Japanese enterprises are increasingly adopting lightweight large language models (LLMs) to balance AI capabilities with cost and security considerations. NTT’s tsuzumi 2, a prominent example, achieves impressive performance on a single GPU, reducing infrastructure demands. Tokyo Online University has successfully deployed tsuzumi 2 for various applications, including course Q&A, teaching material creation, and student guidance. The single-GPU design addresses data privacy concerns and minimizes electricity costs, crucial for sustainable AI deployment.
Evaluations by NTT have demonstrated that tsuzumi 2 can match the performance of larger models in specific tasks, such as financial-system inquiry handling. Its reinforced knowledge in finance, medicine, and the public sector enables targeted domain-specific applications. The model’s inherent data sovereignty ensures data residency and strengthens security protocols, addressing a key concern for many organizations.
FUJIFILM Business Innovation is leveraging tsuzumi 2 with its REiLI technology to enhance document analysis capabilities. This integration enables the processing of sensitive corporate information without transmitting it externally, striking a balance between AI capability, security, and cost-effectiveness.