cstr/siglip-large-256-GGUF

SigLIP ViT-L/16 Vision Encoder 256px (GGUF)

GGUF conversion of google/siglip-large-patch16-256 for use with CrispEmbed.

• Architecture: SigLIP ViT-L/16 vision encoder
• Parameters: 304M
• Output: 1024-dimensional L2-normalized embeddings
• Input: 256x256 RGB image with SigLIP normalization (mean=0.5, std=0.5)
• Size: ~1.2 GB
• Source: google/siglip-large-patch16-256

Usage

# Embed a single image
crispembed -m siglip-large-256 --image photo.jpg

# Batch processing
crispembed -m siglip-large-256 --image-dir ./photos/ --output embeddings.bin

About SigLIP

SigLIP replaces CLIP's softmax contrastive loss with a sigmoid loss, enabling better scaling and stronger performance on retrieval benchmarks. It uses a simpler normalization scheme (mean=0.5, std=0.5) compared to CLIP's ImageNet statistics.

Notes

• All output embeddings are L2-normalized.
• This is a GGUF conversion; weights are numerically equivalent to the original HuggingFace model.

Provenance and EU AI Act Art. 53 note

• Upstream model: google/siglip-large-patch16-256 — published by google.
• Upstream licence: apache-2.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF/GGML). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.