SambaNova launches one trillion parameter Generative AI model

“The entire AI industry is talking about building the smartphone of AI — an integrated hardware and software system — and today, SambaNova is the first to deliver a version of that to the enterprise,” said Rodrigo Liang, Co-founder and CEO of SambaNova Systems. “This past fall, we announced the SN40L, the smartest AI chip (rivaling Nvidia), and today we’ve integrated that chip with the first 1T parameter model for the enterprise. Samba-1 rivals GPT-4, however, it’s better suited for the enterprise as it can be delivered on-premises or in private clouds so that customers can fine-tune the model with their private data without ever disclosing it into the public domain.”

Available within SambaNova Suite™, Samba-1 includes a growing list of specialty AI models that are quick to deploy, manage and maintain. Along with SambaNova's SN40L chip that was recently announced, SambaNova now offers a fully optimized trillion parameter model that can be fine-tuned and deployed in private environments at 1/10th the hardware footprint, showing the true value of SambaNova’s full stack platform.

"SambaNova's release of Samba-1, a trillion parameter AI model, that integrates with their SN40L chip for improved performance, marks an important development in enterprise AI" said Matt Eastwood, SVP of Research, IDC. "Customers are looking for full stack solutions that include support for 1T parameter models. The integration of SambaNova's innovative silicon means that businesses can run on as little as one rack and drive much greater efficiencies than other similarly sized models."

“With Samba-1, enterprise customers of all sizes now have access to massive 1T parameter capabilities at the levels of simplicity and economics associated with significantly smaller models,” stated Liang. “With privacy and security at the forefront, SambaNova has taken Fortune 10 AI capabilities and made them accessible to all enterprises in less time and at a lower cost.”

Samba-1 models have been trained across a variety of different use cases, tasks, and languages, and all work together as a single Composition of Experts (CoE) to solve business problems. These experts are built on top of:

• Llama2 7B / 13B / 70B

• Mistral 7B

• DeepSeek Coder 1.3B / 6.7B / 33B

• Falcon 40B

• DePlot

• CLIP

• Llava

“Samba-1 is suited for enterprise clients that require a full stack AI solution, based on open standards, that they can deploy and see value from quickly,” said Senthil Ramani, Global Lead, Data & AI, Accenture. “SambaNova followed the full-stack approach from the beginning, which meant different design points and engineering decisions were made. The result is a privatized AI platform that is efficient, sustainable, cost effective, and secure.”

There are many enterprise benefits of Samba-1. They include:

• Enterprise-Ready: A private, secure 1+ trillion parameter model that provides the highest-level of accuracy and enterprise control.

• Better Performance and Accuracy: The Composition of Experts architecture aggregates multiple specialist models, which increases performance and accuracy while making fine-tuning modular.

• Lower cost and easy to manage: Deployed as a single endpoint, Samba-1 simplifies management and cost.

• Inference at 10X less cost: The inference cost is 10x less than large, monolithic models

• Control access to data: Controlled access to expert models in the same way you control access to all your data.

• Own the model: Customers own their model and fine-tune it within their own environment, with their own data. The model is built using open standards to eliminate vendor lock in.

• Fast and easy deployment: Deployment flexibility, as part of full-stack generative AI platform.