DALL-E

Training compute

4.7×10²² FLOP

Parameters

12B

Published

Jan 5, 2021

DALL-E is an AI model developed by OpenAI (United States), first published in January 2021. It works in the image generation domain, on tasks such as text-to-image and image generation.

Training it took an estimated 4.7×10²² FLOP of compute (estimation method: third-party estimation). The model has 12,000,000,000 parameters. It was trained on roughly 320B datapoints. Training ran on 1,024 NVIDIA Tesla V100 DGXS 16 GB. The compute alone is estimated at $126K in 2023 dollars.

Access: API access. Its weights are not openly released. The reference paper has 6,403 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: OpenAI
Country of organization: United States
Domain: Image generation
Task: Text-to-image, Image generation
Training compute: 4.7×10²² FLOP
Compute estimation method: Third-party estimation
Parameters: 12,000,000,000
Dataset size: 320B
Training hardware: NVIDIA Tesla V100 DGXS 16 GB
Chips used: 1,024
Training power draw: 519.7 kW
Training cost (2023 USD): $126K
Numerical format: FP16
Model accessibility: API access
Open weights: No
Citations: 6,403
Epoch confidence: Likely