DBRX is an open, state-of-the-art, general-purpose LLM trained using Mosaic AI Training. Mosaic AI Training helps overcome infrastructure, performance, and scientific challenges in LLM training. It provides a training stack, distributed training capabilities, distributed checkpointing, training performance optimizations, GPU fault tolerance, network fabric fault tolerance, and experiment tracking. It enables users to build their own custom LLMs tailored to specific business contexts and language domains.

9m read timeFrom databricks.com
Post cover image
Table of contents
Mosaic AI Training stackDistributed trainingDistributed checkpointingTraining performance optimizationsGPU fault toleranceNetwork fabric fault toleranceExperiment trackingStart training your own custom LLM

Sort: