Multiverse Computing, a Spanish AI startup, has secured $215 million in a Series B funding round to enhance its CompactifAI technology, which aims to compress large language models to fit on smartphones. This method reportedly reduces parameter count by 70% and model memory by 93%, while maintaining 97-98% of the original model's accuracy. The compacted version of the Llama-2 7B model is claimed to run 25% faster at inference and is significantly cheaper to operate. Investors included Bullhound Capital, HP Tech Ventures, and Toshiba. The startup's technology employs tensor networks from quantum physics to optimize AI models, effectively reducing redundancies while preserving essential correlations. Additionally, Multiverse Computing is already serving over 100 clients, including Bosch and the Bank of Canada, and has plans to expand its offerings. This breakthrough could address the computational hurdles of massive AI models, making advanced AI applications more accessible on everyday devices.

Source 🔗