Databricks introduces DBRX as new standard for open-source LLMs #innovation

Databricks has launched DBRX, a new open-source large language model that outperforms established options like GPT-3.5. The 132 billion parameter DBRX model excels in language understanding, programming, and math tasks, surpassing other open-source LLMs and even a closed-source model. It demonstrated superior performance on coding tasks and matched or exceeded GPT-3.5 in various benchmarks. The efficiency of the model’s architecture makes it faster and more compute-efficient than other models. DBRX was pretrained on 12 trillion tokens of text and code data and leverages technologies like rotary position encodings and curriculum learning. Customers can interact with DBRX via APIs or finetune it on their proprietary data. Partners and analysts praise DBRX’s potential to accelerate enterprise adoption of open, customised large language models. The model is already being integrated into Databricks’ AI products. Overall, DBRX sets a new standard for open-source LLMs and provides enterprises with a platform to build customised reasoning capabilities based on their data.

Source link

Source link: https://www.artificialintelligence-news.com/2024/03/28/databricks-claims-dbrx-new-standard-open-source-llms/