Thursday, April 4, 2024

Viking AI: "Silo AI launches Nordic LLM family to protect Europe’s linguistic diversity"

From The Next Web, April 4:

The open models aim to protect Europe’s AI sovereignty  

Europe’s linguistic diversity has received a boost into the AI era with the launch of new LLMs for Nordic languages.

Fittingly named Viking, the model family is the brainchild of Finnish startup Silo AI. The company claims to host Europe’s largest private AI lab and has pledged to protect the continent’s digital sovereignty.

Viking is the latest step of that mission. By integrating Danish, Finnish, Icelandic, Norwegian, and Swedish into its open-source models, Silo AI plans to expand digital inclusivity.

Peter Sarlin, the startup’s CEO and co-founder, wants the LLMs to “bridge gaps” in digital communication.

“We aim to cater to the nuances of these low-resource languages and ensure that LLMs reflect the values and culture of the Nordic people using them,” Sarlin told TNW.

All five Viking languages are low-resource, meaning they lack the linguistic data needed to train high-performing LLMs. That threatens their future in the digital world — and the physical world too.

Without LLMs in these languages, communities could be deprived of countless services, from machine translation to personalised healthcare.

Businesses would also lose access to potential markets. And as AI spreads deeper into our lives, linguistic diversity and cultural heritage could disappear.

Silo AI’s solution to this problem focuses on Europe.

Nordic languages and Europe’s digital sovereignty
Before launching Viking, Silo AI earned plaudits for Poro, an open model for English, Finnish, and programming languages.

Viking expands the repertoire. The new suite of models was built to master the low-resource Nordic languages, without compromising the high-resource English language. 

According to the startup’s evaluations, Viking offers “unparalleled proficiency” in Nordic languages.

Additional tests suggest the family’s English performance is on par with other open models, such as Mistral and Meta’s Llama. Viking also reputedly “excels” in processing multiple languages and coding....

....MUCH MORE

And, if you are wondering, Silo AI was developed and trained on the EU's non-NVDA-accelerated, Lumi supercomputer in Finland which we've visited a few times in the context of the Top 500 list of supercomputers and back in 2021 in "Finland May Soon Be Home To The Fastest Supercomputer In The World".

As of the most recent Top 500 list (November 2023) Lumi is the world's fifth fastest supercomputer:

....The (again) upgraded LUMI system, another HPE Cray EX system installed at EuroHPC center at CSC in Finland is now the No. 5 with a performance of 380 Pflop/s. The European High-Performance Computing Joint Undertaking (EuroHPC JU) is pooling European resources to develop top-of-the-range Exascale supercomputers for processing big data. One of the pan-European pre-Exascale supercomputers, LUMI, is located in CSC’s data center in Kajaani, Finland.....