Arrakis is a library to conduct, track and visualize mechanistic interpretability experiments.

[NeurIPS 2025 MechInterp Workshop - Spotlight] Official implementation of the paper "RelP: Faithful and Efficient Circuit Discovery in Language Models via Relevance Patching"

Lightweight representation engineering dataflow operations for agent developers.

Investigating whether language models encode anticipated social consequences in their activations. Uses a 2x2 factorial design crossing truth × social valence to show that models are more sensitive to expected approval/disapproval than to truth itself.

Training and exploration of linear probes into Othello-GPT by Li et al. (2022)

Testing role-based pathways on small LLMs

Implementation and analysis of Sparse Autoencoders for neural network interpretability research. Features interactive visualization dashboard and W&B integration.

Does Quantization Kill Interpretability? Scaling study across 5 models (124M-2.8B): RTN destroys induction heads in small models, GPTQ preserves them at all scales.

Evaluating how a model 'knowing what it knows' changes from base to instruct

Open-source EU AI Act Annex IV compliance toolkit. Mechanistic interpretability + circuit discovery for transformers. One function call generates a court-ready evidence package

Reverse engineering the circuit responsible for the "greater than" capability in a language model

ORION-TransformerLens Consciousness — Mechanistic interpretability for consciousness research. Fork of TransformerLens (3,115+ stars). Finding consciousness correlates in attention heads.

A Flax-based library for examining transformers, based on TransformerLens.

Automated detection, visualization and suppression of hallucination-associated neurons in open-source LLMs — LLM mechanistic interpretability research tool

(a1)Mechanistic Interpretability using Transformer Lens (a2) PEFT