LIMoE-H/14

Training compute

1.8×10²² FLOP

Parameters

5.6B

Published

Jun 6, 2022

LIMoE-H/14 is an AI model developed by Google (United States), first published in June 2022. It works in the multimodal, vision and language domain, on tasks such as image classification.

Training it took an estimated 1.8×10²² FLOP of compute (estimation method: operation counting). The model has 5,600,000,000 parameters. It was trained on roughly 4.6T datapoints. Training ran on Google TPU v2,Google TPU v3,Google TPU v4.

Access: Unreleased. Its weights are not openly released. The reference paper has 317 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google
Country of organization: United States
Domain: Multimodal, Vision, Language
Task: Image classification
Training compute: 1.8×10²² FLOP
Compute estimation method: Operation counting
Parameters: 5,600,000,000
Dataset size: 4.6T
Training hardware: Google TPU v2, Google TPU v3, Google TPU v4
Model accessibility: Unreleased
Open weights: No
Citations: 317
Epoch confidence: Confident