cstr/clip-text-base-GGUF

CLIP Text Encoder Base (GGUF)

GGUF conversion of the CLIP text encoder (base) for use with CrispEmbed. Extracted from openai/clip-vit-base-patch16.

• —Architecture: CLIP text transformer with causal attention
• —Parameters: 63M
• —Output: 512-dimensional L2-normalized embeddings
• —Tokenizer: BPE tokenizer (embedded in GGUF), max 77 tokens
• —Size: ~244 MB

Usage

# Embed a single text
crispembed -m clip-text-base "a photo of a cat"

# Embed from file
crispembed -m clip-text-base --input queries.txt --output embeddings.bin

Cross-modal pairing

Output embeddings live in the same vector space as cstr/clip-vit-base-patch16-GGUF. Use both for zero-shot image-text retrieval:

crispembed -m clip-text-base "a photo of a cat"               # text embedding
crispembed -m clip-vit-base-patch16 --image photo.jpg          # vision embedding
# cosine similarity measures image-text alignment

Notes

• —All output embeddings are L2-normalized.
• —BPE tokenizer is bundled inside the GGUF file; no external vocab files needed.
• —This is a GGUF conversion; weights are numerically equivalent to the original HuggingFace model.

Provenance and EU AI Act Art. 53 note

• —Upstream model: openai/clip-vit-base-patch16 — published by openai.
• —Upstream licence: mit. 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.