A groundbreaking collaboration between Silo AI’s SiloGen and the University of Turku has given birth to ‘Poro’—a remarkable family of open-source large language models. These models, called LLMs, cover all official European languages and even coding languages.
A few days back, SiloGen announced the release of the initial model checkpoints for this multilingual LLM family. What makes this project stand out is its goal: ensuring European digital independence and making these language models accessible to everyone.
This initiative combines top-notch minds, access to huge computing power via Europe’s supercomputer LUMI, a vast amount of data, and special software to train these language models.
Just two months into the project, we’re thrilled to unveil the first checkpoint for Poro 34B. Named after the Finnish word for reindeer, this 34 billion-parameter model works wonders in English, Finnish, and code. A statement from Silo AI.
And guess what?
More languages and cool features are on the way! Future Poro versions will cover more European languages and come with enhanced capabilities.
But here’s the really awesome part: as one of the pioneers in this field, they’re sharing their progress. Through ‘Poro Research Checkpoints,’ they’ll give other researchers a peek into the model’s training journey. This move will help folks who don’t have the means to create such colossal models on their own.
Even though it’s only 30% trained, Poro is already outperforming existing models in the Finnish language test. And it’s not just about Finnish—Poro is set to level up in other languages too in the coming days.
So, what does this mean for you?
Imagine a language wizard that understands multiple languages and codes—Poro is on its way to becoming just that. And it’s not only smart but also open to sharing its secrets with the world. Now that’s something to be excited about.
