2 min Applications

Databricks moves to focus more on custom AI with MosaicML acquisition

Databricks moves to focus more on custom AI with MosaicML acquisition

Databricks is set to acquire MosaicML for $1.3 billion. In doing so, it has acquired a promising player in the AI market.

Databricks hopes to complete the deal in July and move all 64 MosaicML employees into its offices. Both companies are located in San Francisco.

MosaicML helps customers run AI models on modest hardware. Its proprietary MPT-7B architecture powers MosaicML models with 7 billion parameters and a context window of 64,000 tokens. In addition, it recently released MPT-30B, which is significantly more powerful with 30 billion parameters and stronger than OpenAI’s GPT-3.

Ownership and security

The AI hype has been going on for some time, with ChatGPT as the initial driver. By now, many companies have the desire to use generative AI for their own purposes, from processing customer data to optimizing production processes. However, privacy and security are pain points here, as AI models largely derive their power from relevant data sets.

Databricks has been dealing with this problem for some time and a few months ago unveiled Dolly 2.0, an AI model that could generate very impressive results for companies using high-quality datasets. This while containing “only” 6 billion parameters. With this methodology, companies can self-assemble a set of data that is relevant to them to maintain ownership over the deployment of AI and handle the data in question safely and carefully.

Faster and cheaper

Thus, the areas of interest of Databricks and MosaicML hardly diverge. MosaicML, however, stands out because of its own Composer application. Indeed, this innovation promises that training neural networks with it is up to seven times faster than without it.

In short: the addition of MosaicML will in all likelihood fit perfectly with the mission that Databricks has envisioned for itself. So now it’s just a matter of waiting until the deal is formally closed.

Also read: Databricks expands DeltaSharing ecosystem