Gemma Scope 2: open suite of tools for language model interpretability(deepmind.google)

https://deepmind.google/blog/gemma-scope-2-helping-the-ai-safety-community-deepen-understanding-of-complex-language-model-behavior/