CLT-Forge: A Scalable Library for Cross-Layer Transcoders and Attribution Graphs
This work addresses scalability issues in mechanistic interpretability for researchers, though it is incremental as it builds on existing CLT methods.
The paper tackles the challenge of scaling Cross-Layer Transcoders (CLTs) for mechanistic interpretability of Large Language Models by introducing an open-source library that integrates distributed training, automated interpretability pipelines, and visualization tools, resulting in a practical solution for more compact and interpretable feature attribution graphs.
Mechanistic interpretability seeks to understand how Large Language Models (LLMs) represent and process information. Recent approaches based on dictionary learning and transcoders enable representing model computation in terms of sparse, interpretable features and their interactions, giving rise to feature attribution graphs. However, these graphs are often large and redundant, limiting their interpretability in practice. Cross-Layer Transcoders (CLTs) address this issue by sharing features across layers while preserving layer-specific decoding, yielding more compact representations, but remain difficult to train and analyze at scale. We introduce an open-source library for end-to-end training and interpretability of CLTs. Our framework integrates scalable distributed training with model sharding and compressed activation caching, a unified automated interpretability pipeline for feature analysis and explanation, attribution graph computation using Circuit-Tracer, and a flexible visualization interface. This provides a practical and unified solution for scaling CLT-based mechanistic interpretability. Our code is available at: https://github.com/LLM-Interp/CLT-Forge.