cstr/siglip-text-base-GGUF

SigLIP Text Base — GGUF (CrispEmbed)

GGUF conversion of the text encoder from google/siglip-base-patch16-224 for CrispEmbed.

Google's SigLIP text encoder produces 768-dim embeddings for cross-modal text-image similarity search. Uses sigmoid pairwise loss (no global batch normalization needed), enabling efficient contrastive learning.

Architecture

• —Text encoder: 12 transformer layers (768-dim, 12 heads, 3072 FFN)
• —Attention: bidirectional (unlike CLIP's causal)
• —Tokenizer: SentencePiece (32K vocab)
• —Pooling: last token (T-1) + linear projection head
• —Output: L2-normalized 768-dim embedding
• —Training: WebLI dataset, sigmoid pairwise loss

Files

Usage

crispembed --backend clip-text -m siglip-text-base-q4_k.gguf \
    --text "a photo of a cat"

License

Apache 2.0.

Credits

• —Original model: Google
• —GGUF conversion & runtime: CrispEmbed

Provenance and EU AI Act Art. 53 note

• —Upstream model: google/siglip-base-patch16-224 — 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). 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.