Sparse Autoencoders Python Packages

glassbox-mech-interp

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.

compresso-pytorch

A PyTorch Framework for Sparse Representation Learning

hypothesaes

HypotheSAEs: hypothesizing interpretable relationships in text datasets using sparse autoencoders. https://arxiv.org/abs/2502.04382

tsfm-lens

Controlled, causal and confirmatory interpretability for time-series foundation models (TimesFM, Chronos, Sundial, Timer, Time-MoE): one command, one report.

interp-lab

An open LLM interpretability tool that helps researchers and AI agents identify LLM activations and features and make Sparse Auto Encoders as needed.

tiny-dashboard

A tiny easily hackable implementation of a feature dashboard.

glassbox-mcp

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.

tsfm-benchmark

Controlled, causal and confirmatory interpretability for time-series foundation models (TimesFM, Chronos, Sundial, Timer, Time-MoE): one command, one report.

reptimeline

Track how discrete representations evolve during neural network training — lifecycle events, phase transitions, ontology discovery, and causal verification

hypersae

High-Performance Hyperbolic Sparse Autoencoders for Mechanistic Interpretability

crosswise

Reliability check for Natural Language Autoencoder explanations - catches fabricated claims against independent SAE evidence, 83% detection at 67% precision.