CLT-Forge: A Scalable Library for Cross-Layer Transcoders and Attribution Graphs

TL;DR

CLT-Forge is an open-source library that enables scalable training and interpretability of cross-layer transcoders, facilitating more compact and interpretable feature attribution graphs for understanding large language models.

Contribution

It introduces a comprehensive framework combining scalable distributed training, automated interpretability, and visualization tools for cross-layer transcoders in LLMs.

Findings

Enables end-to-end training of CLTs at scale

Provides automated interpretability pipeline

Facilitates compact and interpretable attribution graphs

Abstract

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.