Xception

Training compute

4.4×10²⁰ FLOP

Parameters

22.9M

Published

Oct 7, 2016

Xception is an AI model developed by Google (United States), first published in October 2016. 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 4.4×10²⁰ FLOP of compute (estimation method: hardware,third-party estimation). The model has 22,855,952 parameters. It was trained on roughly 350M datapoints. Training ran on 60 NVIDIA Tesla K80 for about 720 hours. The compute alone is estimated at $13K in 2023 dollars.

The reference paper has 17,674 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google
Country of organization: United States
Domain: Vision
Task: Image classification
Training compute: 4.4×10²⁰ FLOP
Compute estimation method: Hardware, Third-party estimation
Parameters: 22,855,952
Dataset size: 350M
Training hardware: NVIDIA Tesla K80
Chips used: 60
Training time: 720 h
Chip-hours: 43.2K
Training power draw: 37.8 kW
Training cost (2023 USD): $13K
Citations: 17,674
Epoch confidence: Confident