Newshot

European cloud provider OVH enters AI models with open-source release of Olala mini 7A1B

AI summary

European cloud services company OVHcloud today announced its first AI model family, Olala, along with its first member, Olala mini 7A1B. Olala mini 7A1B uses a hybrid attention mechanism combining Differential-TPA and Mamba-3, with 7 billion total parameters and 1 billion active parameters, in BF16 data format. The model will be released open source under the Apache 2.0 license on Hugging Face. It can run on GPUs with 8GB of VRAM, achieving 160 TPS token throughput on an RTX A4500 and 50 TPS on CPU-only platforms. OVH said Olala mini 7A1B was trained on three European public supercomputers using NVIDIA GH200, A100, and H100, at a total cost of 7 million euros. The initial version of Olala mini 7A1B has a 32K context length, and version 1.1, expected by the end of this year, will expand to 128K context and offer better math capabilities. OVH will continue developing AI models, with Olala 2.0 to have a parameter scale of >100B.

Translation(model translation, not the source's own text)

European Cloud Provider OVH Enters AI Models with Open-Source Release of Olala mini 7A1B

IT Home, November 17 news: European cloud service provider OVHcloud announced today local time the launch of its first AI model family, Olala, and simultaneously unveiled the first member of the series, Olala mini 7A1B.

Olala mini 7A1B adopts a hybrid attention mechanism combining Differential-TPA and Mamba-3, with 7 billion total parameters and 1 billion activated parameters, using the BF16 data format. The model will be released as open source under the Apache 2.0 license on Hugging Face.

It can run on GPUs with 8GB of VRAM, achieving a token throughput of 160 TPS on an RTX A4500, and can reach 50 TPS even on a CPU-only platform.

OVH stated that Olala mini 7A1B was trained on three European public supercomputers using NVIDIA's GH200, A100, and H100, with a total cost of 7 million euros (IT Home note: approximately 51.428 million RMB at current exchange rates).

The initial version of the Olala mini 7A1B model has a context length of 32K, and version 1.1, expected to be released by the end of this year, will expand to 128K context and feature better mathematical capabilities. OVH will continue to develop AI models, and Olala 2.0 will have a parameter scale of >100B.

Related Articles

  • Unitree 王兴兴 says robot model generalization is the most thorny issue right now: hopes robots can immediately work in millions of households without retraining
  • AI large model weekly rankings: Google Gemini 4 Carbon makes a strong debut at the top, domestic models firmly hold the top five spots on the video chart
  • Vertical large model for international Chinese language education unveiled, providing intelligent services to global Chinese learners and other groups
  • No GPU needed: enthusiast runs 2.8 trillion parameter Kimi K3 AI model on discontinued Optane memory, outputting 1.5 tokens per second
  • Google releases Gemini 4 Hydrogen model: focused on everyday programming and agent workflows, $1.5 per million Token input
  • Baidu 李彦宏 launches "Token Exchange" for large models: claims to be neutral and trustworthy, scores model providers

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗