ResNeXt-101 32x48d

Training compute

8.7×10²¹ FLOP

Parameters

829M

Published

May 2, 2018

ResNeXt-101 32x48d is an AI model developed by Facebook (United States), first published in May 2018. It works in the vision domain, on tasks such as image classification. It counts among the frontier models: the systems trained with the most compute of their moment.

Training it took an estimated 8.7×10²¹ FLOP of compute (estimation method: operation counting). The model has 829,000,000 parameters. It was trained on roughly 940M datapoints. Training ran on 336 NVIDIA V100 for about 496 hours. The compute alone is estimated at $134K in 2023 dollars.

Access: Open weights (non-commercial). Its weights are openly available. The reference paper has 1,462 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Facebook
Country of organization: United States
Domain: Vision
Task: Image classification
Training compute: 8.7×10²¹ FLOP
Compute estimation method: Operation counting
Parameters: 829,000,000
Dataset size: 940M
Training hardware: NVIDIA V100
Chips used: 336
Training time: 496 h
Training power draw: 209.2 kW
Training cost (2023 USD): $134K
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Citations: 1,462
Epoch confidence: Confident