hn.today

OVH releases first model with EU public funding and supercomputer access

twitter.com15 points0 comments
Screenshot of OVH releases first model with EU public funding and supercomputer access

OVH has introduced Olala Mini 7A1B, the first model in its planned large-language-model family. Its 7-billion-parameter architecture activates about 1 billion parameters per request, using 256 experts per layer and a hybrid attention design combining Differential-TPA and Mamba-3. The sparse approach is intended to reduce memory use and increase speed: OVH says the model runs with as little as 8 GB of GPU memory, reaches 160 tokens per second on an RTX A4500, and can generate 50 tokens per second on a CPU. It currently supports a 32,000-token context window. OVH plans to release the model and its weights under the Apache 2.0 licence on Hugging Face, alongside a technical report and AI Act compliance documentation.

Training used three European public supercomputers - Jupiter, MeluXina and Jean Zay - with public computing support valued at €7 million. Qwant has already adopted Olala Mini for its quick-answer feature, reporting responses 20% faster and operating costs one-third those of its previous model. OVH’s planned version 1.1 will extend the context window to 128,000 tokens and improve mathematical reasoning. Version 2.0 is intended to produce specialized models for enterprise agents, cloud DevOps and cybersecurity, and is expected to exceed 100 billion parameters. OVH also plans for Olala to power Workspace models that can run locally on users’ computers.

Read on twitter.com0 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in AI

Human.md

Human.md

Human.md argues that people working alongside increasingly capable AI should bring intentional rule-breaking, unexpected inspiration, empathy, emotional resonance, and delight. It urges creators to experiment and ship many prototypes rather than rely on polished opinions or merely meet a brief. (human-md.ai)

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.