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