Reduced memory usage
Lower memory requirements allow for the generation of longer samples on devices with limited memory, like single GPUs or CPUs.
RecurrentGemma
A family of open models using a novel recurrent architecture for faster processing of long sequences.
Explore RecurrentGemma
RecurrentGemma is based on Griffin, a hybrid model architecture that mixes gated linear recurrences with local sliding window attention.
Capabilities
Higher throughput
Performs inference at significantly higher batch sizes. Capable of generating substantially more tokens per second — especially for long sequences.
High performance
RecurrentGemma matches Gemma's performance while requiring less memory and achieving faster inference.
Download RecurrentGemma
Hugging Face
