PaliGemma 2
PaliGemma 2: SigLIP encoder with the full Gemma 2 range (2B, 9B, 27B) at 224, 448 and 896 px for transfer learning.
Broader transfer coverage than PaliGemma, including OCR, table and molecular-structure recognition, music scores, fine-grained captions and radiography reports. Paper arXiv 2412.03555.
- Date
- Wednesday, 4 December 2024
- Lab
- Google DeepMind
- Kind
- open-weights
- Access
- open weights (restricted license)
Model sizes quoted as 3B/10B/28B in some sources (vision encoder included).
Sources
This record was checked against its sources on 6 October 2026. How we check