File size: 939 Bytes
3e8b112 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 | pix-siglip2-base-224
Copyright 2025 Google LLC (SigLIP 2 model weights and tokenizer)
This repository redistributes Google's SigLIP 2 base/16 at 224
(https://huggingface.co/google/siglip2-base-patch16-224, commit
75de2d55ec2d0b4efc50b3e9ad70dba96a7b2fa2), licensed under the Apache License,
Version 2.0 (see LICENSE).
Modifications (made for swift-photo-index, https://github.com/arraypress/swift-photo-index):
- The vision and text towers were exported, unchanged in architecture, to an
Apple Core AI asset (pix-siglip2-base-224.aimodel) with coreai-torch 0.4.3,
weights cast from float32 to float16, as two entry points (image, text) that
return the pooled embeddings (get_image_features / get_text_features).
- The Gemma tokenizer from the same commit was rewritten into two flat binary
files (pix-siglip2-support/vocab.bin, merges.bin) with the same ids and
merge order.
The export script is included as export_siglip2.py.
|