iGPT-L

Training compute

8.9×10²¹ FLOP

Parameters

1.4B

Published

Jun 17, 2020

iGPT-L is an AI model developed by OpenAI (United States), first published in June 2020. It works in the image generation and vision domain, on tasks such as image completion.

Training it took an estimated 8.9×10²¹ FLOP of compute (estimation method: hardware). The model has 1,362,000,000 parameters. It was trained on roughly 2.7B datapoints. Training ran on NVIDIA Tesla V100 DGXS 32 GB. The compute alone is estimated at $30K in 2023 dollars.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 1,694 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: OpenAI
Country of organization: United States
Domain: Image generation, Vision
Task: Image completion
Training compute: 8.9×10²¹ FLOP
Compute estimation method: Hardware
Parameters: 1,362,000,000
Dataset size: 2.7B
Training hardware: NVIDIA Tesla V100 DGXS 32 GB
Chip-hours: 60K
Training cost (2023 USD): $30K
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 1,694
Epoch confidence: Confident