AmoebaNet-A (F=448)

Training compute

3.9×10²⁰ FLOP

Parameters

469M

Published

Feb 5, 2018

AmoebaNet-A (F=448) is an AI model developed by Google Brain (United States), first published in February 2018. It works in the vision domain, on tasks such as image classification.

Training it took an estimated 3.9×10²⁰ FLOP of compute (estimation method: hardware). The model has 469,000,000 parameters. It was trained on roughly 1.1M datapoints. Training ran on 450 NVIDIA Tesla K40s for about 168 hours. The compute alone is estimated at $12K in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 3,322 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google Brain
Country of organization: United States
Domain: Vision
Task: Image classification
Training compute: 3.9×10²⁰ FLOP
Compute estimation method: Hardware
Parameters: 469,000,000
Dataset size: 1.1M
Training hardware: NVIDIA Tesla K40s
Chips used: 450
Training time: 168 h
Chip-hours: 75.6K
Training power draw: 229.2 kW
Training cost (2023 USD): $12K
Numerical format: FP32
Model accessibility: Unreleased
Open weights: No
Citations: 3,322
Epoch confidence: Confident