Florence

Training compute

4.8×10²² FLOP

Parameters

893M

Published

Nov 22, 2021

Florence is an AI model developed by Microsoft (United States), first published in November 2021. It works in the vision domain, on tasks such as image captioning, visual question answering, image classification and object detection.

Training it took an estimated 4.8×10²² FLOP of compute (estimation method: hardware). The model has 893,000,000 parameters. It was trained on roughly 7.5B datapoints. Training ran on 512 NVIDIA A100 SXM4 40 GB for about 240 hours. The compute alone is estimated at $107K in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 1,122 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft
Country of organization: United States
Domain: Vision
Task: Image captioning, Visual question answering, Image classification, Object detection
Training compute: 4.8×10²² FLOP
Compute estimation method: Hardware
Parameters: 893,000,000
Dataset size: 7.5B
Training hardware: NVIDIA A100 SXM4 40 GB
Chips used: 512
Training time: 240 h
Chip-hours: 122.9K
Training power draw: 412.8 kW
Training cost (2023 USD): $107K
Numerical format: FP16
Model accessibility: Unreleased
Open weights: No
Citations: 1,122
Epoch confidence: Confident