BERT-Large

Training compute

2.9×10²⁰ FLOP

Parameters

340M

Published

Oct 11, 2018

BERT-Large is an AI model developed by Google (United States), first published in October 2018. It works in the language domain, on tasks such as question answering and text autocompletion.

Training it took an estimated 2.9×10²⁰ FLOP of compute (estimation method: operation counting,hardware,third-party estimation). The model has 340,000,000 parameters. It was trained on roughly 2.6B datapoints. Training ran on 64 Google TPU v2 for about 96 hours. The compute alone is estimated at $2K in 2023 dollars.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 114,811 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google
Country of organization: United States
Domain: Language
Task: Question answering, Text autocompletion
Training compute: 2.9×10²⁰ FLOP
Compute estimation method: Operation counting, Hardware, Third-party estimation
Parameters: 340,000,000
Dataset size: 2.6B
Training hardware: Google TPU v2
Chips used: 64
Training time: 96 h
Chip-hours: 6.1K
Training power draw: 37.0 kW
Training cost (2023 USD): $2K
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 114,811
Epoch confidence: Confident