MSRA (C, PReLU)

Training compute

2.4×10¹⁹ FLOP

Parameters

87M

Published

Feb 6, 2015

MSRA (C, PReLU) is an AI model developed by Microsoft Research (United States), first published in February 2015. 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 2.4×10¹⁹ FLOP of compute (estimation method: hardware). The model has 87,048,800 parameters. It was trained on roughly 1.3M datapoints. Training ran on NVIDIA Tesla K40t for about 588 hours. The compute alone is estimated at $1K in 2023 dollars.

The reference paper has 20,475 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft Research
Country of organization: United States
Domain: Vision
Task: Image classification
Training compute: 2.4×10¹⁹ FLOP
Compute estimation method: Hardware
Parameters: 87,048,800
Dataset size: 1.3M
Training hardware: NVIDIA Tesla K40t
Training time: 588 h
Training cost (2023 USD): $1K
Numerical format: FP32
Citations: 20,475
Epoch confidence: Confident