CircuitKIT : Circuit Discovery, Evaluation, and Application Toolkit for Mechanistic Interpretability
The paper introduces a source-available library meant to make circuit analysis easier to run, compare, and apply.
CircuitKIT connects discovery, evaluation, diagnostics, and intervention workflows through a typed, serializable circuit representation. The authors frame it as a fix for fragmented mechanistic-interpretability tooling, where researchers often combine separate implementations and hand-write contrastive prompts. It includes discovery algorithms, interfaces for turning structured data into discovery tasks, and modules for pruning, editing, steering, and selective fine-tuning. ArXiv · AI/CL/LG's note
CircuitKIT connects discovery, evaluation, diagnostics, and intervention workflows through a typed, serializable circuit representation. The authors frame it as a fix for fragmented mechanistic-interpretability tooling, where researchers often combine separate implementations and hand-write contrastive prompts. It includes discovery algorithms, interfaces for turning structured data into discovery tasks, and modules for pruning, editing, steering, and selective fine-tuning. ArXiv · AI/CL/LG's note
score 5