SigLino 30M
SigLino 30M is a small dense vision encoder distilled from DINOv3 and SigLIP2. It produces image features for downstream tasks, rather than generating a text answer.
How this encoder is configured
The card describes 12 layers, a hidden width of 384 and 16-by-16 image patches. The family combines knowledge from two vision teachers. Its published evaluations include image retrieval and segmentation, which need task-specific use of the extracted features.SigLino 30M model card
Evaluate the downstream feature task
Practical guidance. Start with image search on a small labeled sample. Check whether similar items remain distinct when backgrounds or viewpoints change. The encoder is one component of a retrieval system; it still needs an index and a defined similarity method.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/siglino-30M |
| License | Apache 2.0SigLino 30M model card |
| Inputs and outputs | Images in; visual feature vectors outSigLino 30M model card |
Related models
A sparse vision encoder with 0.6B total and 0.15B active parameters that selects two of 28 experts.
Launch date not confirmedDownloadable