Gemma Scope Release
Google DeepMind announced Gemma Scope on July 31, 2024 as an open suite of sparse autoencoders for language-model interpretability, providing hundreds of freely available sparse autoencoders for Gemma 2 9B and 2B.
Historical classification
This verified record belongs to LINEAiGE's artificial-intelligence history layer beginning in 1956.
Why it matters
A dated release milestone for an open interpretability toolset built for layer-level study of Gemma 2 models.
People, organizations & technologies
ORGANIZATIONS
Google DeepMind
TECHNOLOGIES
Gemma Scope, Gemma 2
Evidence & verification
STATE
verified
CONFIDENCE
high
LAST VERIFIED
2026-09-25T21:02:00+08:00
Google DeepMind's dated announcement directly states that on July 31, 2024 it was announcing Gemma Scope and describes it as hundreds of freely available open sparse autoencoders for Gemma 2 9B and 2B. The current official Gemma Scope product page independently describes Gemma Scope as interpretability tools for Gemma models and says the original Gemma Scope enables layer-level analysis of Gemma 2. No development geography, ancestry, influence, causality, universal priority, or unsupported relationship edge is inferred.
Sources
Gemma Scope: helping the safety community shed light on the inner workings of language models
Google DeepMind
PUBLISHED
2024-07-31
Trace the chronology
Record identity
LINEAiGE IDgemma-scope-release-2024