Aurora

Training compute

4.5×10²¹ FLOP

Parameters

1.3B

Published

May 28, 2024

Aurora is an AI model developed by Microsoft Research (United States), first published in May 2024. It works in the earth science domain, on tasks such as weather forecasting.

Training it took an estimated 4.5×10²¹ FLOP of compute (estimation method: hardware). The model has 1,300,000,000 parameters. Training ran on 32 NVIDIA A100 for about 420 hours.

Access: Unreleased. Its weights are not openly released. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft Research
Country of organization: United States
Domain: Earth science
Task: Weather forecasting
Training compute: 4.5×10²¹ FLOP
Compute estimation method: Hardware
Parameters: 1,300,000,000
Training hardware: NVIDIA A100
Chips used: 32
Training time: 420 h
Training power draw: 25.3 kW
Numerical format: BF16
Model accessibility: Unreleased
Open weights: No
Epoch confidence: Confident