MoE-LoRA 不用固定 k 值了，CARE 根据 token 难度自动选专家，在 LLaMA 和 Qwen 上效率更高，还能顺手做 OOD 检测。
Mixture-of-Experts (MoE) LoRA 通常对每个 token 固定激活 k 个专家，但易在简单 token 上浪费计算、难 token 上欠拟合。CARE 利用路由器输出分布作为不确定性信号，以核采样方式按置信度阈值自适应选择专家，且当已选专家分歧时小幅扩展。在 LLaMA-3.1-8B 和 Qwen2.5-7B 的 8 个常识基准、数学、代码和知识任务上，CARE 在相同计算量下优于固定 top-k MoE-LoRA，并能在激活更少专家时匹敌固定 k=4 基线。置信度与分歧信号还提升了 OOD 检测效果，超越 MSP、熵和多遍代理方法。
Spend Experts Where You Are Unsure: Confidence-Adaptive Routing for Mixture-of-Experts LoRA
Mixture-of-Experts (MoE) variants of Low-Rank Adaptation (LoRA) route every token to a fixed number of experts $k$. Tokens differ in how uncertain the model is about them, so a single k over-spends on easy tokens and under-serves hard ones. We observe that the router's output distribution is already a per-token uncertainty signal: peaked mass indicates confidence, while a flat distribution indicates ambiguity. We introduce CARE (Confidence-Adaptive Routing of Experts), which admits experts in a nucleus fashion. Experts are activated in decreasing router weight until their cumulative mass reaches a threshold, with a small extension when the admitted experts disagree. A budget thermostat calibrates the threshold so that the average number of active experts matches any target. CARE is a drop-in, single-forward-pass rule with no extra parameters. Across eight commonsense benchmarks on LLaMA-3.1-8B and Qwen2.5-7B, as well as math, code, and knowledge tasks, CARE improves over fixed top-k MoE-LoRA at matched compute and matches the fixed-k=4 baseline while activating fewer experts. The same confidence and disagreement signals also improve out-of-distribution detection over MSP, entropy, and multi-pass proxies. We support the design with nucleus fidelity, budget optimality, and an epistemic reading of disagreement, and we release code.