How adapting models brings generative AI to the enterprise

There’s a lot of hype and activity around generative AI models, especially in the short time since ChatGPT first launched.

ChatGPT – and the GPT-3 large language model (LLM) on which it is based – are trained on public data, which serves as an excellent foundation for consumer applications, but lacks the customization, privacy, and security that an enterprise needs . That’s where Rodrigo Liang, co-founder and CEO of SambaNova Systemswants to make a difference with today’s launch of its company’s SambaNova Suite, which aims to help enterprises build and deploy custom generative AI models.

>>Follow VentureBeat’s ongoing generative AI coverage<

SambaNova started in 2017 and was largely focused on hardware for AI, raising a whopping $676 million in April 2021 to support its efforts. In recent years, the company has expanded its initial focus on hardware to also build out support for both machine learning training and inference on various models, with its dataflow-as-a-service offering. The new SambaNova Suite extends the dataflow service with a collection of capabilities that enable organizations to tailor both open-source and proprietary generative AI models to their specific requirements.

“The focus of SambaNova is how to bring generative AI capabilities more to the enterprise,” Liang told VentureBeat. “There are certain things that businesses need to be successful, and that’s what we’re doing for them today.”

Nvidia isn’t the only AI hardware vendor for generative AI

There is a growing list of vendors building on generative AI models to enable enterprise use cases.

Content creation is a particularly vibrant field of generative AI customization for enterprises. Jasper AIfor example, recently announced its Jasper for Business offering designed to tailor AI to a specific business. Font style emerged from stealth Monday with its enterprise content creation platform for generative AI, alongside $65 million in funding. Quantive last week announced its foray into generative AI to help organizations with their business strategy.

SambaNova’s main differentiator from others in the generative AI space is that it has its own hardware to help optimize business use cases. Rather than relying solely on Nvidia GPUs like many in the industry, SambaNova has developed its own custom silicon optimized for both machine learning training and inference.

“What we’ve done is … take the AI ​​software stack that people really want to use with PyTorch, TensorFlow and complex models like GPT, and we’re taking it all the way back to the silicon,” Liang said. “We have silicon that’s actually tailored to run these big models for the enterprise, rather than the other way around.”

The team behind SambaNova, including Liang, has a background in building microprocessors for Sun Microsystems and Oracle. Liang said SambaNova has created a microprocessor for AI that is extremely power efficient and performative to run these generative AI applications for enterprises.

Liang emphasized that the custom silicon also enables continuous training and inference capability so that the data that feeds the generative AI models can be kept up to date.

“In business you need real-time information and so you don’t want your models to really lag behind,” he said.

Privacy, responsible AI and the enterprise

With the new SambaNova Suite, Liang said his company is looking to solve the key challenges enterprises face with generative AI. Those challenges include customization for a company’s specific data, as well as the ability to mitigate bias and provide responsible and explainable AI.

With its platform, SambaNova enables an organization to run customized training in a private environment on any data the organization has, including unstructured data that can be found in a company’s Slack discussion channels. Liang took it a step further, saying the platform provides organizations with transparency about how a given AI model actually works.

“SambaNova is built to show you exactly how the model arrived at a certain conclusion,” Liang said. “We store all the processes around how we train and refine the model so that when an auditor comes or someone wants to check for bias or why something happened a certain way, you can actually work through the flow and verify that your results are correct. has been done.”