circuit-discovery

Tracking exactly what happens to the internal "circuitry" (induction heads) of a 2-layer attention-only Transformer when forced to undergo domain adaptation from prose to structured Python code.

Python package for mechanistic interpretability providing statistical audits of activation patching for TransformerLens with False Discovery Rate (FDR) control, Benjamini-Hochberg correction, and sign-flip p-values.

Circuit-level regression testing for AI systems. Catch mechanistic drift that behavioral evals miss.

Open-source EU AI Act Annex IV documentation toolkit. Mechanistic interpretability + circuit discovery for transformers. One function call generates a structured, hash-chained evidence package.

A framework for identifying neural circuit components via convergent evidence across weight deltas, activations, and latent geometry.

Circuit-level interpretability evidence does not survive defensible analytic variation. Pre-registered multiverse over 15,840 specifications; the derived EU AI Act Annex IV claim flips across 73.2% of specification pairs. arXiv:2608.13754

Official code of the project "Query Circuits: Explaining How Language Models Answer User Prompts" accepted to ICML 2026

Official implementation of "How Do Linear Probes Emerge? A Circuit-Tracing Framework with Concept-Targeted Attribution" (EMNLP 2026)

Hierarchical Sparse Circuit Extraction from Billion-Parameter Language Models through Scalable Attribution Graph Decomposition

Mechanistic interpretability of subject-verb agreement in gpt2-small: zero/mean/resample ablation, ACDC circuit discovery, and causal scrubbing validation.

Mechanistic interpretability CLI for transformer models on Apple Silicon. Analyze per-layer predictions, monitor activation drift, compare models, discover circuits. MLX-based, no GPU needed.

Practical mechanistic interpretability tools — activation caching, linear probes, activation patching, circuit discovery, and visualization for transformer models

Mechanistic interpretability and safety auditing for Decision Transformers. Features circuit mapping, causal patching, TopK SAEs, and behavioral steering.