Segment Anything Model

Training compute

7.8×10²¹ FLOP

Parameters

636M

Published

Apr 5, 2023

Segment Anything Model is an AI model developed by Meta AI (United States), first published in April 2023. It works in the vision domain, on tasks such as image segmentation.

Training it took an estimated 7.8×10²¹ FLOP of compute (estimation method: hardware). The model has 636,000,000 parameters. It was trained on roughly 1.1B datapoints. Training ran on 256 NVIDIA A100 for about 68 hours. The compute alone is estimated at $16K in 2023 dollars.

Access: Open weights (unrestricted). Its weights are openly available. It is built on top of ViT-Huge/14. The reference paper has 13,486 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Meta AI
Country of organization: United States
Domain: Vision
Task: Image segmentation
Training compute: 7.8×10²¹ FLOP
Compute estimation method: Hardware
Parameters: 636,000,000
Dataset size: 1.1B
Training hardware: NVIDIA A100
Chips used: 256
Training time: 68 h
Training power draw: 204.1 kW
Training cost (2023 USD): $16K
Numerical format: FP32
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Base model: ViT-Huge/14
Citations: 13,486
Epoch confidence: Confident