Skip to content

Gemma Scope

Inspect Gemma 2 activations with sparse autoencoders

Visit official site
Category AI models

What Gemma Scope is

The toolkit is described as a set of sparse autoencoders for Gemma 2 9B and 2B that acts like a microscope for internal model activations.

What you can do with it

  • Study activations across different model layers
  • Work with several sparse autoencoder variants
  • Follow Google Colab notebooks to learn the flow

Getting started

  • Find Gemma Scope on Hugging Face
  • Pick a Gemma 2 variant such as 2B or 9B
  • Read the technical report and blog posts

Pros & cons

✓ What we like

  • An interactive demo is offered with Neuronpedia
  • Blog posts and a report explain the approach

! What to watch out for

  • It targets the Gemma 2 9B and 2B variants
  • Getting started follows the Colab notebook tutorials

FAQ

Which models are covered?

The material names Gemma 2 9B and 2B variants.

How do I learn to use it?

Google Colab notebook tutorials, a technical report and blog posts are listed.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review