SEER

Training compute

1.8×10²² FLOP

Parameters

1.3B

Published

Jul 29, 2021

SEER is an AI model developed by Facebook AI Research and INRIA (United States and France), first published in July 2021. It works in the vision domain, on tasks such as image embedding and image classification.

Training it took an estimated 1.8×10²² FLOP of compute (estimation method: hardware). The model has 1,300,000,000 parameters. It was trained on roughly 1B datapoints. Training ran on 512 NVIDIA V100 for about 196 hours. The compute alone is estimated at $34K in 2023 dollars.

Access: Open weights (non-commercial). Its weights are openly available. The reference paper has 301 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 embedding, Image classification
Training compute: 1.8×10²² FLOP
Compute estimation method: Hardware
Parameters: 1,300,000,000
Dataset size: 1B
Training hardware: NVIDIA V100
Chips used: 512
Training time: 196 h
Chip-hours: 98.3K
Training power draw: 310.4 kW
Training cost (2023 USD): $34K
Numerical format: FP16
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Citations: 301
Epoch confidence: Confident