OVH has introduced Olala mini, a 7-billion-parameter AI model designed to run on GPUs with as little as 8GB of memory.

Cloud provider OVH has introduced Olala, a new family of large language models (LLMs), beginning with Olala mini 7A1B. The compact model is built on a 7-billion-parameter architecture and is designed to run on GPUs with as little as 8GB of memory.

Olala mini uses one billion active parameters distributed across 256 experts per layer, alongside a hybrid attention architecture combining Differential-TPA and Mamba-3. The model uses BF16 format and activates only part of its architecture for each request, reducing its computing and memory requirements.

The model was trained using three European public supercomputers: Jupiter, MeluXina, and Jean Zay. The training effort involved computing resources equivalent to €7 million in computing hours.

OVH plans to officially launch Olala mini at the OVHcloud AI Summit on 19 November, releasing its model weights, pre-trained base model, technical report, and documentation covering compliance with the European Union’s AI Act.

Octave Klaba, CEO of OVHcloud, said: “AI is strategic for OVH. As well as offering AI services to our customers, we must be able to create models for our own internal needs and for those of our customers.”

The company’s development roadmap will see Version 1.1 extending the context window to 128,000 tokens and improving mathematical reasoning. A subsequent version 2.0 is intended to support the development of agentic AI for businesses and cybersecurity.

Stay updated on crypto and AI by following our socials

Leave a Reply

Your email address will not be published. Required fields are marked *

Instagram