Snowflake and Nvidia announced a partnership to help enterprises build customised generative AI (GenAI) applications using the enterprises’ own data without sharing any proprietary data. The collaboration, revealed at this week’s Snowflake Summit 2023, combines Snowflake Data Cloud’s data computing and security services with the large language model (LLM)–customisation capabilities of Nvidia’s NeMo platform.
“We want to be the platform of choice for building generative AI experiences, assistants, copilots and user apps,” said Snowflake SVP of Product Christian Kleinerman. “So think of Snowflake as a way to bring generative AI experiences and applications to run close to the enterprise data without having to ship enterprise data to a number of third-party endpoints.”
Data Cloud is an ecosystem of partners, customers, data providers and data service providers that can share data through the Snowflake platform using its Secure Data Sharing technology. It has nine workloads — united locations for high-volume data that allow users to run analytics and gain insights — including application, data engineering, data lake, collaboration, data science, data warehousing, marking analytics, unistore and cybersecurity.
Snowflake said it would host and run Nvidia’s NeMo platform in the Data Cloud, which enables users to use their proprietary data on the Snowflake platform — data that can range in scope from hundreds of terabytes to petabytes of raw and curated business information. The end product: tailored LLMs that power business-specific applications and services, including chatbots, summarisation and intelligent search.
Manuvir Das, VP of enterprise computing at Nvidia, highlighted the advantages of creating custom GenAI models endowed with the knowledge and skills specific to a company’s operations and based on its proprietary data. “That would be a better option for the company than just a generic foundational model.”
“The difference is the data that the company has that’s stored in the data lakes, the data warehouses” such as the Snowflake platform, he explained.
Snowflake taps Nvidia NeMo for LLM customisation
The NeMo LLM framework is a cloud-native software platform for building, customizing and deploying genAI and LLM models with billions of parameters. “NeMo is really an end-to-end platform for what we call model making; for creating custom models that enterprise companies can use,” said Das.
He noted that Nvidia and Snowflake’s engineering teams are working together to insert NeMo’s model-making engine into Data Cloud to empower enterprise customers to produce and utilise these custom models for various use cases in their business.
This integration will allow Snowflake customers to leverage foundation models and fine-tune them using their data stored in the Data Cloud or to start from scratch and train entirely new models. But the foundation models can provide a head start, enabling AI developers and data scientists to efficiently train custom models.
During this process, data can be extracted from the repository within the Snowflake Data Cloud, curated, prepared and used for training custom models. Once trained, these models can be stored within the Snowflake Data Cloud, enabling applications to access and utilise them seamlessly, Das explained.
The platform also offers the Nemo Guardrails software, ensuring users’ applications adhere to specific business topics, safety protocols and security requirements. “It’s a framework for helping practitioners control what models do to keep them within the parameters of what you would like the model to do,” he added.
Custom GenAI models for multiple verticals
As of April 2023, over 8,000 organisations worldwide are using Snowflake Data Cloud from industries spanning advertising, media and entertainment, financial services, health care and life sciences, manufacturing, retail and consumer-packaged goods, technology and telecom, according to the vendor.
And the Snowflake-Nvidia collaboration aims to transform these industries by bringing customised GenAI applications to different verticals to power business-specific applications and services.
For example, using custom GenAI, healthcare insurance models can answer complex questions about procedures covered under various plans, or financial services models can provide details about specific lending opportunities available to retail and business customers based on specific circumstances.
“Data is essential to creating generative AI applications that understand the complex operations and unique voice of every company,” Nvidia Founder and CEO Jensen Huang said. “Together, Nvidia and Snowflake will create an AI factory that helps enterprises turn their valuable data into custom generative AI models to power groundbreaking new applications — right from the cloud platform they use to run their businesses.”