Kosmos-2.5

Training compute

2.2×10²¹ FLOP

Parameters

1.3B

Published

Aug 21, 2024

Kosmos-2.5 is an AI model developed by Microsoft (United States), first published in August 2024. It works in the multimodal, language and vision domain, on tasks such as character recognition (ocr), document classification, language modeling/generation and 2 more.

Training it took an estimated 2.2×10²¹ FLOP of compute (estimation method: operation counting). The model has 1,300,000,000 parameters. It was trained on roughly 260B datapoints.

Access: Open weights (unrestricted). Its weights are openly available. It is built on top of Pix2Struct-Large. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft
Country of organization: United States
Domain: Multimodal, Language, Vision
Task: Character recognition (OCR), Document classification, Language modeling/generation, Visual question answering, Document representation
Training compute: 2.2×10²¹ FLOP
Compute estimation method: Operation counting
Parameters: 1,300,000,000
Dataset size: 260B
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Base model: Pix2Struct-Large
Epoch confidence: Confident