PocketSummarize

Overview

PocketSummarize is an on-device iOS summarization app that uses the Core ML–converted all-MiniLM-L6-v2 embedding model to generate high-quality summaries from any user-entered text. The app is optimized for mobile performance and privacy, with no internet connection required.

---

Folder Structure

PocketSummarize/
├─ Models/
│  ├─ embeddings/
│  │  ├─ AllMiniLML6V2.mlmodel
│  ├─ llm/
│  │  └─ vocab.txt
├─ Preview Content/
├─ Scripts/
│  └─ convert_minilm_to_coreml/
├─ Sources/
│  ├─ App/
│  │  └─ PocketSummarizeApp.swift
│  ├─ Models/
│  ├─ Services/
│  │  └─ SummaryEngine.swift
│  ├─ Utils/
│  │  ├─ MiniLMTokenizer.swift
│  │  └─ MLHelpers.swift
│  ├─ ViewModels/
│  └─ Views/
│     ├─ ContentView.swift
│     ├─ SummaryView.swift
│     └─ TokenDebugView.swift
├─ Assets/
├─ LICENSE
└─ README.md

---

Getting Started

1. Clone the project

git clone https://github.com/Abhishek6353/PocketSummarize.git
cd PocketSummarize

2. Open in Xcode

open .

3. Ensure model + vocab are added to your target

• In Xcode: select AllMiniLML6V2.mlmodel → Target Membership → ON
• Do the same for vocab.txt

4. Run the app

Use the iOS Simulator or physical device to test the summarization flow.

---

Usage Example

Paste or type any article or paragraph, tap Summarize, and view the output.

Inside SummaryEngine.swift:

let engine = try SummaryEngine(seqLen: 64)
let result = try await engine.summarize(inputText)

---

Model License & Attribution

The original model:

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

Converted model files included here comply with the same license.

---

Code License

All Swift code, tokenizer logic, and summarization engine are licensed under the MIT License.

---

Maintainer

Abhishek GitHub: @Abhishek6353

---

Star the Repo ⭐

If this tool helped you build your on-device NLP app faster, consider starring the repo!