AlexaTM 20B

Training compute

2×10²³ FLOP

Parameters

19.8B

Published

Aug 2, 2022

AlexaTM 20B is an AI model developed by Amazon (United States), first published in August 2022. It works in the language domain, on tasks such as language modeling, translation and question answering.

Training it took an estimated 2×10²³ FLOP of compute (estimation method: hardware). The model has 19,750,000,000 parameters. It was trained on roughly 1.3T datapoints. Training ran on 128 NVIDIA A100 for about 2.9K hours. The compute alone is estimated at $268K in 2023 dollars.

Access: API access. Its weights are not openly released. The reference paper has 90 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Amazon
Country of organization: United States
Domain: Language
Task: Language modeling, Translation, Question answering
Training compute: 2×10²³ FLOP
Compute estimation method: Hardware
Parameters: 19,750,000,000
Dataset size: 1.3T
Training hardware: NVIDIA A100
Chips used: 128
Training time: 2,880 h
Chip-hours: 368.6K
Training power draw: 102.6 kW
Training cost (2023 USD): $268K
Numerical format: BF16
Model accessibility: API access
Open weights: No
Citations: 90
Epoch confidence: Confident