ERNIE 3.0

Training compute

2.3×10²² FLOP

Parameters

10B

Published

Jul 5, 2021

ERNIE 3.0 is an AI model developed by Baidu (China), first published in July 2021. It works in the language domain, on tasks such as language modeling, language modeling/generation, text classification and question answering.

Training it took an estimated 2.3×10²² FLOP of compute (estimation method: operation counting). The model has 10,000,000,000 parameters. It was trained on roughly 375B datapoints. Training ran on 384 NVIDIA V100. The compute alone is estimated at $39K in 2023 dollars.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 590 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Baidu
Country of organization: China
Domain: Language
Task: Language modeling, Language modeling/generation, Text classification, Question answering
Training compute: 2.3×10²² FLOP
Compute estimation method: Operation counting
Parameters: 10,000,000,000
Dataset size: 375B
Training hardware: NVIDIA V100
Chips used: 384
Training power draw: 232.9 kW
Training cost (2023 USD): $39K
Numerical format: FP16
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 590
Epoch confidence: Confident