AraGPT2-Mega

Training compute

2×10²¹ FLOP

Parameters

1.5B

Published

Dec 31, 2020

AraGPT2-Mega is an AI model developed by American University of Beirut (Lebanon), first published in December 2020. It works in the language domain, on tasks such as language modeling/generation and question answering.

Training it took an estimated 2×10²¹ FLOP of compute (estimation method: third-party estimation,hardware,operation counting). The model has 1,500,000,000 parameters. Training ran on 128 Google TPU v3 for about 216 hours.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 145 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: American University of Beirut
Country of organization: Lebanon
Domain: Language
Task: Language modeling/generation, Question answering
Training compute: 2×10²¹ FLOP
Compute estimation method: Third-party estimation, Hardware, Operation counting
Parameters: 1,500,000,000
Training hardware: Google TPU v3
Chips used: 128
Training time: 216 h
Training power draw: 117.0 kW
Numerical format: BF16
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 145
Epoch confidence: Confident