---
library_name: gemma.cpp
license: gemma
extra_gated_heading: Access RecurrentGemma on Hugging Face
extra_gated_prompt: To access RecurrentGemma on Hugging Face, you’re required to review
  and agree to Google’s usage license. To do this, please ensure you’re logged-in
  to Hugging Face and click below. Requests are processed immediately.
extra_gated_button_content: Acknowledge license
---

# RecurrentGemma Model Card

**Model Page**: [RecurrentGemma]( https://ai.google.dev/gemma/docs/recurrentgemma/model_card)

This model card corresponds to the 2B instruction-tuned version of the RecurrentGemma model for gemma.cpp, the C++ implementation of the Gemma model family. You can also visit the model card of the [2B base model](https://huggingface.co/google/recurrentgemma-2b-sfp-cpp).

SFP checkpoints are pre-quantized to 8-bit using Switching Floating Point (SFP), a method that switches between eXmY floating point representations based on the values stored. gemma.cpp uses the [Highway](https://github.com/google/highway) SIMD library for optimal performance across most CPU architectures.

Visit [https://github.com/google/gemma.cpp](https://github.com/google/gemma.cpp) for more information and usage examples!

**Resources and technical documentation:**

*   [Responsible Generative AI Toolkit](https://ai.google.dev/responsible)
*   [RecurrentGemma on Kaggle](https://www.kaggle.com/models/google/recurrentgemma)

**Terms of Use:** [Terms](https://www.kaggle.com/models/google/recurrentgemma/license/consent/verify/huggingface?returnModelRepoId=google/recurrentgemma-2b-it-sfp-cpp)

**Authors:** Google