Foundation models are trained on billions of parameters of data, but most of this data is general purpose and from the public domain. While useful in some scenarios, enterprises must often train these base foundation models on their own proprietary data, a step called “fine-tuning.” Tuning helps to maximize a model’s productivity in terms of overall accuracy for any specific use case. Given the potentially sensitive nature of this data and an organization’s data security standards, uploading proprietary data to a 3^rd party cloud can be time-consuming to achieve, if not prohibited altogether. This issue can be especially true for the financial, medical and government sectors. For this reason, if an enterprise has the necessary infrastructure, an on-premise deployment can be a better option when data privacy compliance and security are high priority requirements.

To help such enterprises, IBM has now released Mistral AI with IBM. As part of IBM’s commitment to making high-quality, performant foundation models widely accessible—and deployable on their own terms in secure environments—Mistral AI with IBM is an IBM watsonx.ai^™ add-on that enables customers to deploy and benefit from Mistral AI models on premises. Mistral AI’s model portfolio represents some of the most powerful models in the market with Mistral Large 2, Mistral AI’s latest general-purpose flagship model, currently available on IBM watsonx.ai. With a tuned version of Mistral Large 2, enterprises can take advantage of what this model does very well, which is to deliver large model performance at a relatively smaller scale and optimized to handle complex enterprise tasks such as RAG, non-English languages, coding (80+ coding languages) and function calling with context window support up to 128k tokens.

With Mistral AI’s flagship model now deployable in both SaaS and on-premises environments, IBM is further extending its open, multi-model strategy. Customers globally and especially in Europe can benefit even more from the combination of IBM, Mistral AI and the watsonx platform in building, deploying and scaling their generative AI use cases in an efficient, flexible and responsible manner.