PaliGemma 2 3B Mix 224

Training compute

4×10²² FLOP

Parameters

2.9B

Published

Feb 19, 2025

PaliGemma 2 3B Mix 224 is an AI model developed by Google (United States), first published in February 2025. It works in the vision, multimodal and language domain, on tasks such as image captioning, visual question answering, object detection and 3 more.

Training it took an estimated 4×10²² FLOP of compute (estimation method: hardware,operation counting). The model has 2,920,000,000 parameters. It was trained on roughly 256B datapoints. Training ran on 256 Google TPU v5e for about 72 hours.

Access: Open weights (non-commercial). Its weights are openly available. It is built on top of Gemma 2 2B,SigLIP 400M. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google
Country of organization: United States
Domain: Vision, Multimodal, Language
Task: Image captioning, Visual question answering, Object detection, Image segmentation, Object recognition, Character recognition (OCR)
Training compute: 4×10²² FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 2,920,000,000
Dataset size: 256B
Training hardware: Google TPU v5e
Chips used: 256
Training time: 72 h
Training power draw: 113.1 kW
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Base model: Gemma 2 2B, SigLIP 400M
Epoch confidence: Confident