SigLino 70M
SigLino 70M is a dense image encoder with 12 layers and a hidden width of 512. It distills DINOv3 and SigLIP2 into one downloadable model.
How this encoder is configured
The model uses 16-by-16 patches and an Apache 2.0 license. It occupies the next published dense size above SigLino 30M. The card reports separate retrieval and segmentation evaluations rather than presenting one universal quality score.SigLino 70M model card
Evaluate the downstream feature task
Practical guidance. Compare it with 30M on the same image collection and resolution. A larger feature extractor can be worthwhile if it resolves similar products or scenes that the smaller model confuses. Include indexing time and embedding storage in the comparison.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/siglino-70M |
| License | Apache 2.0SigLino 70M model card |
| Inputs and outputs | Images in; visual feature vectors outSigLino 70M 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