AI BriefingGoogle Releases Gemma 2 2B and Safety/Interpretability Tools
·2024.07.31 09:00
Key point
Google has released Gemma 2 2B for on-device use, the safety classifier ShieldGemma, and the interpretability tool Gemma Scope.
Details
Just one month after the release of Gemma 2, Google has expanded its model lineup with the new releases of Gemma 2 2B, ShieldGemma, and Gemma Scope.
Key updates:
- Gemma 2 2B: A 2.6B parameter model optimized for on-device environments. It shares the same architecture as the existing 9B and 27B models, and incorporates sliding attention and logit soft-capping techniques.
- ShieldGemma: A series of safety classifiers trained based on Gemma 2. It helps developers filter application inputs and outputs to ensure safety.
- Gemma Scope: A family of sparse autoencoders for the Gemma 2 2B and 9B models, providing interpretability tools to understand the model's internal workings.
Gemma 2 2B can be easily used via Hugging Face's transformers library and llama.cpp, and weights are also provided in GGUF format.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.