zidage/siglip2-base-coreml-macos

SigLIP2 Base CoreML macOS

Precompiled CoreML distribution for Alcedo/Puerh Mind semantic embeddings on macOS.

This repository stores compiled .mlmodelc bundles packaged as zip files so end users do not need Xcode or coremlcompiler installed at runtime.

Contents

• ImageEncoder.mlmodelc.zip: CoreML image encoder, input image, output embedding
• TextEncoder.mlmodelc.zip: CoreML text encoder, input tokens, output embedding
• tokenizer.zip: tokenizer files used by the text encoder
• manifest.json: file sizes and SHA-256 checksums consumed by Alcedo

Model Details

• Source conversion: palmier-io/siglip2-base-coreml
• Original architecture: google/siglip2-base-patch16-256
• Embedding dimension: 768
• Image size: 256
• Text context length: 64 tokens
• Output: L2-normalized embeddings
• Platform: macOS native CoreML

Checksums

ImageEncoder.mlmodelc.zip  91,698,362   f3255dad62bda6c50021b4eac3bf764423dd52b198005480273e800faa1babb8
TextEncoder.mlmodelc.zip   258,591,067  ba64d0cac0695b5c0cd18c898382a5455ed74ac666c27421ee94047a3561a72e
tokenizer.zip              5,460,173    c37f2a8e8555d8561109564c4f60ee962b0072abddcfcfd599d321469d6d1ef5
manifest.json              669          5dd52b1bb33595d78da3377f6f974c2f195fc628a2554f9d09c232bfd35fb397

Attribution

The CoreML packages are based on the conversion published at palmier-io/siglip2-base-coreml, which itself targets google/siglip2-base-patch16-256.