GShard (600B)

Training compute

1.3×10²² FLOP

Parameters

600B

Published

Jun 30, 2020

GShard (600B) is an AI model developed by Google (United States), first published in June 2020. It works in the language domain, on tasks such as translation.

Training it took an estimated 1.3×10²² FLOP of compute (estimation method: third-party estimation,hardware). The model has 600,000,000,000 parameters. It was trained on roughly 1T datapoints. Training ran on Google TPU v3 for about 96 hours.

Access: Unreleased. Its weights are not openly released. The reference paper has 1,992 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google
Country of organization: United States
Domain: Language
Task: Translation
Training compute: 1.3×10²² FLOP
Compute estimation method: Third-party estimation, Hardware
Parameters: 600,000,000,000
Dataset size: 1T
Training hardware: Google TPU v3
Training time: 96 h
Chip-hours: 96.4K
Model accessibility: Unreleased
Open weights: No
Citations: 1,992
Epoch confidence: Confident