Lensly models
ExecuTorch (.pte, XNNPACK, arm64 Android) exports used by Lensly, a private
on-device photo search app. Nothing here is trained by the Lensly project; these are format conversions.
| File | What | Source model | Licence |
|---|---|---|---|
siglip2_b32_256/siglip2_b32_256_image_xnnpack_int8_attn.pte |
image encoder, int8 | google/siglip2-base-patch32-256 | Apache-2.0 |
siglip2_b32_256/siglip2_b32_256_text_xnnpack_int8.pte |
text encoder, int8, 64-token padding baked in | same | Apache-2.0 |
siglip2_b32_256/tokenizer_ascii.json |
pruned English tokenizer | same | Apache-2.0 |
siglip2_b32_256/tokenizer.json |
full tokenizer | same | Apache-2.0 |
sface/sface_xnnpack_fp32.pte |
face embedder (128-d) | OpenCV Zoo SFace (2021dec) | Apache-2.0 |
The app downloads these once, verifies each file's SHA-256 against a manifest pinned in its source
(src/core/models/registry.ts), and then runs fully offline. Photos never leave the phone.
Export scripts: tools/model-export/ in the app repository.
- Downloads last month
- 17
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support
Model tree for santaklux/lensly-models
Base model
google/siglip2-base-patch32-256