OVHcloud launches first AI model family Olala, 7B model runs on 8GB GPU

OVHcloud, a cloud company in Europe, has launched its first AI model family, Olala. The first model is Olala mini 7A1B. It is open source under Apache 2.0 on Hugging Face and can run on a GPU with 8GB of memory.

The model uses a mix of Differential-TPA and Mamba-3 attention. It has 7 billion total parameters and 1 billion active parameters. It reaches 160 tokens per second on an RTX A4500.

It was trained on three public supercomputers in Europe with NVIDIA GH200, A100 and H100 chips. The total cost was 7 million euros. The first version has a 32K context. Version 1.1 is due by the end of the year with 128K context and better math skills. OVH plans Olala 2.0 with over 100B parameters.

SponsoredDomoAI
AI Tools Mentioned

Comments

Loading comments…