The Larger The Model, The Better

Composition of Experts

A Composition of Experts (CoE) model uniquely combines the capabilities of multiple, individual small models to create a single, incredibly powerful model of almost unlimited size. Delivering the performance and ease of training that is only possible with small models, the accuracy and multi-modal capabilities of the largest models, and access control not found in Mixture of Experts models, the SambaNova Suite Composition of Experts models are enabling generative AI in the enterprise.

Model Ownership

In the enterprise, models have become more than just a tool, they have become a valuable asset within any organization. Fine-tuned on the customers most valuable, internal data, these models uniquely understand the customer organization and, like other assets, will continue to grow in value over time. SambaNova customer always retain ownership of their models in perpetuity.

Samba-1

SambaNova provides the latest open source models, pre-trained on public data, to our customers. The models can then be fine tuned on their data for greater accuracy.

This offers them the most up-to-date capabilities and functionality, in the fastest time. Unlike closed models, this gives the customer visibility into the weights and training data to meet regulatory requirements.

Latest Models

Take advantage of the latest open source models including:

• Llama 2
• Mistral
• Falcon
• DeepSeek
• BLOOMChat
• LLaVA
• CLIP

Training

Any model is only as good as the data it is trained on. SambaNova uses the latest open source models which can be trained on customer data for the highest accuracy. With models based on a Composition of Experts model, each expert model can be further refined as necessary, dramatically increasing the accuracy of the system.