Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: google/siglip2-base-patch16-224
|
| 3 |
+
tags:
|
| 4 |
+
- mlx
|
| 5 |
+
- siglip2
|
| 6 |
+
- image-retrieval
|
| 7 |
+
- semanticgallery
|
| 8 |
+
language:
|
| 9 |
+
- en
|
| 10 |
+
library_name: transformers
|
| 11 |
+
pipeline_tag: zero-shot-image-classification
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# SemanticGallery MLX SigLIP2 Stage 1
|
| 15 |
+
|
| 16 |
+
This repository hosts the published Stage 1 weights used by SemanticGallery.
|
| 17 |
+
|
| 18 |
+
- Base model: `google/siglip2-base-patch16-224`
|
| 19 |
+
- Runtime target: `MLX` on Apple Silicon
|
| 20 |
+
- Training path: public text-image alignment on Flickr30k and Screen2Words
|
| 21 |
+
|
| 22 |
+
SemanticGallery quick start downloads these Stage 1 weights automatically on first run. Stage 2 remains local and adapts the image encoder to the user's own gallery.
|