analiza aiqbench
CLT-Forge is an open-source library designed to scale Cross-Layer Transcoders for mechanistic interpretability of Large Language Models. By integrating distributed training with compressed activation caching and automated pipelines using Circuit-Tracer, it addresses the redundancy issues inherent in current feature attribution graphs. This tool offers a practical framework for researchers aiming to analyze how LLMs process information through sparse, interpretable features across layers.
CLT-Forge: A Scalable Library for Cross-Layer Transcoders and Attribution Graphs
arXiv:2603.21014v2 Announce Type: replace-cross 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.
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: this https URL.
9 pages, 7 figures, 1 table. Code: this https URL. Demonstration video: this https URL
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2603.21014 [cs.LG]
(or arXiv:2603.21014v2 [cs.LG] for this version)
Text integral sub licența deschisă a publicației · 177 cuvinte
Subiecte conexe
Vertical Advantage: Transforming Industries with Lakebase and Agentic AI
Databricks · 27 aug. 2026
How Do Linear Probes Emerge? A Circuit-Tracing Framework with Concept-Targeted Attribution
arXiv · 31 aug. 2026
Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Models
arXiv · 16 sept. 2026
Planning or Improvisation? Stress-Testing the Poetry Planning Site on Open Models and Open Cross-Layer Transcoders
arXiv · 17 sept. 2026