Gemma 2: Improving Open Language Models at a Practical Size
dev.to·13h·
Discuss: DEV
🌱Tiny Languages
Preview
Report Post

Meet Gemma 2 — small models, big smarts

There’s a new family of language models that is made to run on more machines and give good results without giant cost. Gemma 2 comes in sizes from tiny to medium, so people and small teams can use them. The team changed how the models learn and talk to themselves so they get smarter for their size. For the smaller ones they used a way where the little model quietly learns from bigger models instead of just guessing the next word, and that helps a lot. The outcome is models that are faster, easier to run, and often match tools that are two or three times larger. You don’t need fancy gear to try them. Everything is shared openly so anyone can download and test. It’s a step toward making useful language tech available to more people,…

Similar Posts

Loading similar posts...