SambaNova Adds Enterprise-Grade GPT Language Modeling to its Dataflow-As-a-Service Platform 

With demands for language modeling soaring, SambaNova Systems has integrated an enterprise-ready Generative Pre-trained Transformer (GPT) language model into its flagship Dataflow-as-a-Service extensible AI services platform with the aim of making it easier for enterprises to add the needed services to their critical business workflows.

Presently, many enterprises that want to perform language model training rely on OpenAI's groundbreaking GPT-3 language modeling services, but GPT-3 services are in high demand and there can be a long wait to use them, limiting their availability, according to SambaNova.

That's where SambaNova came in to add GPT to its Dataflow-as-a-Service platform to expand the application's capabilities for customers. The value of the OpenAI's GPT-3 language modeling is that it uses deep learning to produce human-like text – like an advanced autocomplete program, according to SambaNova.

The addition of GPT to the dataflow platform is aimed at helping enterprises with mission-critical, text-based use cases from document classification to invoice automation, electronic health records management or sentiment analysis.

The release of the new GPT features for SambaNova's Dataflow-as-a-Service adds another option for enterprises that want to bring Natural Language Processing (NLP) into their production workflows. According to the company, customers will be able to be up and running with a customized language model within a month using their platform, compared to nine months or a year with other GPT services such as GPT-3.

Marshall Choy of SambaNova Systems

"Deploying large language models takes expertise and significant staffing resources, Marshall Choy, vice president of product for SambaNova, told EnterpriseAI. "SambaNova is further extending and differentiating its leadership position in large models and NLP with a complete GPT service offering."

The GPT-enabled platform offers ease of use and flexibility with low code interfaces such as API, GUI and command-line interfaces (CLI), while providing services to enterprise users on an as-needed subscription model, said Choy.

A GPT-3 Alternative for Enterprises

"This is a breakthrough after GPT-3 and an alternative to GPT-3," said Choy. "SambaNova Dataflow-as-a-Service GPT extends GPT-3 capabilities enabling customers to achieve higher levels of accuracy with domain-specific pre-training to deliver domain-specific task understanding – instead of more generalized capabilities." It is also a GPT-3 alternative through its enhanced accessibility for bringing large language model computing to customer data centers using a broad range of interfaces, he added.

"The wait list for GPT-3 access is a validation of the high level of interest in the ML community for large language models," said Choy. "We know of users who have had to wait for one year to gain access. The beta program [for the newly expanded platform] is advancing nicely and a couple of tasks, text completion and search have now come out of beta. Flexibility is also improving soon as we make available a pre-training option in addition to the current inference-only offering."

The new GPT-enabled SambaNova Dataflow-as-a-Service application also provides low-code and no-code development options inclusive of API/SDK access, in addition to GUI and CLI interfaces to perform both training and inference.

"SambaNova raises the level of abstraction and user interaction above the ML model with Dataflow-as-a-Service GPT," said Choy. "Instead of integrating infrastructure and then training, tuning, deploying and optimizing models using open source frameworks as the interface, Dataflow-as-a-Service interacts at the API level with the service, which thereby abstracts away all the complexities of infrastructure including the ML model itself."

A public dataset pre-trained model and needed infrastructure is provided, giving customers the ability to run pre-trained domain-specific language models that are then fine-tuned for their tasks, he said.

The company can provide the services to enterprise customers through a wide range of infrastructure configurations based on customer desired needs, said Choy. "Based on downstream task accuracy requirements, we scale model parameters and tokens. Based on time to train requirements, we scale the infrastructure to train the model from one to 16 or more racks."

Analysts on SambaNova GPT Capabilities

Matt Eastwood, IDC analyst

Matt Eastwood, an analyst with IDC, told EnterpriseAI that there is room in the marketplace for offerings like SambaNova's GPT services.

"The computing landscape continues to grow and also become more specialized with more and more customers looking to differentiate in the marketplace by using data analytics and AI to inform more prescriptive decision making in near real time," said Eastwood. "IDC believes that the fastest growing AI training server market is accelerated training at the edge and the fastest growing AI inferencing server market is accelerated inferencing in the cloud. By 2025, cloud represents 56.2 percent of the AI server infrastructure market, edge 18.6 percent and on-premises 25.2 percent."

What that means is the market will set up a set of dependencies with cloud service providers who will then be looking for ways to differentiate themselves while also offering customers choices, said Eastwood.

"The focus across enterprises and service providers alike will be in gaining an advantage through the enablement of more cost efficient and effective machine learning," he said. "To this end, GPT-3 will provide a degree of competition, but this market is large, and it is growing quickly. We believe there continues to be a significant amount of room for both innovation and differentiation."

SambaNova's GPT integration into its existing platform "provides a path to more quickly stand up AI based applications" for enterprises, said Eastwood. "Any technology that helps customers [create] a path from large language model capabilities towards automating ML models and infrastructure will be given a close look by organizations looking to continue their digital transformation via AI."

R. Ray Wang, analyst

Another analyst, R. Ray Wang, principal analyst and founder of Constellation Research, said SambaNova is addressing a need in the marketplace with its updated platform.

"We are moving from the craft stage of AI/ML language model creation to a phase where tools are more advanced and can remove the barriers of entry by providing hard-to-find resources without costly integration," said Wang. "The GPT language model allows companies to leap-frog existing efforts. There is massive demand among early adopters who no longer want manual approaches and for those new entrants who want to get into the market."

Rob Enderle, principal analyst with Enderle Group, agreed, calling the SambaNova product enhancement an "attractive, well-timed product" that will help meet the demand for robust natural language interfaces needed by enterprises.

"Natural language processing is still in its infancy, making this anyone’s game for now," said Enderle. "This market is unlikely to fully mature until a [commodity] AI is created and reaches critical mass, which is not expected until the 2040 timeframe."

The SambaNova GPT services will be interesting to watch, said Enderle. "On paper, it looks strong, but I would be more confident if I had customer references to look at and they were not part of this [news]."

Rob Enderle, analyst

GPT-3 has been garnering more headlines in the last few years.

In May, Microsoft announced that it was for the first time integrating the GPT-3 language generator into its Microsoft Power Apps software to make it easier for enterprise workers to build no-code applications. The action marked the first time that Microsoft is using OpenAI’s GPT-3 model within one of its own applications aimed at customers. The move came eight months after Microsoft licensed the GPT-3 natural language AI model from OpenAI in September of 2020.

In April, SambaNova announced that it had received $676 million in new Series D funding as it ramped up its efforts to take on Nvidia in the AI marketplace.