The Quantum Sieve Tracer: A Hybrid Framework for Layer-Wise Activation Tracing in Large Language Models
This work addresses mechanistic interpretability for AI researchers by providing a high-resolution tool to analyze attention circuits, though it is incremental as it builds on existing causal tracing methods with quantum enhancements.
The paper tackled the challenge of separating sparse semantic signals from noise in Large Language Models by introducing the Quantum Sieve Tracer, a hybrid quantum-classical framework for tracing factual recall circuits, and found that ablating specific heads in Llama's layer 9 improved factual recall, revealing divergent mechanisms between models.
Mechanistic interpretability aims to reverse-engineer the internal computations of Large Language Models (LLMs), yet separating sparse semantic signals from high-dimensional polysemantic noise remains a significant challenge. This paper introduces the Quantum Sieve Tracer, a hybrid quantum-classical framework designed to characterize factual recall circuits. We implement a modular pipeline that first localizes critical layers using classical causal tracing, then maps specific attention head activations into an exponentially large quantum Hilbert space. Using open-weight models (Meta Llama-3.2-1B and Alibaba Qwen2.5-1.5B-Instruct), we perform a two-stage analysis that reveals a fundamental architectural divergence. While Qwen's layer 7 circuit functions as a classic Recall Hub, we discover that Llama's layer 9 acts as an Interference Suppression circuit, where ablating the identified heads paradoxically improves factual recall. Our results demonstrate that quantum kernels can distinguish between these constructive (recall) and reductive (suppression) mechanisms, offering a high-resolution tool for analyzing the fine-grained topology of attention.