From Verizon Media:

SambaNova, an AI chip startup, launches Samba-1, an AI system with 56 generative open source models for enterprise tasks like text rewriting and language translation. CEO Rodrigo Liang claims Samba-1 allows for easier fine-tuning and flexibility in addressing multiple AI use cases compared to other models like OpenAI.

Unlike single-model systems, Samba-1 lets customers control how requests are routed among the 56 models, reducing the cost of fine-tuning. The multi-model strategy allows for comparing responses to ensure reliability. Samba-1 can be deployed on-premises or in a hosted environment, collapsing training costs for tasks that require multiple models.

Though other vendors offer competitive pricing and routing tools for large generative models, SambaNova sells a set-it-and-forget-it package. Samba-1 offers enterprises a full-stack solution with custom models trained on private data, hosted on a single server rack at one-tenth the cost of alternatives. This packaged solution may appeal to some enterprises seeking simplicity and cost efficiency.



Read more at Verizon Media: SambaNova now offers a bundle of generative AI models