Version

**TriRoute: Unified Learned Routing for Adaptive Attention, Experts, KV-Cache**

A new paper introduces TriRoute, a unified learned routing approach that jointly optimizes adaptive attention, Mixture-of-Experts, and KV-cache allocation. Unlike methods that act on a single axis (MoE for FFN sparsification or MoD for block skipping), TriRoute coordinates multiple conditional computation dimensions within one framework, potentially reducing per-token inference costs.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2607.06601)

**详情页**：https://ai.daily.yangsir.net/daily/20260710-T0-08

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*