TORRENT - Throughput ORchestration for Resource-Efficient Neural Transformers

• Unité de recherche: INNOSUISSE
• Numéro de projet: 138.404 INNO-ICT
• Titre du projet anglais: TORRENT - Throughput ORchestration for Resource-Efficient Neural Transformers

Mots-clé (Anglais)

ICT, Artificial/local/extended/networked intelligence

Résumé des résultats (Abstract) (Anglais)

Klepsydra provides a highly efficient AI inference engine for edge AI based on a lock-free and asynchronous programming approach. This feasibility study investigates whether it can be further developed for server-based AI, to increase throughput, energy and operational efficiency of data centres.