DINOv2

Training compute

7.4×10²¹ FLOP

Parameters

1.1B

Published

Apr 14, 2023

DINOv2 is an AI model developed by Facebook AI Research and INRIA (United States and France), first published in April 2023. It works in the vision domain, on tasks such as image representation and image classification.

Training it took an estimated 7.4×10²¹ FLOP of compute (estimation method: hardware). The model has 1,140,000,000 parameters. It was trained on roughly 36.4B datapoints. Training ran on NVIDIA A100 SXM4 40 GB. The compute alone is estimated at $10K in 2023 dollars.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 8,133 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Facebook AI Research, INRIA
Country of organization: United States, France
Domain: Vision
Task: Image representation, Image classification
Training compute: 7.4×10²¹ FLOP
Compute estimation method: Hardware
Parameters: 1,140,000,000
Dataset size: 36.4B
Training hardware: NVIDIA A100 SXM4 40 GB
Training cost (2023 USD): $10K
Numerical format: FP16
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 8,133
Epoch confidence: Confident