跳到正文
arXiv cs.AI· Firas Spanioli, Yaniv David·· 10 小时前AI 评分34

GateWatch:用布尔电路实现低开销 LLM 监控

GateWatch: Low-Overhead LLM Monitoring with Boolean Circuits

AI 导读

GateWatch 用少量二值化 SAE 特征上的显式布尔逻辑构成紧凑可解释分类器,在 Gemma-2-9B 的 24 个任务上平均仅用 3.8 个 SAE 特征和 3.5 个逻辑门,即达到 80.49% 平均测试准确率,仅比完整 16K 特征 SAE 探针低 1.17 个百分点,探针内存减少 99.98%。

正文 · 原文

View PDF

Abstract:As large language models (LLMs) and LLM-based agents are increasingly deployed, monitoring their behavior becomes important for detecting unsafe or unintended actions. Activation probes offer an efficient approach, but existing methods face trade-offs between accuracy, interpretability, and computational overhead. Probes on raw activations lack interpretable inputs, whereas sparse autoencoder (SAE) probes use interpretable features but often rely on additive classifiers and large SAE encoders, limiting interpretability and increasing memory overhead. We introduce GateWatch, a compact, interpretable classifier that expresses its entire decision rule as an explicit Boolean logic over a small number of binarized SAE features. We explore three approaches to constructing these circuits under a specified gate budget. For efficient deployment, GateWatchServe computes only selected SAE features, enables streaming of their binary values across layers during inference, and evaluates multiple GateWatch classifiers in parallel using compact lookup tables. Across 24 tasks on Gemma-2-9B, GateWatch achieves 80.49% mean test accuracy using only 3.8 SAE features and 3.5 logic gates per probe on average, just 1.17 percentage points below full 16K-feature SAE probes, while matching or outperforming them on 10 tasks and reducing probe memory by 99.98%. Inside vLLM, 1,000 concurrent multi-layer, multi-token GateWatch classifiers retain 97-98% of baseline throughput and 99.8% of KV-cache capacity.
Comments: 23 pages
Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2611.24619 [cs.AI]
  (or arXiv:2611.24619v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.24619

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Firas Spanioli [view email]
[v1] Wed, 18 Nov 2026 15:37:04 UTC (2,781 KB)

来源:arXiv cs.AI · arxiv.org