“DBRX was trained on 3072 NVIDIA H100s connected by 3.2Tbps Infiniband.”
Databricks built a 132B-parameter mixture-of-experts model on 12T tokens over three months and benchmarked it against GPT-3.5. The headline rival is a model two generations old. The weights ship under “an open license” that the launch post never names. The real product is the pitch that any enterprise can now train its own model from scratch on Databricks compute.