Model Family
DBRX
Databricks' open-weights 132B mixture-of-experts LLM.
Definition
DBRX is Databricks' fine-grained MoE LLM with 132B total parameters and 36B active per token. Released openly in 2024, it outperformed Llama 2 70B on benchmarks while staying cheap to serve via expert routing.
Common use cases
- Enterprise self-hosting
- Fine-tuning
- RAG