CTR-BERT

Training compute

6.5×10¹⁹ FLOP

Parameters

70M

Published

Dec 6, 2021

CTR-BERT is an AI model developed by Amazon (United States), first published in December 2021. It works in the recommendation domain, on tasks such as click-through rate prediction.

Training it took an estimated 6.5×10¹⁹ FLOP of compute (estimation method: hardware). The model has 70,000,000 parameters. It was trained on roughly 1.2B datapoints. Training ran on 8 NVIDIA A100.

Access: Unreleased. Its weights are not openly released. The reference paper has 52 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: Amazon
Country of organization: United States
Domain: Recommendation
Task: Click-through rate prediction
Training compute: 6.5×10¹⁹ FLOP
Compute estimation method: Hardware
Parameters: 70,000,000
Dataset size: 1.2B
Training hardware: NVIDIA A100
Chips used: 8
Training power draw: 6.4 kW
Model accessibility: Unreleased
Open weights: No
Citations: 52
Epoch confidence: Likely