LaMDA

Training compute

3.5×10²³ FLOP

Parameters

137B

Published

Feb 10, 2022

LaMDA is an AI model developed by Google (United States), first published in February 2022. It works in the language domain, on tasks such as language modeling.

Training it took an estimated 3.5×10²³ FLOP of compute (estimation method: hardware). The model has 137,000,000,000 parameters. It was trained on roughly 2.1T datapoints. Training ran on 1,024 Google TPU v3 for about 1.4K hours. The compute alone is estimated at $230K in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 1,863 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google
Country of organization: United States
Domain: Language
Task: Language modeling
Training compute: 3.5×10²³ FLOP
Compute estimation method: Hardware
Parameters: 137,000,000,000
Dataset size: 2.1T
Training hardware: Google TPU v3
Chips used: 1,024
Training time: 1,385 h
Chip-hours: 1.4M
Training power draw: 927.2 kW
Training cost (2023 USD): $230K
Model accessibility: Unreleased
Open weights: No
Citations: 1,863
Epoch confidence: Confident