Upstage and Amazon Web Services (AWS) are collaborating to provide high-performance and efficient AI solutions for enterprise use to government and public institutions as well as regulated industries in the Asia-Pacific (APAC) region and the United States. Upstage announced on the 19th that it will work with AWS to develop and expand distribution channels for its language model “SOLAR.”
The company plans to use AWS infrastructure when building, training, and deploying foundation models in the future. By leveraging AWS’s machine learning infrastructure, such as Amazon SageMaker, AWS Trainium, and AWS Inferentia chips, Upstage aims to expand its large language model (LLM) “SOLAR” and its AI-based document processing solution suite.
Upstage has already been training the SOLAR model for several years using Amazon SageMaker. It claims to have significantly reduced training time through advanced data preprocessing and continuous pre-training techniques. Additionally, using AWS infrastructure, the company has implemented a self-developed deep expansion scaling (DUS) method, enabling more efficient model expansion and performance improvement.
The two companies aim to provide top-notch generative AI solutions optimized for the performance, security, and compliance needs of the government, public institutions, and regulated industries. Besides, Amazon has also made a minority equity investment in Upstage.
John Jones, AWS’s Global Vice President for Startups and Venture Capital, stated, “Upstage is elevating the use of AI by using AWS to help public sector stakeholders enhance efficiency and productivity while ensuring the safe protection of everyone’s data.”
In April, the two companies jointly launched an AI initiative to support the introduction of generative AI for educational institutions and non-profit organizations. Upstage provides free API access to its entire range of AI model products, while AWS offers free infrastructure support alongside tools such as AWS Skill Builder and the experimental generative AI sandbox, Party Rock. Over 200 institutions, including Seoul National University and KAIST, are participating in this program.