HALO: Hybrid Adaptive Latent Reasoning for Language Models

HALO：面向语言模型的混合自适应潜在推理

Abstract: We study how to improve a frozen pretrained language model with a small amount of adaptive extra computation. A simple approach is to add additional refinement steps on top of the backbone hidden states, but fixed extra refinement can be wasteful: a one-step refinement head may be too weak, while forcing a second full-sequence refinement step everywhere can increase compute without improving transfer.

摘要： 我们研究了如何通过少量的自适应额外计算来改进冻结的预训练语言模型。一种简单的方法是在主干隐藏状态之上增加额外的细化步骤，但固定的额外细化可能会造成浪费：单步细化头可能过于薄弱，而强制在所有位置进行第二次全序列细化步骤则会增加计算量，却未必能提升迁移效果。

We introduce HALO, a hybrid adaptive latent-refinement method that combines a coarse refinement stage with selective second-stage latent refinement on a subset of tokens chosen by token scoring and monotonic token halting. On the main public benchmark comparison built from MMLU-Pro and GPQA-Diamond, HALO achieves the best overall average among the paper-facing methods, outperforming the frozen backbone, fixed-1, and fixed-2.

我们引入了 HALO，这是一种混合自适应潜在细化方法。它结合了粗略细化阶段与选择性第二阶段潜在细化，后者通过标记评分（token scoring）和单调标记停止（monotonic token halting）机制，仅对部分标记进行处理。在基于 MMLU-Pro 和 GPQA-Diamond 构建的主要公共基准测试中，HALO 在所有论文提及的方法中取得了最佳的总体平均成绩，表现优于冻结主干模型、固定单步（fixed-1）和固定双步（fixed-2）方法。

Internal analysis further shows that HALO reaches nearly the same token-accuracy level as fixed-2 while using fewer average applied refine steps than fixed-1 and far fewer than fixed-2. These results suggest that the key advantage is not simply more refinement, but a better allocation of refinement: HALO achieves the strongest paper-facing result while also using less measured controller compute than either fixed baseline.

内部分析进一步表明，HALO 在达到与固定双步（fixed-2）几乎相同的标记准确率的同时，其平均应用的细化步骤少于固定单步（fixed-1），且远少于固定双步（fixed-2）。这些结果表明，其核心优势不在于单纯增加细化次数，而在于更优的细化分配：HALO 在实现最强性能的同时，其控制器计算消耗也低于任何一种固定基准方法。