The company used Apache Spark, Databricks notebooks, Unity Catalog, and MLflow to develop DBRX. Databricks' aim is not just the performance of the model, but to show how it was built using its proprietary tools, which could improve customer chatbots or internal question answering. The company's strategic partnership with Microsoft, which resulted in Azure Databricks, is also noteworthy. However, Microsoft's move into Databricks's lakehouse market and its partnership with OpenAI could pose a challenge.
Key takeaways:
- Databricks has launched an open source large language model (LLM) called DBRX, which it claims outperforms rivals in language understanding, programming, and math.
- DBRX was developed by Mosaic AI, which Databricks acquired for $1.3 billion, and trained on Nvidia DGX Cloud. It uses a mixture-of-experts (MoE) architecture and possesses 132 billion parameters, but only 36 billion are active on any one input.
- Databricks is making DBRX available for free on GitHub and Hugging Face, hoping customers will use it as the basis for their own LLMs, potentially improving customer chatbots or internal question answering.
- The launch of DBRX comes as Databricks faces increasing competition from Microsoft, which has moved into Databricks's lakehouse market and promises users enterprise-grade LLMs with its $10 billion OpenAI partnership.