用可组合流程简化大模型机制可解释性研究的复现与协作。
MURANO: Design, Run, and Reproduce Mechanistic Interpretability Experiments as Composable Pipelines
- 将加载、记录、归因、干预、评估等操作拆解为可组合步骤
- 通过命名结果传递实现跨工具无缝衔接，支持流程化实验
- 适合跨学科研究人员复现或开展可解释性实验
本文提出 Murano，一个开源框架，用于设计、运行和复现大规模语言模型的机制可解释性研究，面向跨学科研究人员。这类研究通常包含加载、记录、归因、干预和评估五个环节，而现有库多仅覆盖其中部分流程，导致研究者需手动适配不同工具的输出格式。为解决此问题，Murano 将上述五类操作表示为可组合的步骤，各步骤以命名结果交换数据，并明确定义输入与输出。管道按顺序执行步骤，组件身份通过标准地址传递。Murano 基于现有可解释性与机器学习库构建。我们通过两个经典研究的复现和一个示范性的稀疏自编码器案例，展示了其有效性。
原文摘要 · Abstract (English)
This paper presents Murano, an open source framework for designing, running, and reproducing mechanistic interpretability studies of large language models, intended for researchers across disciplines. These studies often combine loading, recording, attribution, intervention, and evaluation, while existing libraries tend to focus on different parts of this workflow. As a result, researchers using several libraries may need to adapt outputs from one for use by another. To bridge this gap, Murano represents operations from these five areas as composable steps. Steps exchange named result artifacts and declare the inputs they require and the outputs they produce. A pipeline executes its steps in the order supplied, and Murano uses canonical addresses when component identities pass between operations. Murano builds on existing interpretability and machine learning libraries. We demonstrate Murano through two reproductions of established interpretability studies and one illustrative sparse autoencoder case study.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。