Google has announced the release of RecurrentGemma 9B, a new deep learning model scaled to 9 billion parameters. The model boasts significant improvements in performance, including more than 25% lower latency and 6-7 times higher tokens per second compared to its predecessor, Gemma. RecurrentGemma 9B is available in both base and instruct-tuned versions. It achieves strong performance metrics, with scores of 60.5 on MMLU, 73.2 on CommonSenseQA, and 39.3 on AGIEval. The model is based on the Griffin architecture, which combines linear recurrence with local attention, and offers faster inference, particularly for long sequences or large batch sizes. The model has been trained on 2 trillion tokens and is now available on platforms such as Hugging Face and Kaggle. For more details, refer to the Griffin paper.
🔥 Introducing our 9B language model, trained on 2 trillion tokens! 🚀 Based on Griffin (https://t.co/kL5TeAbmVV) and delivers: 💪 Powerful performance ⚡️ Lightning-fast inference Pretrained and instruction-tuned models now available on HF & Kaggle! Start building today! 🏗️ https://t.co/s8sjsO51Vi
RecurrentGemma-9B is out! https://t.co/rSTQn2SlhR https://t.co/Il5UudfpZk - Uses Griffin architecture, combining linear recurrence with local attention - Downstream evals comparable to Mistral and Gemma - Faster inference, especially for long sequences or large batch sizes 1/n https://t.co/l3MLNebAzq
RecurrentGemma 9B by Google is out 🔥 ⚡️Super fast for long sequences: Good throughput+latency 👀Base and instruct tuned versions 🏆Similar quality as Gemma Check the y-axis below 🤯 Models: https://t.co/Py7CImb6el Griffin paper: https://t.co/KhWcw0euaY https://t.co/4FNQePDTbb
Welcome RecurrentGemma 9B 🔥 > Same performance as Gemma with more than 25% lower latency and 6-7x higher tokens/ sec ⚡ > Base (9B) and Instruct (9B-IT) models released. > MMLU - 60.5, CommonSenseQA 73.2, AGIEval 39.3 - pretty strong base model to fine-tune further. > Based on… https://t.co/J3ctP4OSlU
📣 🧠 Exciting news for researchers pushing the boundaries of efficient deep learning! We've scaled RecurrentGemma to 9 billion parameters. 🧵↓