Italian AI startup iGenius has launched Colosseum 355B, a large language model built on Nvidia technology. The model aims to serve clients in highly regulated industries with specific data protection needs.
iGenius specializes in AI development for sectors like finance and public administration. Its AI solution, Crystal, is designed for these industries, leveraging the capabilities of its LLMs.
Colosseum 355B was created using the Nvidia AI Enterprise software platform and the DGX Cloud AI platform, which organizes over 3,000 Nvidia H100 GPUs. Nvidia reported that the development took two months. The model, optimized to fit on a single H100 GPU node, features 355 billion parameters and supports over 50 languages, excelling in coding tasks.
The model was pre-trained using a method known as FP8 precision, significantly reducing memory usage and cutting inference costs by 50% without compromising accuracy or quality.
Industries subject to strict regulations often approach AI technology with caution. iGenius plans to address this with its new language model, which can operate on a single H100 GPU, allowing organizations to run it within their own firewall without cloud hosting. This enables companies to host their proprietary LLM and maintain full control over the model.
According to iGenius CEO Uljan Sharka, the company is in contact with potential users in the financial services sector, heavy industry, and government. Sharka even flew to Brussels on Thursday to present the new model to the European Commission.