NTTDATA tsuzumi 2 Instruct (preview版)
durch 株式会社NTTデータ
NTTDATA tsuzumi 2 Instruct inference APIs and hosted fine-tuning in Azure AI Studio
【ご利用に関する注意事項】
本モデルは現在プレビュー版のため、動作保証しておりません。
また、利用後の返金は承ることができませんのであらかじめご了承ください。
「tsuzumi 2」は、NTTが開発した大規模言語モデルです。1GPUで動作可能な軽量モデルでありながら、超大型モデルに迫るトップクラスの性能と、高い日本語処理能力を備えています。
パラメータサイズが小さいため、学習やチューニング(※)に必要なコストを抑えられる点も特長です。さらに、金融・自治体・医療分野に関する知識を幅広く学習しています。
Microsoft Foundryとの統合により、セキュアかつスピーディな導入が実現可能です。
tsuzumi 2は、モデル専用の計算リソースを確保して利用するマネージド コンピューティングを採用しています。
これにより、トークン量を気にすることなく安定したスループットを維持でき、セキュアで一貫したパフォーマンスのAI環境を構築することが可能です。
※本モデルではチューニング機能のご利用はできません。チューニング機能が搭載されたモデルをご希望の場合は下記窓口までお問合せください。
nttd_tsuzumi@hml.nttdata.co.jp
【Important Notes Regarding Use】
This model is currently a preview version, so we do not guarantee its operation.
Please also note that we cannot accept refunds after use.
“tsuzumi 2” is a large-scale language model developed by NTT. It is a lightweight model capable of running on a single GPU, yet it delivers top-tier performance rivaling ultra-large models and possesses high Japanese language processing capabilities.
Its small parameter size is another key feature, enabling reduced costs for training and tuning*. Furthermore, it has extensively learned knowledge related to the financial, municipal government, and medical fields.
Integration with Microsoft Foundry enables secure and rapid deployment.
tsuzumi 2 employs managed computing, which ensures dedicated computational resources for the model.
This allows for stable throughput without worrying about token counts, enabling the construction of a secure AI environment with consistent performance.
*Tuning functionality is not available for this model. If you require a model with tuning capabilities, please contact the following support channel:
nttd_tsuzumi@hml.nttdata.co.jp
■提供事業者/責任者(Service provider/person responsible)
株式会社NTTデータ NTT DATA Japan Corporation
東京都江東区豊洲3-3-3 豊洲センタービル Toyosu Center Building, 3-3, Toyosu 3-chome, Koto-ku, Tokyo
インダストリーセールス事業部 Industrial Sales Division
050-5546-7240 nttd_tsuzumi@hml.nttdata.co.jp
坂本良平 Ryouhei Sakamoto