phi-3.5-Vision

Training compute

8.8×10²² FLOP

Parameters

4.2B

Published

Apr 23, 2024

phi-3.5-Vision is an AI model developed by Microsoft (United States), first published in April 2024. It works in the vision domain, on tasks such as visual question answering.

Training it took an estimated 8.8×10²² FLOP of compute (estimation method: hardware,operation counting). The model has 4,200,000,000 parameters. Training ran on 256 NVIDIA H100 SXM5 80GB for about 144 hours.

Access: Open weights (unrestricted). Its weights are openly available. It is built on top of phi-3-mini 3.8B. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft
Country of organization: United States
Domain: Vision
Task: Visual question answering
Training compute: 8.8×10²² FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 4,200,000,000
Training hardware: NVIDIA H100 SXM5 80GB
Chips used: 256
Training time: 144 h
Training power draw: 354.2 kW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Base model: phi-3-mini 3.8B
Epoch confidence: Confident