DBRX

Training compute

2.6×10²⁴ FLOP

Parameters

132B

Published

Mar 27, 2024

DBRX is an AI model developed by Databricks (United States), first published in March 2024. It works in the language domain, on tasks such as chat and code generation.

Training it took an estimated 2.6×10²⁴ FLOP of compute (estimation method: operation counting). The model has 132,000,000,000 parameters. It was trained on roughly 12T datapoints. Training ran on NVIDIA H100 SXM5 80GB.

Access: Open weights (restricted use). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Databricks
Country of organization: United States
Domain: Language
Task: Chat, Code generation
Training compute: 2.6×10²⁴ FLOP
Compute estimation method: Operation counting
Parameters: 132,000,000,000
Dataset size: 12T
Training hardware: NVIDIA H100 SXM5 80GB
Numerical format: BF16
Model accessibility: Open weights (restricted use)
Open weights: Yes
Epoch confidence: Confident