GPT-1

Training compute

1.8×10¹⁹ FLOP

Parameters

117M

Published

Jun 1, 2018

GPT-1 is an AI model developed by OpenAI (United States), first published in June 2018. It works in the language domain, on tasks such as question answering, text classification and language modeling.

Training it took an estimated 1.8×10¹⁹ FLOP of compute (estimation method: operation counting,third-party estimation). The model has 117,000,000 parameters. It was trained on roughly 1.3B datapoints. Training ran on 8 NVIDIA Quadro P600 for about 720 hours.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 13,799 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: OpenAI
Country of organization: United States
Domain: Language
Task: Question answering, Text classification, Language modeling
Training compute: 1.8×10¹⁹ FLOP
Compute estimation method: Operation counting, Third-party estimation
Parameters: 117,000,000
Dataset size: 1.3B
Training hardware: NVIDIA Quadro P600
Chips used: 8
Training time: 720 h
Training power draw: 664 W
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 13,799
Epoch confidence: Likely