OVH announces Olala mini 7A1B

corporate.ovhcloud.com2 min read

OVHcloud has announced Olala mini 7A1B, its first family of compact 7-billion-parameter large language models built on a hybrid attention architecture. Trained using European public supercomputers with support equivalent to seven million euros in computing hours, the model will be available open source on Hugging Face under the Apache 2.0 license. Qwant has already adopted the model for its flash answer feature, reporting answers twenty percent faster and running costs divided by three. Open article →