Oracle has announced a new Generative AI service for Oracle Cloud Infrastructure, designed to allow companies to integrate Artificial Intelligence into their line of business applications. So, OCI Generative AI is a managed service built on Oracle Cloud infrastructure in collaboration with the enterprise AI platform Cohere.
Cohere’s managed service and models will work in conjunction with AI Vector Search, a feature of Oracle Database 23c that delivers Retrieval Augmented Generation (RAG), an AI technique that combines large pre-trained language models and enterprise data to deliver answers with a higher level of precision.
OCI Generative AI will also be the foundation for generative AI capabilities integrated into Oracle’s software-as-a-service application suite. These include the Oracle Fusion Cloud and Oracle NetSuite application suite, as well as applications for industry, such as Oracle Cerner.
On the other hand, Oracle has announced the upcoming availability of OCI instances powered by Ampere’s Nvidia H100 Tensor Core, Nvidia L40S, and AmpereOne GPUs. The new OCI Compute instances are designed to run a variety of cloud workloads, making them more accessible to enterprises.
OCI’s generative AI service has three models: Command, which takes a user’s prompt and generates text; Summarize, which performs extractive summaries based on user parameters; and Embed, which translates text into numerical vectors that models can understand. Oracle has also made several improvements to services it already had in place.
Thus, in Oracle Digital Assistant it has added generative AI functions to allow the integration of large language models and other generative capabilities in the assistants. In OCI Language, on the one hand it has added health information (Healthcare NLP) with natural language processing for functions such as clinical trial notes, patient progress and electronic health records.
In addition, it has added a document translation function (Document Translation Experience) with formats such as Word, PowerPoint, HTML, JSON or Excel. OCI Vision has the ability to recognize faces and parts of the face in images. OCI Speech now allows the service to integrate speaker information into transcribed sections of the audio, and OCI Data Science adds a central repository for managing features developed by data science teams: Feature Store.
OCI Compute instances with Nvidia GPUs will include bare metal instances powered by H100 and L40S GPUs, which can help reduce the time it takes to train large AI models if you have those powered by the H100 GPU. Those with the L40s are appropriate for training medium and small models.
Likewise, Oracle has confirmed that there will be A2 instances of OCI Compute driven by the AmpereOne CPUs with up to 320 cores in the bare metal configuration, and with up to 156 cores in the flexible virtual machine instances. In both cases, they will be available throughout 2024.