sparse-autoencoder-training

Train and analyze sparse autoencoders for neural network activations.

3|Updated Apr 4, 2026

Search Tags:#superposition #interpretability #sae #saelens #transformerlens #monosemanticity #feature-steering

Ready to run

npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill sparse-autoencoder-training-twjarviszyra-web

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: sparse-autoencoder-training
Source: https://github.com/twjarviszyra-web/hermes-unbound/tree/main/optional-skills/mlops/saelens
Command: npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill sparse-autoencoder-training-twjarviszyra-web

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill removes the friction of understanding what a language model represents by turning dense activations into sparse, interpretable features.

Core Features & Use Cases

Load pretrained Sparse Autoencoders for common model layers and inspect the features they activate on real prompts.
Train custom SAEs with configurable sparsity, architecture, and logging to study superposition and monosemanticity.
Analyze feature attribution, ablation, and steering to test how specific features influence predictions and behavior.
Use it for mechanistic interpretability research, safety analysis, and feature browsing with Neuronpedia.

Quick Start

Ask the skill to load a pretrained SAE for a target TransformerLens model and identify the top activating features for a prompt.

Frequently Asked Questions about sparse-autoencoder-training

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I train a sparse autoencoder for language model activations?▼

To train a sparse autoencoder for language model activations, configure sparsity, architecture, and logging parameters to study superposition and monosemanticity. This workflow supports custom SAE training integrated directly with TransformerLens.

What is monosemanticity analysis in neural network interpretability?▼

Monosemanticity analysis in neural network interpretability identifies sparse, interpretable features within dense model activations. By training sparse autoencoders, researchers isolate individual features to study how specific concepts are represented.

Can I load pretrained SAEs with TransformerLens and Neuronpedia?▼

Yes, you can load pretrained SAEs with TransformerLens and Neuronpedia. This capability allows you to load pretrained sparse autoencoders for common model layers, inspect features on real prompts, and browse them using Neuronpedia.

Does SAELens support feature steering and ablation workflows?▼

SAELens supports feature steering and ablation workflows through activation encoding and decoding. You can analyze feature attribution, ablate specific features, and test how steering influences model predictions and behavior.

What is the best way to analyze superposition in language models?▼

The best way to analyze superposition in language models is by training custom sparse autoencoders. Transforming dense activations into sparse features reveals how models represent more concepts than available neurons by overlapping representations.