skills.rest
sheawinkler
sheawinklerCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

sparse-autoencoder-training

Train sparse autoencoders on transformer activations using SAELens configurations.

78|16|Updated Apr 23, 2026
Search Tags:#interpretability#feature-discovery#train#mechanistic-interpretability#sae#saelens#transformer-lens
One-click install
npx skills add https://github.com/sheawinkler/hermes-agent-ultra --skill sparse-autoencoder-training-sheawinkler
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sparse-autoencoder-training
Source: https://github.com/sheawinkler/hermes-agent-ultra/tree/main/optional-skills/mlops/saelens
Command: npx skills add https://github.com/sheawinkler/hermes-agent-ultra --skill sparse-autoencoder-training-sheawinkler

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

SAELens training enables researchers to train and analyze Sparse Autoencoders (SAEs) to decompose neural activations into sparse, interpretable features, supporting mechanistic interpretability.

Core Features & Use Cases

  • Train SAEs on transformer activations to learn sparse representations
  • Analyze feature activations, reconstruction quality, and interpretability metrics
  • Apply SAEs for feature discovery, ablations, and steering guidance in research workflows
  • Reference SAELens workflows and tutorials to accelerate experimentation
  • Study monosemantic representations in language models by identifying interpretable features

Quick Start

Run the training workflow on a GPT-2 activation stream to extract and evaluate sparse features.

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 sparse autoencoders to extract interpretable features from transformer activations?▼

Train sparse autoencoders on transformer activations to decompose neural network internal states into sparse, monosemantic, and interpretable features for mechanistic interpretability research.

What is sparse autoencoder feature discovery in language models?▼

Sparse autoencoder feature discovery is the process of decomposing neural activations into sparse representations, allowing researchers to identify monosemantic interpretable features within transformer language models.

Can I use SAELens configurations to train sparse autoencoders on GPT-2 activations?▼

Yes, you can use SAELens-based training configurations to train sparse autoencoders on GPT-2 activation streams to extract and evaluate interpretable sparse features.

What is the best way to analyze reconstruction quality and feature activations in mechanistic interpretability?▼

The best way to analyze reconstruction quality is to train sparse autoencoders using SAELens workflows to evaluate feature activations and interpretability metrics across transformer models.

Does sparse autoencoder training work for feature ablations and steering guidance in research workflows?▼

Yes, sparse autoencoder training supports applying learned features for feature ablations and steering guidance, enabling deeper mechanistic interpretability analysis within transformer research workflows.

Related Skills

Explore More Data & Analytics Skills→
GregsGreyCodeGregsGreyCode
6

sparse-autoencoder-training

Train and analyze sparse autoencoders on transformer activations using SAELens.

Community
Advanced
SignmanalSignmanal
1

sparse-autoencoder-training

Decompose polysemantic transformer activations into sparse interpretable features with SAELens.

Community
Advanced
nelohenriqnelohenriq
1

sparse-autoencoder-training

Automate Sparse Autoencoder training and analysis for transformer models using SAELens.

Community
Advanced
ragnarokhaaragnarokhaa
4

sparse-autoencoder-training

Train Sparse Autoencoders to decompose neural network activations into interpretable features.

Community
Advanced