Today, Datastonesthe fast-growing data infrastructure company led by Ali Ghodsi, announced a partnership with the Paris-based company mistralthe well-funded startup that has made waves in the worldwide AI community with its growing family of high-performance large language models (LLMs) – lots of them open source.
As a part of the agreement, Databricks will invest an undisclosed amount in Mistral, expand its Series A round and convey select Mistral LLMs into its data intelligence platform.
The move will directly integrate the models, making it easier for enterprise users to make use of them with their data for generative AI applications – without changing the safety, privacy and governance that the Databricks platform already offers .
The development marks the addition of one other notable distribution partner for Mistral, which is moving aggressively through industry partnerships. Most recently, the corporate announced an analogous partnership with Snowflake, which directly competes with Databricks with its data cloud offerings, and with Microsoft, the latter of which drew criticism of the company takeover the interest of the regulatory authorities.
Select models to integrate natively
In a blog post published Today, Databricks confirmed that the partnership with Mistral will end in the native integration of two of the corporate's text generation models – Mistral 7B and Mixtral 8x7B, each open source.
The former is a small transformation model with 7 billion parameters, trained with a context length of 8KB, which could be very efficient to make use of. The latter is now a sparse mixture of expert models (SMoE) that supports a context length of 32 KB and might handle English, French, Italian, German and Spanish. Mixtral 8x7B even outperforms Meta's Llama 2 70B (on which it was trained) and OpenAI's GPT-3.5 in several benchmarks, including GSM-8K and MMLU, while providing faster inference.
Databricks Data Intelligence Platform users can now find each models within the platform's marketplace, complete with details about their capabilities and different uses.
According to Databricks, users can experiment with the models within the “Mosaic AI Playground” available through the platform console, use them as optimized model endpoints through “Mosaic AI Model Serving,” or use them with their proprietary data hosted on the platform (Mosaic AI Foundation Model Adaptation), adapting a selected use case as a goal.
“Mistral AI models can now be used and customised in quite a lot of ways on Databricks, which provides probably the most comprehensive set of tools for constructing, testing and deploying end-to-end generative AI applications. Whether you begin with a side-by-side comparison of pre-trained models or use models via pay-per-token, there are several ways to start quickly.
While the news is great for Databricks customers seeking to develop Gen AI apps and solutions with their data assets on the platform, it can be crucial to notice that the partnership doesn’t mention Mistral Large, Mistral's newest model , which is correct behind GPT. 4 and outperforms Anthropic's Claude 2, Google's Gemini Pro, and GPT-3.5 with native knowledge of 5 languages ​​and a context window of 32,000 tokens. Snowflake has integrated this model into the Cortex service of its data cloud, together with Mixtral 8x7B and Mistral 7B.
When contacted by VentureBeat, a Databricks spokesperson said there was nothing yet to say concerning the Mistral Large integration. Other Open source, commercially viable models Databricks offers Metas Llama-2, CodeLlama, Stable Diffusion XL and the MPT family from Mosaik.
Mistral continues its partnership
Databricks and Snowflake are usually not the one partners for Mistral.
The company, which raised Europe's largest ever seed round in June 2023 and a large Series A soon followed, has focused heavily on industry exposure to expand its reach and position as a vendor-trusted player within the dominated AI space -Category to consolidate by OpenAI, Anthropic and Google.
Just a couple of weeks ago, the corporate secured a $16 million investment from Microsoft so as to add its models to the Azure cloud platform. The deal makes Mistral only the second company after OpenAI to supply its models on the Microsoft platform.
Then it was signed too separate partnerships with IBMmaking the Mistral 8x7B available on WatsonX and with confusion and Amazon. It might be interesting to see what other partnerships the startup can secure to extend its exposure and advance AI use cases across industries.
