Databricks claims its open source foundational LLM outsmarts GPT-3.5

In the AI gold rush, analytics outfit wants to provide the shovels

Analytics platform Databricks has launched an open source foundational large language model, hoping enterprises will opt to use its tools to jump on the LLM bandwagon.

The biz, founded around Apache Spark, published a slew of benchmarks claiming its general-purpose LLM – dubbed DBRX – beat open source rivals on language understanding, programming, and math. The developer also claimed it beat OpenAI's proprietary GPT-3.5 across the same measures.

DBRX was developed by Mosaic AI, which Databricks acquired for $1.3 billion, and trained on Nvidia DGX Cloud. Databricks claims it optimized DBRX for efficiency with what it calls a mixture-of-experts (MoE) architecture – where multiple expert networks or learners divide up a problem.

Databricks explained that the model possesses 132 billion parameters, but only 36 billion are active on any one input.

Joel Minnick, Databricks marketing vice president, told The Register: "That is a big reason why the model is able to run as efficiently as it does, but also runs blazingly fast. In practical terms, if you use any kind of major chatbots that are out there today, you're probably used to waiting and watching the answer get generated. With DBRX it is near instantaneous."

But the performance of the model itself is not the point for Databricks. The biz is, after all, making DBRX available for free on GitHub and Hugging Face.

Databricks is hoping customers use the model as the basis for their own LLMs. If that happens it might improve customer chatbots or internal question answering, while also showing how DBRX was built using Databricks's proprietary tools.

Databricks put together the dataset from which DBRX was developed using Apache Spark and Databricks notebooks for data processing, Unity Catalog for data management and governance, and MLflow for experiment tracking.

Minnick revealed that enterprise investment in LLMs was delayed by fears over third-party ownership and governance. "Having to move data out to third parties, not having ownership over the model weights, not being able to fully control the governance of the data end-to-end – these are things that slow them down," he explained.

"What we set out to build was an extremely efficient … model that enterprises can use to go and bring to their own applications for their own specific use cases."

Hyoun Park, CEO and chief analyst at Amalgam Insights, observed the significance of DBRX is that Databricks can show how the model was built, step-by-step, as a process for other enterprises to follow and fine tune.

"That combination of lineage, visibility, repeatability, and model ownership in end-to-end model tuning, testing, and operationalization is important."

Park noted that he understood Databricks had built over 50,000 custom models for clients already. "It's that combination of model building experience and the ability to do it at scale with a high performing model on par with the best private and open source efforts that makes this announcement notable to me from an enterprise IT perspective."

The DBRX news plays out against a changing competitive backdrop for Databricks. The biz has a long-term strategic partnership with Microsoft, which resulted in Azure Databricks – wherein users are promised integrated data services closely tied to the Redmond giant's cloud platform.

But since the offer launched in 2017, Microsoft has moved into Databricks's lakehouse market – where users are offered data warehousing and data lakes in one environment – and promises users enterprise-grade LLMs with its $10 billion OpenAI partnership. In its Fabric environment, Microsoft can also offer "mirroring" from its transactional database systems Azure Cosmos DB and Azure SQL DB, offering access to analytics services without moving data.

An open question hanging over the strategies of both Databricks and Microsoft is when the flood of expected investment in LLM technologies is going to arrive. In January, Gartner predicted enterprise spending on the technology won't be coming this year, and would have little impact on other IT investment. ®

More about


Send us news

Other stories you might like