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

2026-03-21

AI Summary

This work addresses the challenge of excessive size and redundancy in attribution maps generated by existing dictionary learning–based mechanistic interpretability methods employing Cross-Layer Transcoders (CLTs), which hinder efficient training and analysis. To overcome these limitations, the authors introduce the first open-source CLT framework that integrates distributed training, activation cache compression, and an automated interpretability pipeline. By leveraging model sharding, Circuit-Tracer–based attribution computation, a unified analysis workflow, and interactive visualization, the proposed system substantially enhances the scalability and practicality of CLT-based interpretability. This tool effectively mitigates attribution map redundancy, offering an efficient, cohesive, and scalable solution for mechanistic interpretability research in large language models.

Technology Category

• Machine Learning: Transparent, Interpretable, Explainable ML
• Computer Vision: Interpretability, Explainability, and Transparency
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

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.

Problem

• mechanistic interpretability
• Cross-Layer Transcoders
• attribution graphs
• Large Language Models
• sparse features

Innovation

• Cross-Layer Transcoders
• Mechanistic Interpretability
• Attribution Graphs
• Distributed Training
• Feature Attribution