Real AI Won the ISCRA Project

At the Data Science Conference 2023 in Belgrade, Real AI won the ISCRA project to build Europe’s first Human-Centered Large Language Model (LLM) on the Leonardo AI Computer Cluster, the world’s fourth largest. This project, named HOMINIS, aims to create ethical, bias-free AI systems using open, sustainable datasets.

ASSEN – During the Data Science Conference 2023 in Belgrade on Thursday, 24 November, it was announced that Real AI won the ISCRA project. Real AI is chosen to build Europe’s first-ever Human-Centered LLM on the world’s 4th largest AI Computer Cluster ‘LEONARDO’.

LEONARDO – the World’s 4the Largest AI Cluster

The Leonardo supercomputer, located at the CINECA data center in Bologna, is a high-performance computing powerhouse. It’s built around the Atos BullSequana XH2000 computer system, incorporating nearly 14,000 Nvidia Ampere GPUs. Leonardo ranks as the fourth fastest supercomputer in the world and second in Europe as of its inauguration in November 2022. This technical prowess positions Leonardo as a crucial asset for advancing AI applications across Europe  .

Recommended : Key Insights Into The Utility Of AI In The College Application Process

REAL AI’s unique opportunity-

“I’m thrilled to spearhead Real AI’s mission with Project HOMINIS, to architect a human-centric AI that resonates with our ethos. The immense capabilities of the Leonardo AI compute cluster, millions of GPU hours, coupled with the strength of our partnerships, set the stage for us to craft the world’s inaugural sustainable, human-centered LLM.” – Tarry Singh, CEO, Real AI B.V.

Real AI will build the first ever Human-Centered LLM (large language model) by allocating substantial computing power in millions of GPU hours from Leonardo, to significantly accelerate the development of a LLM from scratch. This ambitious project positions Real AI at the forefront of AI development, committed to environmental stewardship and technological innovation.

Project HOMINIS –

Human-centered Open-source Model for Intelligent, Neutral and Inclusive Systems. Project HOMINIS aims to revolutionize AI by creating ethical, bias-free AI systems using open, inspectable datasets. Addressing toxicity in web-scale datasets, it proposes sustainable alternatives to AI foundation models, reducing environmental impact. The project’s four primary goals include: 1) Compiling high-value, diverse datasets from various sources, including scientific papers and knowledge bases, while eliminating biased content; 2) Conducting extensive ablation studies on transformer models and exploring alternative architectures; 3) Optimizing the foundation model using advanced techniques, leading to a preliminary release for community collaboration; 4) Implementing instruction tuning for ethical alignment, culminating in the final release of a responsibly trained AI model. Additionally, HOMINIS seeks to reduce energy consumption using innovative methods like Flash Attention and routing, enhancing data processing efficiency, and improving model interference and knowledge integration.

Real AI’s partnerships with UNINA AND NVIDIA –

UNINA, The University of Naples Federico II, has a notable record in significant initiatives. For example, the first Apple iOS Developer Academy in 2016, the Cisco Digital. Transformation Lab in 2018 and most recently (together with Real AI) is a founding member of Europe’s first Human-Centered AI Masters.

NVIDIA is the world’s largest AI company, having transformed the GPU into a powerful computing brain that intersects with virtual reality, high-performance computing, and AI.

Latest HRtech Interview Insights : HRTech Interview With Lavonne Monroe, VP Of Global Talent Acquisition And Onboarding At HPE