EN
As generative AI transforms the enterprise, it is becoming increasingly vital that organizations choose the right models for their business. Selecting the right model can mean the difference between providing correct, meaningful responses compared to incomplete or even wrong answers. When a consumer grade AI provides an erroneous response, it can be a source of comedy. For the enterprise, however, this can be catastrophic.
While choosing the right model is vitally important, a proliferation of new open source models are constantly being released and updated. This can make it difficult for developers to know which one to choose when building their applications. That is why SambaNova created Sambaverse.
Sambaverse is a unique environment where developers can freely test out hundreds of different models and directly compare their responses to prompts from a single endpoint. This will help ensure that they are able to choose the best available models for a given application.
Developers can choose from up to six models at a time, all from a curated list of some of the most popular models found on Hugging Face. They can then enter any prompt and each model will simultaneously run and provide a response. The prompt only has to be entered once to run across all the different models. With Sambaverse, developers can compare and contrast the speed and accuracy of each model’s response to help determine which one will best meet their needs.
The ability to try out multiple different models from a single endpoint is made possible by the Samba-1 Composition of Experts (CoE) model architecture. This architecture aggregates multiple, individual smaller models into a single large model. This large model delivers the benefits of both very large models and smaller models, without the drawbacks of either.
SambaNova is committed to supporting the developer community and promoting the development of responsible AI. We are furthering this commitment by making Sambaverse freely available to the developer community.
