~/AI MODELS/european-cloud-provider-ovhcloud-releases-open-source-olala-mini-7a1b-ai-model

European Cloud Provider OVHcloud Releases Open-Source Olala mini 7A1B AI Model

European cloud provider OVHcloud announced its first AI model family, Olala, launching the initial Olala mini 7A1B model. The model features 7 billion total parameters, 1 billion active parameters, and a hybrid architecture combining Differential-TPA and Mamba-3. This release marks a significant entry by a major European cloud provider into the AI model space, emphasizing efficient open-source infrastructure trained on European public supercomputers. It offers developers a cost-effective, high-performance model that runs efficiently on consumer-grade hardware. Olala mini 7A1B uses the BF16 data format, supports an initial 32K context window, and is open-sourced under the Apache 2.0 license on Hugging Face. It can run on GPUs with just 8GB of VRAM, achieving a token throughput of 160 TPS on an RTX A4500 and 50 TPS on a CPU-only platform.

## BACKGROUND

OVHcloud trained this model across three European public supercomputers utilizing NVIDIA GH200, A100, and H100 hardware with a total cost of 7 million euros. The company plans to release version 1.1 with a 128K context window by the end of the year and aims to develop an Olala 2.0 model with over 100 billion parameters.

## KEYWORDS

#AI Models#Open Source#OVHcloud#Mamba#Machine Learning

$ subscribe --daily

European Cloud Provider OVHcloud Releases Open-Source Olala mini 7A1B AI Model | Daily News