DBRX: A New State-of-the-Art Open LLM
Product Information
Key Features of DBRX: A New State-of-the-Art Open LLM
DBRX is a fine-grained mixture-of-experts (MoE) architecture that provides substantial improvements in compute-efficiency for training and inference. It surpasses GPT-3.5 and is competitive with Gemini 1.0 Pro.
Fine-Grained Mixture-of-Experts (MoE) Architecture
DBRX uses a fine-grained MoE architecture, which provides substantial improvements in compute-efficiency for training and inference.
Transformer-Based Decoder-Only Large Language Model
DBRX is a transformer-based decoder-only large language model that was trained using next-token prediction.
132B Total Parameters
DBRX has 132B total parameters, of which 36B parameters are active on any input.
Trained on 3072 NVIDIA H100s
DBRX was trained on 3072 NVIDIA H100s connected by 3.2Tbps Infiniband.
Suite of Databricks Tools
DBRX was trained using a suite of Databricks tools, including Unity Catalog, Apache Spark, and Databricks notebooks.
Use Cases of DBRX: A New State-of-the-Art Open LLM
Natural Language Processing (NLP)
Code Modeling
Mathematics and Problem Solving
Conversational AI
Pros and Cons of DBRX: A New State-of-the-Art Open LLM
Pros
- State-of-the-art performance in established open LLMs
- Substantial improvements in compute-efficiency for training and inference
- Competitive with Gemini 1.0 Pro and GPT-3.5
- Especially capable in code modeling
Cons
- Large model size (132B total parameters)
- Requires significant computational resources for training and inference
How to Use DBRX: A New State-of-the-Art Open LLM
- 1
Get started with DBRX on Databricks by downloading the model from the Databricks Marketplace
- 2
Deploy the model on Model Serving for production applications
- 3
Use the Databricks Foundation Model APIs for pay-as-you-go pricing and query the model from the AI Playground chat interface