AllMiniLML6V2-coreml

Overview

This repository provides a Core ML–converted version of the all-MiniLM-L6-v2 embedding model, along with a fully runnable SwiftUI demo, tokenizer, vocab, and the conversion script used to generate the model.
It is designed to help iOS developers easily run on-device text embeddings without needing to perform the Core ML conversion themselves.

---

Features

• ✅ Core ML conversion of all-MiniLM-L6-v2
• ✅ Fully functional SwiftUI demo app
• ✅ MiniLM tokenizer implemented in Swift
• ✅ vocab.txt included for tokenization
• ✅ convert_minilm_to_coreml.py script
• ✅ Organized, production-ready folder structure
• ✅ MIT license for code + Apache-2.0 license for the model

---

Folder Structure

AllMiniLML6V2-coreml/
├─ Models/
│  ├─ embeddings/
│  │  └─ AllMiniLML6V2.mlmodel
│  └─ llm/
│     └─ vocab.txt
├─ Scripts/
│  └─ convert_minilm_to_coreml.py
├─ Sources/
│  └─ App/
│     ├─ AllMiniLML6V2CoreMLApp.swift
│     ├─ Utils/
│     │  └─ MiniLMTokenizer.swift
│     └─ Views/
│        └─ ContentView.swift
├─ Assets/
│  └─ Assets.xcassets
├─ LICENSE
└─ README.md

---

Getting Started

1. Clone the repository

git clone https://github.com/<your-username>/AllMiniLML6V2-coreml.git
cd AllMiniLML6V2-coreml

2. Open in Xcode

Open the folder in Xcode:

open .

3. Add model + vocab to target membership

Inside Xcode:

• Select AllMiniLML6V2.mlmodel
• Check Target Membership → Your App
• Do the same for vocab.txt

4. Run the app

Build and run the SwiftUI demo.
The ContentView already loads the tokenizer and runs the embedding pipeline.

---

Usage Example

Inside ContentView.swift:

let tokenizer = try MiniLMTokenizer(vocabURL: vocabURL)
let tokens = tokenizer.encode("Hello world")
let embeddings = try model.prediction(inputIds: tokens.inputIds, attentionMask: tokens.attentionMask)

---

Model License & Attribution

The original model:

• sentence-transformers/all-MiniLM-L6-v2
• Hugging Face: https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2
• Licensed under: Apache License 2.0

Your converted model is a derivative under the same license for model files.

See full details inside the LICENSE file.

---

Code License

All custom code, scripts, and Swift components in this repository
are licensed under the MIT License.

---

Conversion Script

The model was converted using:

python convert_minilm_to_coreml.py   --model_name_or_path sentence-transformers/all-MiniLM-L6-v2   --output AllMiniLML6V2.mlmodel

---

Contact

Maintainer: Abhishek
GitHub: https://github.com/Abhishek6353

---

Star the Repo ⭐

If this helped you, please consider starring the repository!