Databricks Spent $10 Million To Develop New DBRX Generative AI Model
On Wednesday, Databricks announced that it invested around $10 million and two months in training to launch its latest generative AI model, DBRX. The company claims that DBRX “outperforms all established open source models on standard benchmarks.”
Databricks provides a cloud-based platform to help enterprises build, scale, and govern data and AI, and its new AI model is similar to OpenAI’s GPT series and Google’s Gemini. It’s available on GitHub and Hugging Face for both research and commercial purposes. The new model offers base (DBRX Base) and fine-tuned (DBRX Instruct) versions, which are adaptable to public or custom data.
In an interview, Naveen Rao, VP of generative AI at Databricks, said, “DBRX was trained to be useful and provide information on a wide variety of topics.” He added, “DBRX has been optimized and tuned for English language usage, but is capable of conversing and translating into a wide variety of languages, such as French, Spanish and German.”
Reports emphasize that using DBRX without being a Databricks customer makes it very challenging. To run DBRX in its standard configuration, users need a server or PC equipped with at least four Nvidia H100 GPUs (or equivalent GPUs totaling around 320GB of memory). With each H100 GPU costing thousands of dollars, this requirement may be within reach for enterprises, but it’s often out of budget for many developers and solopreneurs.
“We’re focused on making the Databricks platform the best choice for customized model building, so ultimately the benefit to Databricks is more users on our platform,” Rao said. “DBRX is a demonstration of our best-in-class pre-training and tuning platform, which customers can use to build their own models from scratch. It’s an easy way for customers to get started with the Databricks Mosaic AI generative AI tools. And DBRX is highly capable out-of-the-box and can be tuned for excellent performance on specific tasks at better economics than large, closed models.”
After finishing the Gen AI model, Jonathan Frankle, chief neural network architect at Databricks and leader of the team that built DBRX, told his team, “We’ve surpassed everything.” According to Wired, “Frankle shared data showing that across about a dozen or so benchmarks measuring the AI model’s ability to answer general knowledge questions, perform reading comprehension, solve vexing logical puzzles, and generate high-quality code, DBRX was better than every other open source model available.”
