Databricks invests in Mistral and brings its AI models to data intelligence platform

Today, Databricks, the fast growing data infrastructure company led by Ali Ghodsi, announced a partnership with Paris-based Mistral, the well-funded startup that’s made waves in the global AI community with its growing family of highly-performant large language models (LLMs) — many of them, open sourced.

Under the engagement, Databricks is investing an undisclosed amount in Mistral, adding to its series A round, and bringing select Mistral LLMs to its data intelligence platform.

The move will see direct integration of the models, making it easier for enterprise users to use them with their data for generative AI applications — without any change to the security, privacy, and governance the Databricks platform already offers.

The development marks the addition of another notable distribution partner for Mistral, which has been moving forward aggressively through industry partnerships. Just recently, it announced a similar partnership with Snowflake, which competes directly with Databricks through its data cloud offerings, as well as Microsoft, the latter of which sparked criticisms of corporate capture and the interest of regulators.

Select models to be integrated natively

In a blog post published today, Databricks confirmed that the partnership with Mistral will result in the native integration of two text-generation models from the company – Mistral 7B and Mixtral 8x7B, both open source.

The former is a small 7 billion parameter transformer model, trained with 8k context length, and is very efficient to serve. Meanwhile, the latter is a sparse mixture of expert models (SMoE), supporting a context length of 32k, and capable of handling English, French, Italian, German, and Spanish. Mixtral 8x7B even outperforms Meta’s Llama 2 70B (from which it was trained on) and OpenAI’s GPT-3.5 across multiple benchmarks, including GSM-8K and MMLU, while boasting faster inference.

Users of the Databricks Data Intelligence Platform can now find both these models on the platform’s marketplace, complete with information about their capabilities and the different ways to use them.

Databricks says users can experiment with the models in the Mosaic AI Playground available through the platform console, use them as optimized model endpoints through Mosaic AI Model Serving or customize them with their proprietary data hosted on the platform (Mosaic AI Foundation Model Adaptation) to target a specific use case.

“Mistral AI models can now be consumed and customized in a variety of ways on Databricks, which offers the most comprehensive set of tools for building, testing and deploying end-to-end generative AI applications. Whether starting with a side-by-side comparison of pre-trained models or consuming models through pay-per-tokens there are several options for getting started quickly.  

While the news is great for Databricks customers looking to build gen AI apps and solutions with their data assets on the platform, it is important to note that the partnership has no mention of Mistral Large, the latest model from Mistral that sits just behind GPT-4 and outperforms Anthropic’s Claude 2, Google’s Gemini Pro and GPT-3.5 with native proficiency across five languages and a context window of 32K tokens. Snowflake has integrated this model into its data cloud’s Cortex service, along with Mixtral 8x7B and Mistral 7B.

When reached out by VentureBeat, a Databricks spokesperson said there is nothing to share on the integration of Mistral Large just yet. Other open-source, commercially usable models offered by Databricks are Meta’s Llama-2, CodeLlama, Stable Diffusion XL and Mosaic’s MPT family.

Mistral continues its partnership spree

Databricks and Snowflake are not the only partners for Mistral.

The company, which raised Europe’s largest-ever seed round in June 2023 and soon followed it up with a massive series A, has focused heavily on industry engagement to expand its reach and cement its position as a vendor-trusted player in the AI category dominated by OpenAI, Anthropic and Google. 

Just a few weeks back, it secured a $16 million investment from Microsoft to add its models to the Azure cloud platform. The deal made Mistral only the second company after OpenAI to offer its models on the Microsoft platform. 

Then, it has also signed separate partnerships with IBM, making the Mistral 8x7B available on WatsonX, and with Perplexity and Amazon. It will be interesting to see what other partnerships the startup can secure to grow its prominence and drive AI use cases across sectors.