ViT-G/14

Training compute

5.8×10²² FLOP

Parameters

1.8B

Published

Jun 8, 2021

ViT-G/14 is an AI model developed by Google Brain and Google Research (United States), first published in June 2021. It works in the vision domain, on tasks such as image classification.

Training it took an estimated 5.8×10²² FLOP of compute (estimation method: hardware,operation counting). The model has 1,843,000,000 parameters. It was trained on roughly 3B datapoints. Training ran on 2,048 Google TPU v3. The compute alone is estimated at $4K in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 1,393 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google Brain, Google Research
Country of organization: United States
Domain: Vision
Task: Image classification
Training compute: 5.8×10²² FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 1,843,000,000
Dataset size: 3B
Training hardware: Google TPU v3
Chips used: 2,048
Training power draw: 1.9 MW
Training cost (2023 USD): $4K
Numerical format: BF16
Model accessibility: Unreleased
Open weights: No
Citations: 1,393
Epoch confidence: Confident