Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
CircuitKIT is a comprehensive toolkit designed to streamline circuit analysis for mechanistic interpretability by integrating discovery, evaluation, and intervention processes into a unified framework. This toolkit addresses the challenges posed by fragmented implementations and manual prompt authoring, facilitating easier comparison and application of various circuit analysis methods. Key results demonstrate that CircuitKIT enhances the efficiency of model explanation and enables effective downstream interventions such as pruning and selective fine-tuning.
CircuitKIT transforms circuit analysis from a fragmented process into a cohesive toolkit, enabling seamless model interventions and comparisons across methods.
Circuit analysis can support not only model explanation but also downstream interventions such as pruning, editing, steering, and selective fine-tuning. However, conducting such analyses currently requires stitching together separate implementations for discovery, evaluation, and intervention, as well as hand-authoring the contrastive prompts required by many discovery methods. This fragmentation makes methods difficult to compare and limits their application beyond canonical tasks. We introduce CircuitKIT, a source-available library that connects the circuit-analysis workflow through a typed, serializable representation. CircuitKIT provides a suite of discovery algorithms, declarative interfaces for mapping structured data into discovery tasks, complementary circuit diagnostics, and downstream application modules. Together, these components provide common infrastructure for conducting and comparing circuit analyses. The library, examples, notebooks, and documentation are released at https://github.com/Lexsi-Labs/CircuitKIT .